PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise

Advanced search

Message boards : Number crunching : a variety of CPU generations, overwelmed by all the subprojects in a good way

Author Message
wolfman1360
Send message
Joined: 18 Feb 18
Posts: 56
ID: 979448
Credit: 64,200,115
RAC: 6
321 LLR Bronze: Earned 10,000 credits (69,252)Cullen LLR Silver: Earned 100,000 credits (434,949)ESP LLR Bronze: Earned 10,000 credits (86,507)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (186,794)PPS LLR Amethyst: Earned 1,000,000 credits (1,106,516)PSP LLR Silver: Earned 100,000 credits (111,724)SoB LLR Jade: Earned 10,000,000 credits (16,939,060)SR5 LLR Gold: Earned 500,000 credits (929,564)SGS LLR Bronze: Earned 10,000 credits (77,792)TRP LLR Amethyst: Earned 1,000,000 credits (1,979,586)Woodall LLR Silver: Earned 100,000 credits (240,757)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (3,181,160)PPS Sieve Jade: Earned 10,000,000 credits (16,895,452)AP 26/27 Amethyst: Earned 1,000,000 credits (1,277,588)GFN Jade: Earned 10,000,000 credits (19,782,869)WW Gold: Earned 500,000 credits (900,000)
Message 149221 - Posted: 8 Mar 2021 | 9:34:21 UTC

Hello!
I've been a bit of a lurker on the Discord and have primarily been focused on biological projects, but figured I'd throw a few machines over to this one for a while. I forgot just how many projects there are and am slightly overwelmed - it doesn't help it's 3:30 in the morning and my insomnia has kicked in full force.

I've got everything from a dual Xeon 5520, a few 3rd generation Ryzen's, my personal machine - an 1800x, a few Ivy bridge i5's as well as a sandy bridge i7 and a dual Xeon e5-2660v3 (with some issues, so I think it will be retired sooner than later.)
The 1800x also has an rx570 - and I have a laptop with a 1050 ti and 8750h. Unfortunately that's all I have for GPU's currently, my 1080 is over at a location I can't get to, thanks Covid. Most of these are running Linux. I have a few other processors in a drawer, but this will do for now.
I truly cannot put into words how much I ador the new options regarding number of CPUS and jobs per location. One of the things that I wasn't the biggest fan of previously was having to go through the app config for each subproject. I think the 5520 - my space heater - is going to at the very least be replaced by an i7-4790 in the coming days, provided I can find the motivation. I think I have an old 500 gb hard drive that should do okay.

As I'm floundering my way through both the forum and the subproject list, what do folks recommend for most of the processors listed? I'm not too up to date on the specifics in regards to what each subproject supports in terms of avx, what processor favors what project, or vice versa. I have no real goals in mind and am not really sure what projects are more about finding primes or what have you. All of these machines are dedicated crunchers though I will probably not be outright disabling hyperthreading and, at least on the ones with it enabled, I guess I'd just set Boinc to use 50% CPU?
Really loving all the community here and looking forward to crunching! If anyone has any team recommendations do let me know, too.
thanks. Time to get some sleep. ;)

Anthony Ayiomamitis
Send message
Joined: 19 Mar 15
Posts: 2877
ID: 386066
Credit: 941,437,990
RAC: 352,490
Found 2 primes in the 2019 Tour de PrimesFound 1 prime in the 2020 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,048,709)Cullen LLR Turquoise: Earned 5,000,000 credits (5,026,008)ESP LLR Turquoise: Earned 5,000,000 credits (5,098,225)Generalized Cullen/Woodall LLR Turquoise: Earned 5,000,000 credits (5,127,156)PPS LLR Sapphire: Earned 20,000,000 credits (20,015,895)PSP LLR Turquoise: Earned 5,000,000 credits (5,117,327)SoB LLR Jade: Earned 10,000,000 credits (15,271,680)SR5 LLR Turquoise: Earned 5,000,000 credits (5,093,822)SGS LLR Turquoise: Earned 5,000,000 credits (5,012,543)TRP LLR Turquoise: Earned 5,000,000 credits (5,050,008)Woodall LLR Turquoise: Earned 5,000,000 credits (5,062,520)321 Sieve (suspended) Double Bronze: Earned 100,000,000 credits (102,500,323)Generalized Cullen/Woodall Sieve (suspended) Double Bronze: Earned 100,000,000 credits (100,500,349)PPS Sieve Double Silver: Earned 200,000,000 credits (243,349,119)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (14,516,062)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,329,228)AP 26/27 Emerald: Earned 50,000,000 credits (58,526,468)GFN Double Bronze: Earned 100,000,000 credits (104,032,849)WW Double Bronze: Earned 100,000,000 credits (100,228,000)PSA Double Bronze: Earned 100,000,000 credits (116,531,700)
Message 149224 - Posted: 8 Mar 2021 | 11:46:13 UTC - in response to Message 149221.

From the GPU side, I would suggest you look at the WW project since it is a badge you do not have and there is some urgency to get going since we just crossed the midpoint of the original leading edge (4611.686) that was cited at the start of the project.

I will let others chime in for the CPU side since I am strictly Intel and I cannot offer anything of substance for your Ryzen CPU's. As far as I know, we do not have any CPU project on a short leash for the moment which would otherwise be of greater interest.

Profile mikey
Avatar
Send message
Joined: 17 Mar 09
Posts: 1290
ID: 37043
Credit: 533,290,723
RAC: 77
Discovered 1 mega prime321 LLR Ruby: Earned 2,000,000 credits (2,038,739)Cullen LLR Ruby: Earned 2,000,000 credits (2,074,615)ESP LLR Ruby: Earned 2,000,000 credits (2,243,517)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,142,353)PPS LLR Turquoise: Earned 5,000,000 credits (7,292,665)PSP LLR Ruby: Earned 2,000,000 credits (2,232,103)SoB LLR Ruby: Earned 2,000,000 credits (2,934,612)SR5 LLR Ruby: Earned 2,000,000 credits (2,053,250)SGS LLR Turquoise: Earned 5,000,000 credits (5,152,318)TRP LLR Ruby: Earned 2,000,000 credits (2,405,015)Woodall LLR Ruby: Earned 2,000,000 credits (2,014,811)321 Sieve (suspended) Sapphire: Earned 20,000,000 credits (23,770,672)Cullen/Woodall Sieve (suspended) Gold: Earned 500,000 credits (944,431)Generalized Cullen/Woodall Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,813,253)PPS Sieve Double Silver: Earned 200,000,000 credits (342,048,709)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,446,797)AP 26/27 Sapphire: Earned 20,000,000 credits (33,140,471)GFN Sapphire: Earned 20,000,000 credits (49,299,387)WW Turquoise: Earned 5,000,000 credits (7,792,000)PSA Sapphire: Earned 20,000,000 credits (20,457,430)
Message 149225 - Posted: 8 Mar 2021 | 11:48:50 UTC - in response to Message 149221.

Hello!
I've been a bit of a lurker on the Discord and have primarily been focused on biological projects, but figured I'd throw a few machines over to this one for a while. I forgot just how many projects there are and am slightly overwelmed - it doesn't help it's 3:30 in the morning and my insomnia has kicked in full force.

I've got everything from a dual Xeon 5520, a few 3rd generation Ryzen's, my personal machine - an 1800x, a few Ivy bridge i5's as well as a sandy bridge i7 and a dual Xeon e5-2660v3 (with some issues, so I think it will be retired sooner than later.)
The 1800x also has an rx570 - and I have a laptop with a 1050 ti and 8750h. Unfortunately that's all I have for GPU's currently, my 1080 is over at a location I can't get to, thanks Covid. Most of these are running Linux. I have a few other processors in a drawer, but this will do for now.
I truly cannot put into words how much I ador the new options regarding number of CPUS and jobs per location. One of the things that I wasn't the biggest fan of previously was having to go through the app config for each subproject. I think the 5520 - my space heater - is going to at the very least be replaced by an i7-4790 in the coming days, provided I can find the motivation. I think I have an old 500 gb hard drive that should do okay.

As I'm floundering my way through both the forum and the subproject list, what do folks recommend for most of the processors listed? I'm not too up to date on the specifics in regards to what each subproject supports in terms of avx, what processor favors what project, or vice versa. I have no real goals in mind and am not really sure what projects are more about finding primes or what have you. All of these machines are dedicated crunchers though I will probably not be outright disabling hyperthreading and, at least on the ones with it enabled, I guess I'd just set Boinc to use 50% CPU?
Really loving all the community here and looking forward to crunching! If anyone has any team recommendations do let me know, too.
thanks. Time to get some sleep. ;)


All but the older than ver 3 Ryziens don't do as well as the newer ones but since you said you have a few Ryzien 3 machines I would start with those by putting them on a cpu project with less than a 5 hour finish time and let it run for a bit and see how it does. Use the default settings at first then start fiddling with them to see what's optimal for your processor and your machine.

As for the RX570 it will do just fine on the projects with less than say 4 hours per task, more than that and it is taking too long for you to get a handle on how it's actually doing.

As a general rule your Intel cpu's will all do very good on the cpu projects as long as you limit them to 50% of the cpu usage in Boinc, that way they only use the actual cores and not the HT ones, and you can pretty much choose any project. Obviously the newer ones will be faster in most cases, higher clock speeds, but some of the older ones had good caches on them and they still do a good job. It's more of a trial and error thing on which is best for which pc but with all the different venues here you can have fun putting each pc, or group of pc's in it's own venue crunching it's own kind of tasks and using it's own settings on how to do it.

Profile Michael GoetzProject donor
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 21 Jan 10
Posts: 13648
ID: 53948
Credit: 285,351,178
RAC: 47,109
The "Shut up already!" badge:  This loud mouth has mansplained on the forums over 10 thousand times!  Sheesh!!!Discovered the World's First GFN-19 prime!!!Discovered 1 mega primeFound 1 prime in the 2018 Tour de PrimesFound 1 prime in the 2019 Tour de PrimesFound 1 prime in the 2020 Tour de PrimesFound 2 primes in the 2021 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,132,712)Cullen LLR Turquoise: Earned 5,000,000 credits (5,038,114)ESP LLR Turquoise: Earned 5,000,000 credits (6,177,890)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,322,963)PPS LLR Sapphire: Earned 20,000,000 credits (20,751,038)PSP LLR Turquoise: Earned 5,000,000 credits (7,956,186)SoB LLR Sapphire: Earned 20,000,000 credits (36,067,618)SR5 LLR Jade: Earned 10,000,000 credits (10,007,110)SGS LLR Ruby: Earned 2,000,000 credits (3,718,606)TRP LLR Turquoise: Earned 5,000,000 credits (5,084,329)Woodall LLR Turquoise: Earned 5,000,000 credits (5,032,821)321 Sieve (suspended) Jade: Earned 10,000,000 credits (10,061,196)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,170,256)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,059,304)PPS Sieve Sapphire: Earned 20,000,000 credits (22,885,121)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,035,522)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,051,121)AP 26/27 Jade: Earned 10,000,000 credits (10,118,303)GFN Emerald: Earned 50,000,000 credits (78,031,938)WW Sapphire: Earned 20,000,000 credits (32,204,000)PSA Jade: Earned 10,000,000 credits (12,445,029)
Message 149232 - Posted: 8 Mar 2021 | 14:59:02 UTC - in response to Message 149221.

Hello!
I've been a bit of a lurker on the Discord and have primarily been focused on biological projects, but figured I'd throw a few machines over to this one for a while. I forgot just how many projects there are and am slightly overwelmed - it doesn't help it's 3:30 in the morning and my insomnia has kicked in full force.

I've got everything from a dual Xeon 5520, a few 3rd generation Ryzen's, my personal machine - an 1800x, a few Ivy bridge i5's as well as a sandy bridge i7 and a dual Xeon e5-2660v3 (with some issues, so I think it will be retired sooner than later.)
The 1800x also has an rx570 - and I have a laptop with a 1050 ti and 8750h. Unfortunately that's all I have for GPU's currently, my 1080 is over at a location I can't get to, thanks Covid. Most of these are running Linux. I have a few other processors in a drawer, but this will do for now.
I truly cannot put into words how much I ador the new options regarding number of CPUS and jobs per location. One of the things that I wasn't the biggest fan of previously was having to go through the app config for each subproject. I think the 5520 - my space heater - is going to at the very least be replaced by an i7-4790 in the coming days, provided I can find the motivation. I think I have an old 500 gb hard drive that should do okay.

As I'm floundering my way through both the forum and the subproject list, what do folks recommend for most of the processors listed? I'm not too up to date on the specifics in regards to what each subproject supports in terms of avx, what processor favors what project, or vice versa. I have no real goals in mind and am not really sure what projects are more about finding primes or what have you. All of these machines are dedicated crunchers though I will probably not be outright disabling hyperthreading and, at least on the ones with it enabled, I guess I'd just set Boinc to use 50% CPU?
Really loving all the community here and looking forward to crunching! If anyone has any team recommendations do let me know, too.
thanks. Time to get some sleep. ;)


Hardware (CPUs):

I divide CPUs into several categories. JUNK, SLOW, AVERAGE, and FAST.

JUNK: Intel CPUs without AVX. That's everything before Sandy Bridge (2xxx), and more recent CPUs in their lower product categories such as Celerons. For AMD it's worse, as all but their most recent CPU lines had worthless AVX implementations. The Junk category therefore includes all AMD CPUs before the Zen2 architecture (the 3xxx series CPUs).

SLOW: Intel CPUs with AVX, but without AVX2/FMA3. These are the Sandy Bridge (2xxx) and Ivy Bridge (3xxx) CPUs. They are roughly twice as fast as otherwise equivalent CPUs lacking AVX. There are no AMD CPUs in this category.

AVERAGE: Intel or AMD CPUs with (functional) AVX2 (also called FMA3). These are Haswell (4xxx) and later Intel CPUs, and Zen2 (3xxx) or later AMD CPUs. Roughly twice as fast as AVX. Most good modern CPUs are in the average category. "Modern" means AMD CPUs from the last 2 or 3 years and Intel CPUs from the last 6 or so years.

FAST: Intel AVX-512 (dual unit) CPUs. Only really expensive consumer and some server CPUs have this feature. AMD doesn't yet support AVX512. Roughly twice as fast as AVX2/FMA3.

Obviously, this is a gross simplification, but it should be good enough.

Projects:

Broadly speaking, there are projects where you can use GPUs, and projects where you can't. Generally speaking, you shouldn't bother using a CPU on a GPU project. It's "bringing a knife to a gun fight."

The GPU projects are all of the GFN projects, PPS-Sieve, AP27, and WW. You probably shouldn't use your CPUs on those projects.

That lease us with all of the LLR projects, which is where most people use their CPUs. It *used* to be that you were always competing with a wingman to be "first" in order to be credited with being the discover of a prime. Today, that only applies to the smallest two projects, SGS and PPSE. Ironically, that means that the smallest projects are not the best place to use your slowest CPUs. Realistically, anything in the junk or slow categories has a very small chance of finding a prime on SGS or PPSE because you'll usually be beaten to the finish line by your wingman.

All of the larger LLR projects now use some fancy math to eliminate the need for double checking, so you don't have a wingman you're competing with. For junk and slow CPUs, my recommendation is to run PPS, which is the smallest of the LLR projects that don't have wingmen.

Average and Fast CPUs can go on any LLR project you wish.

As for GPUs, all the GPU projects have wingmen, so you'll be competing to be first. There's no primes to discover on PPS-Sieve, and the credit rewards are good, so that's a good choice for slower GPUs. Also a good choice for slow (or fast) GPUs is WW. It's new, and it has a limited lifetime (we're currently 12% done, after a couple of months). That would probably be my recommendation for your GPUs. Faster GPUs can give the variously-sized GFN projects a try.

AP27 is also a good choice for GPUs because it produces a lot of small discoveries.

If you want to find your first prime, the most important thing to remember is that primes are rarer as the numbers get larger. Double the number of digits in a number, and it takes about 10 times as long to find a prime. This means SGS is the best bet for finding a prime on a CPU, and GFN-15 is the best bet on a GPU.
____________
My lucky number is 75898524288+1

Profile BurProject donor
Volunteer tester
Avatar
Send message
Joined: 25 Feb 20
Posts: 462
ID: 1241833
Credit: 279,947,726
RAC: 547,227
321 LLR Ruby: Earned 2,000,000 credits (2,092,823)Cullen LLR Ruby: Earned 2,000,000 credits (2,002,841)ESP LLR Amethyst: Earned 1,000,000 credits (1,445,099)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,180,764)PPS LLR Amethyst: Earned 1,000,000 credits (1,225,852)PSP LLR Ruby: Earned 2,000,000 credits (2,064,832)SoB LLR Amethyst: Earned 1,000,000 credits (1,669,219)SR5 LLR Ruby: Earned 2,000,000 credits (2,065,004)SGS LLR Amethyst: Earned 1,000,000 credits (1,202,156)TRP LLR Ruby: Earned 2,000,000 credits (2,089,856)Woodall LLR Ruby: Earned 2,000,000 credits (2,112,258)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,107,153)PPS Sieve Turquoise: Earned 5,000,000 credits (5,080,097)AP 26/27 Ruby: Earned 2,000,000 credits (4,697,966)GFN Turquoise: Earned 5,000,000 credits (7,489,336)WW Double Silver: Earned 200,000,000 credits (239,400,000)PSA Amethyst: Earned 1,000,000 credits (1,022,470)
Message 149242 - Posted: 8 Mar 2021 | 18:43:16 UTC

To add my 2 cents:

I like WW for GPU, because if something is found, it's major. I think more so than the next Mersenne prime.

Chances are very low though, iirc finding another Wieferich prime is expected within 10^1000 and we only search to 2^64 or 10^18... But the good part is, even if nothing is found, we did at least establish a new boundary. I'm not sure if the results are accepted by the math community in a way that it can be quoted in scientific literature, but I hope so.


For CPU I find the conjectures the most interesting, because it's not just another prime that will eventually be kicked out of the Top 5000, but it has a lasting effect. Of course same as always: the lower the chance to find something, the higher the reward and vice versa. Seventeen or Bust has smallest odds and biggest primes, The Riesel Problem has the best odds.

A much higher chance to find something but still of some significance is the Fermat Divisor search. Primes here are not unlikely to be (x)(G)Fermat divisors. So it's an interesting alternative.

Otherwise, as already pointed out, if you're mainly after finding primes for entry in Caldwell's list, go for the Proth subprojects for CPU or GFN for GPU.
____________
1281979 * 2^485014 + 1 is prime ... no further hits up to: n = 4,800,000

wolfman1360
Send message
Joined: 18 Feb 18
Posts: 56
ID: 979448
Credit: 64,200,115
RAC: 6
321 LLR Bronze: Earned 10,000 credits (69,252)Cullen LLR Silver: Earned 100,000 credits (434,949)ESP LLR Bronze: Earned 10,000 credits (86,507)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (186,794)PPS LLR Amethyst: Earned 1,000,000 credits (1,106,516)PSP LLR Silver: Earned 100,000 credits (111,724)SoB LLR Jade: Earned 10,000,000 credits (16,939,060)SR5 LLR Gold: Earned 500,000 credits (929,564)SGS LLR Bronze: Earned 10,000 credits (77,792)TRP LLR Amethyst: Earned 1,000,000 credits (1,979,586)Woodall LLR Silver: Earned 100,000 credits (240,757)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (3,181,160)PPS Sieve Jade: Earned 10,000,000 credits (16,895,452)AP 26/27 Amethyst: Earned 1,000,000 credits (1,277,588)GFN Jade: Earned 10,000,000 credits (19,782,869)WW Gold: Earned 500,000 credits (900,000)
Message 149243 - Posted: 8 Mar 2021 | 18:54:50 UTC

Thanks for all the help so far, this definitely answers a lot.
As far as threads are concerned, what is more benifitial?
Let's give a Ryzen 7 3700x, for instance. Would running 4x 2 threads give more throughput than 2x 4 CPUs? Ultimately - what should my end goal be, or does that also depend heavily on subproject?

If I remember correctly, SGS and are all single threaded anyway and only the LLR projects are multithreaded, so will lack of hyperthreading in the single threaded projects be hurtful?

I've almost run the 5520 dry of tasks. I'm going to get the 4790 up and running and throw it over here. I will likely never have anything supporting avx512 - I am an AMD guy at this point, both for price and performance, especially in Canada.

Profile Chooka
Send message
Joined: 15 May 18
Posts: 211
ID: 1014486
Credit: 390,512,214
RAC: 381,023
Found 1 prime in the 2019 Tour de PrimesFound 5 primes in the 2020 Tour de PrimesFound 4 primes in the 2021 Tour de Primes321 LLR Ruby: Earned 2,000,000 credits (2,216,329)Cullen LLR Ruby: Earned 2,000,000 credits (2,826,433)ESP LLR Ruby: Earned 2,000,000 credits (3,928,263)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,218,841)PPS LLR Sapphire: Earned 20,000,000 credits (27,292,881)PSP LLR Turquoise: Earned 5,000,000 credits (7,765,873)SoB LLR Turquoise: Earned 5,000,000 credits (5,915,028)SR5 LLR Ruby: Earned 2,000,000 credits (2,095,507)SGS LLR Ruby: Earned 2,000,000 credits (2,037,347)TRP LLR Ruby: Earned 2,000,000 credits (4,118,961)Woodall LLR Ruby: Earned 2,000,000 credits (2,247,622)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,279,140)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (6,057,633)PPS Sieve Double Silver: Earned 200,000,000 credits (229,743,763)AP 26/27 Ruby: Earned 2,000,000 credits (2,919,046)GFN Sapphire: Earned 20,000,000 credits (34,893,546)WW Emerald: Earned 50,000,000 credits (51,956,000)
Message 149244 - Posted: 8 Mar 2021 | 19:04:33 UTC

Great break down Michael!
This explains why my 3930K is almost twice as slow as my 4930K. It's pretty woeful (although it does have 2 x GPU's where as the 4930K only has one....so there is some variation there.)

My old Dell 4xxx series crunchers are pretty good albeit only 4 cores.
____________

Profile Rafael
Volunteer tester
Avatar
Send message
Joined: 22 Oct 14
Posts: 899
ID: 370496
Credit: 413,035,696
RAC: 237,742
321 LLR Turquoise: Earned 5,000,000 credits (8,236,942)Cullen LLR Turquoise: Earned 5,000,000 credits (8,028,695)ESP LLR Turquoise: Earned 5,000,000 credits (8,437,609)Generalized Cullen/Woodall LLR Turquoise: Earned 5,000,000 credits (8,011,297)PPS LLR Turquoise: Earned 5,000,000 credits (8,392,935)PSP LLR Turquoise: Earned 5,000,000 credits (9,998,922)SoB LLR Turquoise: Earned 5,000,000 credits (9,310,410)SR5 LLR Turquoise: Earned 5,000,000 credits (9,959,965)SGS LLR Turquoise: Earned 5,000,000 credits (8,944,665)TRP LLR Turquoise: Earned 5,000,000 credits (8,917,699)Woodall LLR Jade: Earned 10,000,000 credits (10,051,780)321 Sieve (suspended) Jade: Earned 10,000,000 credits (10,033,828)Generalized Cullen/Woodall Sieve (suspended) Jade: Earned 10,000,000 credits (10,037,204)PPS Sieve Jade: Earned 10,000,000 credits (10,305,147)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,000,053)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,030,160)AP 26/27 Sapphire: Earned 20,000,000 credits (22,935,939)GFN Sapphire: Earned 20,000,000 credits (35,932,489)WW Emerald: Earned 50,000,000 credits (50,712,000)PSA Double Bronze: Earned 100,000,000 credits (170,761,999)
Message 149252 - Posted: 8 Mar 2021 | 21:22:41 UTC - in response to Message 149243.

As far as threads are concerned, what is more benifitial?
Let's give a Ryzen 7 3700x, for instance. Would running 4x 2 threads give more throughput than 2x 4 CPUs? Ultimately - what should my end goal be, or does that also depend heavily on subproject?

It depends heavily on the subproject, or rather the size of the FFT used. The rule is to try and use as few threads per task as possible while still keeping everything within CPU cache. In your case, you have 32MB of cache, so something like Sob that's going at 3200k*8 =~25MB of cache is probably better left as 1x8threads rather than 2x4t, whereas something smaller like ESP at 1600k*8 = 12.8MB would be better at 2x4t since it all fits.

With that said, this is just an educated guess and the best option is for you to test yourself. You might also be fine with suboptimal settings in the name of doing individual tasks faster, or perhaps leave a core free for the GPU or your daily driver needs. It's all up to you.


If I remember correctly, SGS and are all single threaded anyway and only the LLR projects are multithreaded

SGS and PPSE use LLR, and while not very efficient, you can multithread it if you want to. You might be mistaking it for LLR2, which is a similar algorithm used for most other CPU projects and is generally a bit slower on it's own, but significantly cuts down on double checking work (along with some other goodies).

so will lack of hyperthreading in the single threaded projects be hurtful?

HyperThreading (or SMT in AMD's case) is generally not recommended for LLR/2 tasks. It is recommended that you either have BOINC use only 50% of the CPUs or turn it off in BIOS.

wolfman1360
Send message
Joined: 18 Feb 18
Posts: 56
ID: 979448
Credit: 64,200,115
RAC: 6
321 LLR Bronze: Earned 10,000 credits (69,252)Cullen LLR Silver: Earned 100,000 credits (434,949)ESP LLR Bronze: Earned 10,000 credits (86,507)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (186,794)PPS LLR Amethyst: Earned 1,000,000 credits (1,106,516)PSP LLR Silver: Earned 100,000 credits (111,724)SoB LLR Jade: Earned 10,000,000 credits (16,939,060)SR5 LLR Gold: Earned 500,000 credits (929,564)SGS LLR Bronze: Earned 10,000 credits (77,792)TRP LLR Amethyst: Earned 1,000,000 credits (1,979,586)Woodall LLR Silver: Earned 100,000 credits (240,757)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (3,181,160)PPS Sieve Jade: Earned 10,000,000 credits (16,895,452)AP 26/27 Amethyst: Earned 1,000,000 credits (1,277,588)GFN Jade: Earned 10,000,000 credits (19,782,869)WW Gold: Earned 500,000 credits (900,000)
Message 149254 - Posted: 8 Mar 2021 | 22:21:32 UTC - in response to Message 149252.

As far as threads are concerned, what is more benifitial?
Let's give a Ryzen 7 3700x, for instance. Would running 4x 2 threads give more throughput than 2x 4 CPUs? Ultimately - what should my end goal be, or does that also depend heavily on subproject?

It depends heavily on the subproject, or rather the size of the FFT used. The rule is to try and use as few threads per task as possible while still keeping everything within CPU cache. In your case, you have 32MB of cache, so something like Sob that's going at 3200k*8 =~25MB of cache is probably better left as 1x8threads rather than 2x4t, whereas something smaller like ESP at 1600k*8 = 12.8MB would be better at 2x4t since it all fits.

With that said, this is just an educated guess and the best option is for you to test yourself. You might also be fine with suboptimal settings in the name of doing individual tasks faster, or perhaps leave a core free for the GPU or your daily driver needs. It's all up to you.

Is there a general rundown of how much cache is needed per subproject?
This is all super helpful info, thank you.
I may be getting projects confused for sure - there are a lot of abreviations to keep in mind.
For now I think I'll get the decent machines crunching PPS-DIV, SOB, and TRP. Perhaps a GPU on WW.
Should I not get less powerful e.g. Xeon 5520 on SGS, the shorter running WUs, with hyperthreading disabled/Boinc at 50% CPU?
Sorry for all the questions. There is a lot of info to grasp and I'd like this to run as efficiently as possible. Given I have so much variety in CPU and especially l3 cache (i3-2130 for instance) this will take some doing.

Profile Michael GoetzProject donor
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 21 Jan 10
Posts: 13648
ID: 53948
Credit: 285,351,178
RAC: 47,109
The "Shut up already!" badge:  This loud mouth has mansplained on the forums over 10 thousand times!  Sheesh!!!Discovered the World's First GFN-19 prime!!!Discovered 1 mega primeFound 1 prime in the 2018 Tour de PrimesFound 1 prime in the 2019 Tour de PrimesFound 1 prime in the 2020 Tour de PrimesFound 2 primes in the 2021 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,132,712)Cullen LLR Turquoise: Earned 5,000,000 credits (5,038,114)ESP LLR Turquoise: Earned 5,000,000 credits (6,177,890)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,322,963)PPS LLR Sapphire: Earned 20,000,000 credits (20,751,038)PSP LLR Turquoise: Earned 5,000,000 credits (7,956,186)SoB LLR Sapphire: Earned 20,000,000 credits (36,067,618)SR5 LLR Jade: Earned 10,000,000 credits (10,007,110)SGS LLR Ruby: Earned 2,000,000 credits (3,718,606)TRP LLR Turquoise: Earned 5,000,000 credits (5,084,329)Woodall LLR Turquoise: Earned 5,000,000 credits (5,032,821)321 Sieve (suspended) Jade: Earned 10,000,000 credits (10,061,196)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,170,256)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,059,304)PPS Sieve Sapphire: Earned 20,000,000 credits (22,885,121)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,035,522)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,051,121)AP 26/27 Jade: Earned 10,000,000 credits (10,118,303)GFN Emerald: Earned 50,000,000 credits (78,031,938)WW Sapphire: Earned 20,000,000 credits (32,204,000)PSA Jade: Earned 10,000,000 credits (12,445,029)
Message 149256 - Posted: 8 Mar 2021 | 22:39:16 UTC - in response to Message 149243.

Thanks for all the help so far, this definitely answers a lot.
As far as threads are concerned, what is more benifitial?
Let's give a Ryzen 7 3700x, for instance. Would running 4x 2 threads give more throughput than 2x 4 CPUs? Ultimately - what should my end goal be, or does that also depend heavily on subproject?

If I remember correctly, SGS and are all single threaded anyway and only the LLR projects are multithreaded, so will lack of hyperthreading in the single threaded projects be hurtful?

I've almost run the 5520 dry of tasks. I'm going to get the 4790 up and running and throw it over here. I will likely never have anything supporting avx512 - I am an AMD guy at this point, both for price and performance, especially in Canada.


Uh oh. Do you understand the difference between HyperThreading and multi-threading? They are *completely* different things, and you need to understand both of them. If you don't know the difference, then ASK. It's a long explanation, and I'd rather spend time answering your questions assuming that you do indeed know what those two terms mean. If you need those terms explained, ask and I'll explain them, and then you can come back and read the following answers to your questions.

First of all LLR and LLR2 are both LLR. SGS, PPS, PPSE, PPS-MEGA, TRP, 321, SoB, etc., are all LLR projects. We run some of them (SGS and PPSE) with the original LLR, and all the others with the more advanced LLR2. (We don't use LLR2 for SGS and PPSE only because it would overload the server.)

For all LLR/LLR2 projects HyperThreading should not be used. If you have a CPU with HyperThreading, like the Intel i7-4790 or the AMD 3700X, you can either turn it off in the BIOS or set BOINC to use "50% of the processors". Note that is NOT "use 50% of the CPU time". Two different settings.

As for multi-threading, in general, it's beneficial to use it for larger numbers, but not for smaller numbers. Most people usually don't multithread SGS, PPSE, PPS, and sometimes PPS-MEGA. Larger numbers usually benefit from multi-threading. That being said, every computer is different. The CPUs are different, the motherboards are different, and the memory is different. What works best for my computer might not work best for your computer. Testing with different numbers of threads on your exact computer is the best way to find out what's best.

A special word about new AMD CPUs like your 3700X. That CPU is divided into 2 CCX units, and moving data between the two units is fairly slow. This CPU is an 8 core, 16 thread CPU. Because we don't want to use HyperThreading, that means we're only interested in the 8 cores. We'll ignore the other 8 HyperThreads. For a really large task like an SoB, normally one would be tempted to use a single task with 8 threads. However, on this CPU, it's better to run two tasks with 4 threads each, and lock each task into a single CCX unit. Under Windows, you can do that with the AffinityWatcher program. If you don't lock the threads into a single CCX, they'll drift back and forth, and that will significantly slow down the task. The slowdown will happen if you run an 8 thread task, which would, by necessity, span both CCX units.

8 core Intel CPUs don't have this problem because they don't have this split architecture.

____________
My lucky number is 75898524288+1

Profile j.sheridanProject donor
Volunteer tester
Send message
Joined: 21 Mar 11
Posts: 816
ID: 91622
Credit: 2,342,089,257
RAC: 2,407,017
Discovered 2 mega primesFound 4 primes in the 2021 Tour de PrimesFound 2 mega primes in the 2021 Tour de Primes321 LLR Jade: Earned 10,000,000 credits (10,011,712)Cullen LLR Jade: Earned 10,000,000 credits (10,025,728)ESP LLR Jade: Earned 10,000,000 credits (10,176,490)Generalized Cullen/Woodall LLR Jade: Earned 10,000,000 credits (10,020,580)PPS LLR Sapphire: Earned 20,000,000 credits (29,281,953)PSP LLR Jade: Earned 10,000,000 credits (10,055,994)SoB LLR Sapphire: Earned 20,000,000 credits (40,958,919)SR5 LLR Jade: Earned 10,000,000 credits (10,885,368)SGS LLR Jade: Earned 10,000,000 credits (10,007,143)TRP LLR Jade: Earned 10,000,000 credits (10,009,531)Woodall LLR Jade: Earned 10,000,000 credits (10,039,527)321 Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,019,388)Cullen/Woodall Sieve (suspended) Double Silver: Earned 200,000,000 credits (265,102,350)PPS Sieve Double Gold: Earned 500,000,000 credits (547,992,812)TRP Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,801,812)AP 26/27 Emerald: Earned 50,000,000 credits (93,546,934)GFN Double Silver: Earned 200,000,000 credits (439,903,440)WW Double Gold: Earned 500,000,000 credits (808,248,000)
Message 149258 - Posted: 8 Mar 2021 | 22:53:49 UTC - in response to Message 149256.
Last modified: 8 Mar 2021 | 22:58:33 UTC

On the 3700x it's still faster to run anything above ~2M fft with 8 threads.

3950x which scales near enough to 2x the 3700x:
Timings for 2880K all-complex FFT length (16 cores, 2 workers): Throughput: 1132.42 iter/sec.
Timings for 2880K all-complex FFT length (16 cores, 4 workers): Throughput: 503.76 iter/sec.

wolfman1360
Send message
Joined: 18 Feb 18
Posts: 56
ID: 979448
Credit: 64,200,115
RAC: 6
321 LLR Bronze: Earned 10,000 credits (69,252)Cullen LLR Silver: Earned 100,000 credits (434,949)ESP LLR Bronze: Earned 10,000 credits (86,507)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (186,794)PPS LLR Amethyst: Earned 1,000,000 credits (1,106,516)PSP LLR Silver: Earned 100,000 credits (111,724)SoB LLR Jade: Earned 10,000,000 credits (16,939,060)SR5 LLR Gold: Earned 500,000 credits (929,564)SGS LLR Bronze: Earned 10,000 credits (77,792)TRP LLR Amethyst: Earned 1,000,000 credits (1,979,586)Woodall LLR Silver: Earned 100,000 credits (240,757)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (3,181,160)PPS Sieve Jade: Earned 10,000,000 credits (16,895,452)AP 26/27 Amethyst: Earned 1,000,000 credits (1,277,588)GFN Jade: Earned 10,000,000 credits (19,782,869)WW Gold: Earned 500,000 credits (900,000)
Message 149260 - Posted: 8 Mar 2021 | 23:07:36 UTC - in response to Message 149256.

Thanks for all the help so far, this definitely answers a lot.
As far as threads are concerned, what is more benifitial?
Let's give a Ryzen 7 3700x, for instance. Would running 4x 2 threads give more throughput than 2x 4 CPUs? Ultimately - what should my end goal be, or does that also depend heavily on subproject?

If I remember correctly, SGS and are all single threaded anyway and only the LLR projects are multithreaded, so will lack of hyperthreading in the single threaded projects be hurtful?

I've almost run the 5520 dry of tasks. I'm going to get the 4790 up and running and throw it over here. I will likely never have anything supporting avx512 - I am an AMD guy at this point, both for price and performance, especially in Canada.


Uh oh. Do you understand the difference between HyperThreading and multi-threading? They are *completely* different things, and you need to understand both of them. If you don't know the difference, then ASK. It's a long explanation, and I'd rather spend time answering your questions assuming that you do indeed know what those two terms mean. If you need those terms explained, ask and I'll explain them, and then you can come back and read the following answers to your questions.

First of all LLR and LLR2 are both LLR. SGS, PPS, PPSE, PPS-MEGA, TRP, 321, SoB, etc., are all LLR projects. We run some of them (SGS and PPSE) with the original LLR, and all the others with the more advanced LLR2. (We don't use LLR2 for SGS and PPSE only because it would overload the server.)

For all LLR/LLR2 projects HyperThreading should not be used. If you have a CPU with HyperThreading, like the Intel i7-4790 or the AMD 3700X, you can either turn it off in the BIOS or set BOINC to use "50% of the processors". Note that is NOT "use 50% of the CPU time". Two different settings.

As for multi-threading, in general, it's beneficial to use it for larger numbers, but not for smaller numbers. Most people usually don't multithread SGS, PPSE, PPS, and sometimes PPS-MEGA. Larger numbers usually benefit from multi-threading. That being said, every computer is different. The CPUs are different, the motherboards are different, and the memory is different. What works best for my computer might not work best for your computer. Testing with different numbers of threads on your exact computer is the best way to find out what's best.

A special word about new AMD CPUs like your 3700X. That CPU is divided into 2 CCX units, and moving data between the two units is fairly slow. This CPU is an 8 core, 16 thread CPU. Because we don't want to use HyperThreading, that means we're only interested in the 8 cores. We'll ignore the other 8 HyperThreads. For a really large task like an SoB, normally one would be tempted to use a single task with 8 threads. However, on this CPU, it's better to run two tasks with 4 threads each, and lock each task into a single CCX unit. Under Windows, you can do that with the AffinityWatcher program. If you don't lock the threads into a single CCX, they'll drift back and forth, and that will significantly slow down the task. The slowdown will happen if you run an 8 thread task, which would, by necessity, span both CCX units.

8 core Intel CPUs don't have this problem because they don't have this split architecture.

Thank you for that detailed explanation.
Extremeley short and utterly inadequate description - hyperthreading = 1 real core being split into 2 logical.
Multithreading = multiple cores being run on a single workunit e.g. 2, 4, 8, hense removing hyperthreading on an i7-4790 = 4 real cores.
yes, this is more basic and there is far more to it of course :)
Does the 3600x suffer from this same issue regarding CCX units?
How does Linux fair in this regard, majority of my hosts are running Ubuntu 20.04.
I do have a 1800x running Windows. Does this still apply on older Ryzens?
I am assuming re: LLR and LLR2 instruction set support remains the same e.g. avx?
So to be clear - is hyperthreading (Boinc at 100% CPU on hyperthreaded processors like Ryzen 3700x) useful for non LLR projects, or should I leave it disabled?

Thanks again

Profile Rafael
Volunteer tester
Avatar
Send message
Joined: 22 Oct 14
Posts: 899
ID: 370496
Credit: 413,035,696
RAC: 237,742
321 LLR Turquoise: Earned 5,000,000 credits (8,236,942)Cullen LLR Turquoise: Earned 5,000,000 credits (8,028,695)ESP LLR Turquoise: Earned 5,000,000 credits (8,437,609)Generalized Cullen/Woodall LLR Turquoise: Earned 5,000,000 credits (8,011,297)PPS LLR Turquoise: Earned 5,000,000 credits (8,392,935)PSP LLR Turquoise: Earned 5,000,000 credits (9,998,922)SoB LLR Turquoise: Earned 5,000,000 credits (9,310,410)SR5 LLR Turquoise: Earned 5,000,000 credits (9,959,965)SGS LLR Turquoise: Earned 5,000,000 credits (8,944,665)TRP LLR Turquoise: Earned 5,000,000 credits (8,917,699)Woodall LLR Jade: Earned 10,000,000 credits (10,051,780)321 Sieve (suspended) Jade: Earned 10,000,000 credits (10,033,828)Generalized Cullen/Woodall Sieve (suspended) Jade: Earned 10,000,000 credits (10,037,204)PPS Sieve Jade: Earned 10,000,000 credits (10,305,147)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,000,053)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,030,160)AP 26/27 Sapphire: Earned 20,000,000 credits (22,935,939)GFN Sapphire: Earned 20,000,000 credits (35,932,489)WW Emerald: Earned 50,000,000 credits (50,712,000)PSA Double Bronze: Earned 100,000,000 credits (170,761,999)
Message 149265 - Posted: 9 Mar 2021 | 0:32:16 UTC - in response to Message 149260.

Does the 3600x suffer from this same issue regarding CCX units?

Has 2 CCX as well. AFAIK the 3300x is the only non apu Ryzen 3xxx that has a single CCX.


How does Linux fair in this regard, majority of my hosts are running Ubuntu 20.04.

Apparently Linux does a lot better than Windows in this regard and there's little need to deal with affinity.

I do have a 1800x running Windows. Does this still apply on older Ryzens?

Yes.

I am assuming re: LLR and LLR2 instruction set support remains the same e.g. avx?

I think LLR2 can't run on 32bit? Not sure about that.

So to be clear - is hyperthreading (Boinc at 100% CPU on hyperthreaded processors like Ryzen 3700x) useful for non LLR projects, or should I leave it disabled?

It's generally useful for non LLR projects and thus should be enabled most of the time.

Profile Michael GoetzProject donor
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 21 Jan 10
Posts: 13648
ID: 53948
Credit: 285,351,178
RAC: 47,109
The "Shut up already!" badge:  This loud mouth has mansplained on the forums over 10 thousand times!  Sheesh!!!Discovered the World's First GFN-19 prime!!!Discovered 1 mega primeFound 1 prime in the 2018 Tour de PrimesFound 1 prime in the 2019 Tour de PrimesFound 1 prime in the 2020 Tour de PrimesFound 2 primes in the 2021 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,132,712)Cullen LLR Turquoise: Earned 5,000,000 credits (5,038,114)ESP LLR Turquoise: Earned 5,000,000 credits (6,177,890)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,322,963)PPS LLR Sapphire: Earned 20,000,000 credits (20,751,038)PSP LLR Turquoise: Earned 5,000,000 credits (7,956,186)SoB LLR Sapphire: Earned 20,000,000 credits (36,067,618)SR5 LLR Jade: Earned 10,000,000 credits (10,007,110)SGS LLR Ruby: Earned 2,000,000 credits (3,718,606)TRP LLR Turquoise: Earned 5,000,000 credits (5,084,329)Woodall LLR Turquoise: Earned 5,000,000 credits (5,032,821)321 Sieve (suspended) Jade: Earned 10,000,000 credits (10,061,196)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,170,256)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,059,304)PPS Sieve Sapphire: Earned 20,000,000 credits (22,885,121)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,035,522)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,051,121)AP 26/27 Jade: Earned 10,000,000 credits (10,118,303)GFN Emerald: Earned 50,000,000 credits (78,031,938)WW Sapphire: Earned 20,000,000 credits (32,204,000)PSA Jade: Earned 10,000,000 credits (12,445,029)
Message 149266 - Posted: 9 Mar 2021 | 1:09:48 UTC - in response to Message 149265.

I am assuming re: LLR and LLR2 instruction set support remains the same e.g. avx?

I think LLR2 can't run on 32bit? Not sure about that.


Yes, the instruction set support in LLR and LLR2 are the same. And, yes, there are 32 bit and 64 bit versions of both.

You're thinking of Windows XP (32 bit) support that no longer works. 32 bits on anything newer than XP will work. It's toolset compatibility with XP that's the problem.


____________
My lucky number is 75898524288+1

Profile Chooka
Send message
Joined: 15 May 18
Posts: 211
ID: 1014486
Credit: 390,512,214
RAC: 381,023
Found 1 prime in the 2019 Tour de PrimesFound 5 primes in the 2020 Tour de PrimesFound 4 primes in the 2021 Tour de Primes321 LLR Ruby: Earned 2,000,000 credits (2,216,329)Cullen LLR Ruby: Earned 2,000,000 credits (2,826,433)ESP LLR Ruby: Earned 2,000,000 credits (3,928,263)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,218,841)PPS LLR Sapphire: Earned 20,000,000 credits (27,292,881)PSP LLR Turquoise: Earned 5,000,000 credits (7,765,873)SoB LLR Turquoise: Earned 5,000,000 credits (5,915,028)SR5 LLR Ruby: Earned 2,000,000 credits (2,095,507)SGS LLR Ruby: Earned 2,000,000 credits (2,037,347)TRP LLR Ruby: Earned 2,000,000 credits (4,118,961)Woodall LLR Ruby: Earned 2,000,000 credits (2,247,622)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,279,140)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (6,057,633)PPS Sieve Double Silver: Earned 200,000,000 credits (229,743,763)AP 26/27 Ruby: Earned 2,000,000 credits (2,919,046)GFN Sapphire: Earned 20,000,000 credits (34,893,546)WW Emerald: Earned 50,000,000 credits (51,956,000)
Message 149269 - Posted: 9 Mar 2021 | 7:13:25 UTC - in response to Message 149258.

On the 3700x it's still faster to run anything above ~2M fft with 8 threads.

3950x which scales near enough to 2x the 3700x:
Timings for 2880K all-complex FFT length (16 cores, 2 workers): Throughput: 1132.42 iter/sec.
Timings for 2880K all-complex FFT length (16 cores, 4 workers): Throughput: 503.76 iter/sec.


And with my 3950X (for the upcoming SoB challenge for anyone interested) -

Timings for 3200K FFT length (16 cores, 1 worker): Throughput: 464.14 iter/sec.
Timings for 3200K FFT length (16 cores, 2 workers): Throughput: 602.66 iter/sec.
Timings for 3200K FFT length (16 cores, 4 workers): Throughput: 289.66 iter/sec.


____________

Profile composite
Volunteer tester
Send message
Joined: 16 Feb 10
Posts: 841
ID: 55391
Credit: 784,066,705
RAC: 430,409
Discovered 2 mega primesFound 1 prime in the 2018 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,477,467)Cullen LLR Gold: Earned 500,000 credits (776,297)ESP LLR Ruby: Earned 2,000,000 credits (3,433,680)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,093,491)PPS LLR Sapphire: Earned 20,000,000 credits (27,332,866)PSP LLR Turquoise: Earned 5,000,000 credits (6,587,988)SoB LLR Sapphire: Earned 20,000,000 credits (34,191,283)SR5 LLR Turquoise: Earned 5,000,000 credits (6,110,877)SGS LLR Ruby: Earned 2,000,000 credits (3,486,285)TRP LLR Turquoise: Earned 5,000,000 credits (7,070,795)Woodall LLR Amethyst: Earned 1,000,000 credits (1,693,614)321 Sieve (suspended) Emerald: Earned 50,000,000 credits (50,256,050)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (50,009,610)PPS Sieve Double Silver: Earned 200,000,000 credits (385,669,989)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Turquoise: Earned 5,000,000 credits (6,616,128)GFN Emerald: Earned 50,000,000 credits (53,771,465)WW Gold: Earned 500,000 credits (932,000)PSA Double Bronze: Earned 100,000,000 credits (102,762,384)
Message 149272 - Posted: 9 Mar 2021 | 8:08:03 UTC

There are other considerations for your subproject setup.

One is how much heat dissipation your CPU cooling can handle. When the FFT fits in cache your cores will idle a lot less than when they have to occasionally wait for data in RAM to fill a cache line, so the CPU will become hotter. This is where prime-finding has a reputation for stress-testing CPUs. I have seen a variation of 20 C within the same CPU between a core that is idle and a core that is busy, but only 10 C where the entire problem does not fit in cache.

If your CPU gets too hot, it protects itelf by slowing down its computations (called throttling). If you let this persist for too long your system wears out sooner (heat degrades the transistors in the CPU and eventually they stop working properly because they no longer respond correctly to the applied voltages). My i7-920 system gave up after 6 years of crunching. It was usual for that system to run at 90 C (I once saw it at 101 C). That's damaging for the CPU. The last couple of years before it died, I had to slow down the base clock to keep the computer stable (not crashing).

In Linux you can use the "sensors" command to see the current temperature of all your cores.

Another consideration is efficiency. The relationship between number of threads and task speedup has diminishing returns, so you will be spending more CPU cycles getting the same answer (in less wall clock time), and thus expending more energy per task with more threads.

Your machine's throughput (average tasks per day) is generally reduced when you are multithreading. Since you are doing fewer tasks per day, you have a smaller chance of finding a prime in a given period of time.

Efficiency directly impacts economics. The 2 computers (and 2 monitors) on my desk cost about CAD$350 per year in direct electricity costs @ $0.098/kWh (i.e. not counting the added cost of air conditioning). I have this number because they are plugged into a home theatre power conditioner which continuously displays line voltage and current drawn. I instantly see the effects of running various tasks - the current runs up to 3.9 amps AC. The 2 LED monitors are not insignificant - together they draw about 0.5 A when they are turned on but a negligible amount when in powersaving mode.

wolfman1360
Send message
Joined: 18 Feb 18
Posts: 56
ID: 979448
Credit: 64,200,115
RAC: 6
321 LLR Bronze: Earned 10,000 credits (69,252)Cullen LLR Silver: Earned 100,000 credits (434,949)ESP LLR Bronze: Earned 10,000 credits (86,507)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (186,794)PPS LLR Amethyst: Earned 1,000,000 credits (1,106,516)PSP LLR Silver: Earned 100,000 credits (111,724)SoB LLR Jade: Earned 10,000,000 credits (16,939,060)SR5 LLR Gold: Earned 500,000 credits (929,564)SGS LLR Bronze: Earned 10,000 credits (77,792)TRP LLR Amethyst: Earned 1,000,000 credits (1,979,586)Woodall LLR Silver: Earned 100,000 credits (240,757)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (3,181,160)PPS Sieve Jade: Earned 10,000,000 credits (16,895,452)AP 26/27 Amethyst: Earned 1,000,000 credits (1,277,588)GFN Jade: Earned 10,000,000 credits (19,782,869)WW Gold: Earned 500,000 credits (900,000)
Message 149274 - Posted: 9 Mar 2021 | 8:38:46 UTC - in response to Message 149269.

On the 3700x it's still faster to run anything above ~2M fft with 8 threads.

I keep seeing a reference to fft. What does this mean and where can I get more information?

wolfman1360
Send message
Joined: 18 Feb 18
Posts: 56
ID: 979448
Credit: 64,200,115
RAC: 6
321 LLR Bronze: Earned 10,000 credits (69,252)Cullen LLR Silver: Earned 100,000 credits (434,949)ESP LLR Bronze: Earned 10,000 credits (86,507)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (186,794)PPS LLR Amethyst: Earned 1,000,000 credits (1,106,516)PSP LLR Silver: Earned 100,000 credits (111,724)SoB LLR Jade: Earned 10,000,000 credits (16,939,060)SR5 LLR Gold: Earned 500,000 credits (929,564)SGS LLR Bronze: Earned 10,000 credits (77,792)TRP LLR Amethyst: Earned 1,000,000 credits (1,979,586)Woodall LLR Silver: Earned 100,000 credits (240,757)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (3,181,160)PPS Sieve Jade: Earned 10,000,000 credits (16,895,452)AP 26/27 Amethyst: Earned 1,000,000 credits (1,277,588)GFN Jade: Earned 10,000,000 credits (19,782,869)WW Gold: Earned 500,000 credits (900,000)
Message 149275 - Posted: 9 Mar 2021 | 8:45:54 UTC - in response to Message 149272.

There are other considerations for your subproject setup.

One is how much heat dissipation your CPU cooling can handle. When the FFT fits in cache your cores will idle a lot less than when they have to occasionally wait for data in RAM to fill a cache line, so the CPU will become hotter. This is where prime-finding has a reputation for stress-testing CPUs. I have seen a variation of 20 C within the same CPU between a core that is idle and a core that is busy, but only 10 C where the entire problem does not fit in cache.

If your CPU gets too hot, it protects itelf by slowing down its computations (called throttling). If you let this persist for too long your system wears out sooner (heat degrades the transistors in the CPU and eventually they stop working properly because they no longer respond correctly to the applied voltages). My i7-920 system gave up after 6 years of crunching. It was usual for that system to run at 90 C (I once saw it at 101 C). That's damaging for the CPU. The last couple of years before it died, I had to slow down the base clock to keep the computer stable (not crashing).

In Linux you can use the "sensors" command to see the current temperature of all your cores.

Another consideration is efficiency. The relationship between number of threads and task speedup has diminishing returns, so you will be spending more CPU cycles getting the same answer (in less wall clock time), and thus expending more energy per task with more threads.

Your machine's throughput (average tasks per day) is generally reduced when you are multithreading. Since you are doing fewer tasks per day, you have a smaller chance of finding a prime in a given period of time.

Efficiency directly impacts economics. The 2 computers (and 2 monitors) on my desk cost about CAD$350 per year in direct electricity costs @ $0.098/kWh (i.e. not counting the added cost of air conditioning). I have this number because they are plugged into a home theatre power conditioner which continuously displays line voltage and current drawn. I instantly see the effects of running various tasks - the current runs up to 3.9 amps AC. The 2 LED monitors are not insignificant - together they draw about 0.5 A when they are turned on but a negligible amount when in powersaving mode.

This is exactly the type of info I need.
My Ryzen 7 1800x is proving this point quite handily by running anywhere from 5-8 c above normal when at 100% utilization.
A lot of these older processors will be retired when Ryzen replacements and funds are available. My way of thinking is the money I spend on electricity will be made up in efficiency and power - and while I'm not paying for electricity now, it certainly doesn't hurt anything, especially with processors from 2009-2010 that refuse to stop working, most especially in the summer.

Clearly there is some extremely well executed code being run here.
I probably will not run this on the laptop for that reason - expensive machine I'd like to hold onto for a while, and I don't like crunching, even part time, with machines that do not have easily removable batteries. The x1 extreme can have it removed, but I need to remove the rear panel in order to access it. Not ideal.

Profile j.sheridanProject donor
Volunteer tester
Send message
Joined: 21 Mar 11
Posts: 816
ID: 91622
Credit: 2,342,089,257
RAC: 2,407,017
Discovered 2 mega primesFound 4 primes in the 2021 Tour de PrimesFound 2 mega primes in the 2021 Tour de Primes321 LLR Jade: Earned 10,000,000 credits (10,011,712)Cullen LLR Jade: Earned 10,000,000 credits (10,025,728)ESP LLR Jade: Earned 10,000,000 credits (10,176,490)Generalized Cullen/Woodall LLR Jade: Earned 10,000,000 credits (10,020,580)PPS LLR Sapphire: Earned 20,000,000 credits (29,281,953)PSP LLR Jade: Earned 10,000,000 credits (10,055,994)SoB LLR Sapphire: Earned 20,000,000 credits (40,958,919)SR5 LLR Jade: Earned 10,000,000 credits (10,885,368)SGS LLR Jade: Earned 10,000,000 credits (10,007,143)TRP LLR Jade: Earned 10,000,000 credits (10,009,531)Woodall LLR Jade: Earned 10,000,000 credits (10,039,527)321 Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,019,388)Cullen/Woodall Sieve (suspended) Double Silver: Earned 200,000,000 credits (265,102,350)PPS Sieve Double Gold: Earned 500,000,000 credits (547,992,812)TRP Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,801,812)AP 26/27 Emerald: Earned 50,000,000 credits (93,546,934)GFN Double Silver: Earned 200,000,000 credits (439,903,440)WW Double Gold: Earned 500,000,000 credits (808,248,000)
Message 149277 - Posted: 9 Mar 2021 | 8:49:54 UTC - in response to Message 149274.

On the 3700x it's still faster to run anything above ~2M fft with 8 threads.

I keep seeing a reference to fft. What does this mean and where can I get more information?


Leaving what it means for someone who can explain the maths better, if you click on a finished task (or on the stderr.txt file of a running task) you'll see something like this:

Using zero-padded FMA3 FFT length 2240K, Pass1=896, Pass2=2560, clm=1, 8 threads, a = 3, L2 = 447*361, M = 161369

Which tells you the size being used for the current task. Note that the size can vary within a sub-project and annoyingly for Ryzen 3000 users cross the boundary for optimum number of threads.

Profile dannyridel
Volunteer tester
Avatar
Send message
Joined: 3 Feb 19
Posts: 919
ID: 1097922
Credit: 20,095,563
RAC: 22,693
Discovered 1 mega prime321 LLR Gold: Earned 500,000 credits (506,806)Cullen LLR Silver: Earned 100,000 credits (112,212)ESP LLR Silver: Earned 100,000 credits (255,803)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (509,419)PPS LLR Ruby: Earned 2,000,000 credits (2,001,172)PSP LLR Gold: Earned 500,000 credits (616,940)SoB LLR Gold: Earned 500,000 credits (533,625)SR5 LLR Silver: Earned 100,000 credits (108,281)SGS LLR Gold: Earned 500,000 credits (501,474)TRP LLR Silver: Earned 100,000 credits (146,999)Woodall LLR Silver: Earned 100,000 credits (117,864)321 Sieve (suspended) Gold: Earned 500,000 credits (506,814)Generalized Cullen/Woodall Sieve (suspended) Bronze: Earned 10,000 credits (30,033)PPS Sieve Ruby: Earned 2,000,000 credits (2,808,043)AP 26/27 Amethyst: Earned 1,000,000 credits (1,273,545)GFN Ruby: Earned 2,000,000 credits (3,545,498)WW Turquoise: Earned 5,000,000 credits (6,148,000)PSA Silver: Earned 100,000 credits (373,034)
Message 149282 - Posted: 9 Mar 2021 | 11:53:00 UTC - in response to Message 149277.

On the 3700x it's still faster to run anything above ~2M fft with 8 threads.

I keep seeing a reference to fft. What does this mean and where can I get more information?


Leaving what it means for someone who can explain the maths better, if you click on a finished task (or on the stderr.txt file of a running task) you'll see something like this:

Using zero-padded FMA3 FFT length 2240K, Pass1=896, Pass2=2560, clm=1, 8 threads, a = 3, L2 = 447*361, M = 161369

Which tells you the size being used for the current task. Note that the size can vary within a sub-project and annoyingly for Ryzen 3000 users cross the boundary for optimum number of threads.

Adding on to that, the cache used for this task is the fft times 8.
In this case, 2240K*8=17.92MB.
____________
SHSIDElectronicsGroup@outlook.com

waiting for a TdP prime...
Proth "SoB": 44243*2^440969+1


wolfman1360
Send message
Joined: 18 Feb 18
Posts: 56
ID: 979448
Credit: 64,200,115
RAC: 6
321 LLR Bronze: Earned 10,000 credits (69,252)Cullen LLR Silver: Earned 100,000 credits (434,949)ESP LLR Bronze: Earned 10,000 credits (86,507)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (186,794)PPS LLR Amethyst: Earned 1,000,000 credits (1,106,516)PSP LLR Silver: Earned 100,000 credits (111,724)SoB LLR Jade: Earned 10,000,000 credits (16,939,060)SR5 LLR Gold: Earned 500,000 credits (929,564)SGS LLR Bronze: Earned 10,000 credits (77,792)TRP LLR Amethyst: Earned 1,000,000 credits (1,979,586)Woodall LLR Silver: Earned 100,000 credits (240,757)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (3,181,160)PPS Sieve Jade: Earned 10,000,000 credits (16,895,452)AP 26/27 Amethyst: Earned 1,000,000 credits (1,277,588)GFN Jade: Earned 10,000,000 credits (19,782,869)WW Gold: Earned 500,000 credits (900,000)
Message 149292 - Posted: 9 Mar 2021 | 22:31:56 UTC - in response to Message 149282.

On the 3700x it's still faster to run anything above ~2M fft with 8 threads.

I keep seeing a reference to fft. What does this mean and where can I get more information?


Leaving what it means for someone who can explain the maths better, if you click on a finished task (or on the stderr.txt file of a running task) you'll see something like this:

Using zero-padded FMA3 FFT length 2240K, Pass1=896, Pass2=2560, clm=1, 8 threads, a = 3, L2 = 447*361, M = 161369

Which tells you the size being used for the current task. Note that the size can vary within a sub-project and annoyingly for Ryzen 3000 users cross the boundary for optimum number of threads.

Adding on to that, the cache used for this task is the fft times 8.
In this case, 2240K*8=17.92MB.

Thanks. Exactly what I was looking for.
I'll keep an eye on this and see how I can best utilize my cores.
I'm assuming I want each workunit to use as few cores as possible but at the same time have the ability for it to still fit in cache?
I hope that makes sense.

Profile composite
Volunteer tester
Send message
Joined: 16 Feb 10
Posts: 841
ID: 55391
Credit: 784,066,705
RAC: 430,409
Discovered 2 mega primesFound 1 prime in the 2018 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,477,467)Cullen LLR Gold: Earned 500,000 credits (776,297)ESP LLR Ruby: Earned 2,000,000 credits (3,433,680)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,093,491)PPS LLR Sapphire: Earned 20,000,000 credits (27,332,866)PSP LLR Turquoise: Earned 5,000,000 credits (6,587,988)SoB LLR Sapphire: Earned 20,000,000 credits (34,191,283)SR5 LLR Turquoise: Earned 5,000,000 credits (6,110,877)SGS LLR Ruby: Earned 2,000,000 credits (3,486,285)TRP LLR Turquoise: Earned 5,000,000 credits (7,070,795)Woodall LLR Amethyst: Earned 1,000,000 credits (1,693,614)321 Sieve (suspended) Emerald: Earned 50,000,000 credits (50,256,050)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (50,009,610)PPS Sieve Double Silver: Earned 200,000,000 credits (385,669,989)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Turquoise: Earned 5,000,000 credits (6,616,128)GFN Emerald: Earned 50,000,000 credits (53,771,465)WW Gold: Earned 500,000 credits (932,000)PSA Double Bronze: Earned 100,000,000 credits (102,762,384)
Message 149293 - Posted: 9 Mar 2021 | 23:38:30 UTC - in response to Message 149292.

I'm assuming I want each workunit to use as few cores as possible but at the same time have the ability for it to still fit in cache?
I hope that makes sense.

Not exactly. You want the total size of all the tasks being worked on simultaneously to fit in the cache, regardless of how many threads are running. If you have an 8-core machine but your CPU cache holds only 2 FFTs, then you should run 2 tasks with at least 1 (and at most 4) threads each.

The operating system is responsible for allocating cores to threads dynamically, but you can force a thread to be run on a particular core (or set of cores) to optimize the use of CPU cache; using "Process Lasso" in Windows, in Linux setting CPU affinity (or cgroups if you are so advanced). With cgroups, if I read this correctly, you could even exclude entire cores from Linux at boot time. However cgroups would be better used to run 2 instances of BOINC to keep them from stomping on each other's assigned cores; maybe one dedicated for feeding a GPU, or to help older AMD CPUs to confine tasks to individual CCX's (but Linux already does a better job at this than Windows).

One of the 6 cores in my i7-5280k always runs hotter than the rest, either because of uneven heat transfer from the CPU to the cooler, or because something is electrically sub-par in that core, but not completely out of spec. That's probably the reason this chip was binned as an i7-5820K instead of an i7-5930K (same part, just different speeds, with some PCIe lanes disabled to guarantee that that they don't look like identical parts).

If I exclude this "damaged" core from Linux I may get a CPU that can be run harder. I doubt however that I could make it go at 3.96 GHz which would be needed to make up for the loss of one working core @ 3.3 GHz. On the other hand, I could simply let BOINC use all the cores except this one to get roughly the same effect.

Profile BurProject donor
Volunteer tester
Avatar
Send message
Joined: 25 Feb 20
Posts: 462
ID: 1241833
Credit: 279,947,726
RAC: 547,227
321 LLR Ruby: Earned 2,000,000 credits (2,092,823)Cullen LLR Ruby: Earned 2,000,000 credits (2,002,841)ESP LLR Amethyst: Earned 1,000,000 credits (1,445,099)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,180,764)PPS LLR Amethyst: Earned 1,000,000 credits (1,225,852)PSP LLR Ruby: Earned 2,000,000 credits (2,064,832)SoB LLR Amethyst: Earned 1,000,000 credits (1,669,219)SR5 LLR Ruby: Earned 2,000,000 credits (2,065,004)SGS LLR Amethyst: Earned 1,000,000 credits (1,202,156)TRP LLR Ruby: Earned 2,000,000 credits (2,089,856)Woodall LLR Ruby: Earned 2,000,000 credits (2,112,258)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,107,153)PPS Sieve Turquoise: Earned 5,000,000 credits (5,080,097)AP 26/27 Ruby: Earned 2,000,000 credits (4,697,966)GFN Turquoise: Earned 5,000,000 credits (7,489,336)WW Double Silver: Earned 200,000,000 credits (239,400,000)PSA Amethyst: Earned 1,000,000 credits (1,022,470)
Message 149358 - Posted: 11 Mar 2021 | 18:40:21 UTC

To get a quick estimate of maximal throughput you can use Prime95. Under Options -> Benchmark enter the FFT values to test and also the range of workers (this is threads) to test. Check "complex FFT" box. I guess it's not 100% comparable to a LLR2 run, but it's probably good for estimating what works and what doesn't.

If you want an accurate test and are familiar with the command prompt, you can download LLR (this should be the newest version) and run a test:

llr2 -tx -d -q"1155*2^1234567+1"


Replace the -tx with -t2, -t3 or whatever number of threads to run. The exponent 1234567 can be chosen according to the subproject. Current exponent n can be checked on the subproject status page. The k = 1155 doesn't have much influence, just change it if LLR2 reports a small factor.

What you want to look at is the "Time per iteration". Multiply this by the number of threads and the setting that gives the smallest product is the winner.

Don't forget to pause BOINC while doing the testing.
____________
1281979 * 2^485014 + 1 is prime ... no further hits up to: n = 4,800,000

Werinbert
Send message
Joined: 9 Jun 13
Posts: 127
ID: 233452
Credit: 309,500,569
RAC: 212,938
321 LLR Gold: Earned 500,000 credits (784,049)Cullen LLR Gold: Earned 500,000 credits (529,186)ESP LLR Gold: Earned 500,000 credits (513,315)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (521,209)PPS LLR Turquoise: Earned 5,000,000 credits (8,770,986)PSP LLR Gold: Earned 500,000 credits (566,492)SoB LLR Gold: Earned 500,000 credits (597,244)SR5 LLR Gold: Earned 500,000 credits (665,436)SGS LLR Gold: Earned 500,000 credits (585,571)TRP LLR Gold: Earned 500,000 credits (562,418)Woodall LLR Gold: Earned 500,000 credits (615,831)321 Sieve (suspended) Jade: Earned 10,000,000 credits (10,014,922)Generalized Cullen/Woodall Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,101,470)PPS Sieve Double Bronze: Earned 100,000,000 credits (100,344,557)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Gold: Earned 500,000 credits (715,073)TRP Sieve (suspended) Gold: Earned 500,000 credits (630,833)AP 26/27 Sapphire: Earned 20,000,000 credits (33,338,578)GFN Sapphire: Earned 20,000,000 credits (38,531,667)WW Double Bronze: Earned 100,000,000 credits (100,076,000)PSA Jade: Earned 10,000,000 credits (10,047,861)
Message 149366 - Posted: 11 Mar 2021 | 23:54:15 UTC - in response to Message 149358.

Don't forget to pause BOINC while doing the testing.


"Pause BOINC?" I don't think my computers understand that comment......

wolfman1360
Send message
Joined: 18 Feb 18
Posts: 56
ID: 979448
Credit: 64,200,115
RAC: 6
321 LLR Bronze: Earned 10,000 credits (69,252)Cullen LLR Silver: Earned 100,000 credits (434,949)ESP LLR Bronze: Earned 10,000 credits (86,507)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (186,794)PPS LLR Amethyst: Earned 1,000,000 credits (1,106,516)PSP LLR Silver: Earned 100,000 credits (111,724)SoB LLR Jade: Earned 10,000,000 credits (16,939,060)SR5 LLR Gold: Earned 500,000 credits (929,564)SGS LLR Bronze: Earned 10,000 credits (77,792)TRP LLR Amethyst: Earned 1,000,000 credits (1,979,586)Woodall LLR Silver: Earned 100,000 credits (240,757)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (3,181,160)PPS Sieve Jade: Earned 10,000,000 credits (16,895,452)AP 26/27 Amethyst: Earned 1,000,000 credits (1,277,588)GFN Jade: Earned 10,000,000 credits (19,782,869)WW Gold: Earned 500,000 credits (900,000)
Message 149368 - Posted: 12 Mar 2021 | 8:07:03 UTC - in response to Message 149358.

To get a quick estimate of maximal throughput you can use Prime95. Under Options -> Benchmark enter the FFT values to test and also the range of workers (this is threads) to test. Check "complex FFT" box. I guess it's not 100% comparable to a LLR2 run, but it's probably good for estimating what works and what doesn't.

If you want an accurate test and are familiar with the command prompt, you can download LLR (this should be the newest version) and run a test:

llr2 -tx -d -q"1155*2^1234567+1"


Replace the -tx with -t2, -t3 or whatever number of threads to run. The exponent 1234567 can be chosen according to the subproject. Current exponent n can be checked on the subproject status page. The k = 1155 doesn't have much influence, just change it if LLR2 reports a small factor.

What you want to look at is the "Time per iteration". Multiply this by the number of threads and the setting that gives the smallest product is the winner.

Don't forget to pause BOINC while doing the testing.

I am still more apt to break things when I start monkeying around in the terminal, but we'll see how I get in the coming weeks.
I hope to participate in the SoB challenge coming up.
My little Ivy bridge i5's appear to be doing particularly well in terms of time completed on these sieves, at least from what I can tell. 4 true cores per WU they're holding their own, at least for now. I think I might have one more sitting in a drawer, I just need to hunt down a motherboard and some ram.
From what I'm seeing these SoB tasks take significantly longer, though, but I haven't completed one yet. I'm running with 0 cache.

Profile composite
Volunteer tester
Send message
Joined: 16 Feb 10
Posts: 841
ID: 55391
Credit: 784,066,705
RAC: 430,409
Discovered 2 mega primesFound 1 prime in the 2018 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,477,467)Cullen LLR Gold: Earned 500,000 credits (776,297)ESP LLR Ruby: Earned 2,000,000 credits (3,433,680)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,093,491)PPS LLR Sapphire: Earned 20,000,000 credits (27,332,866)PSP LLR Turquoise: Earned 5,000,000 credits (6,587,988)SoB LLR Sapphire: Earned 20,000,000 credits (34,191,283)SR5 LLR Turquoise: Earned 5,000,000 credits (6,110,877)SGS LLR Ruby: Earned 2,000,000 credits (3,486,285)TRP LLR Turquoise: Earned 5,000,000 credits (7,070,795)Woodall LLR Amethyst: Earned 1,000,000 credits (1,693,614)321 Sieve (suspended) Emerald: Earned 50,000,000 credits (50,256,050)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (50,009,610)PPS Sieve Double Silver: Earned 200,000,000 credits (385,669,989)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Turquoise: Earned 5,000,000 credits (6,616,128)GFN Emerald: Earned 50,000,000 credits (53,771,465)WW Gold: Earned 500,000 credits (932,000)PSA Double Bronze: Earned 100,000,000 credits (102,762,384)
Message 149369 - Posted: 12 Mar 2021 | 8:11:50 UTC - in response to Message 149358.
Last modified: 12 Mar 2021 | 8:24:54 UTC

If you want an accurate test and are familiar with the command prompt, you can download LLR (this should be the newest version) and run a test:

llr2 -tx -d -q"1155*2^1234567+1"


Replace the -tx with -t2, -t3 or whatever number of threads to run. The exponent 1234567 can be chosen according to the subproject. Current exponent n can be checked on the subproject status page. The k = 1155 doesn't have much influence, just change it if LLR2 reports a small factor.

What you want to look at is the "Time per iteration". Multiply this by the number of threads and the setting that gives the smallest product is the winner.

Sorry to disagree with you, I think you have it wrong. The devil is in the details.
The overall approach is valid because you are concerned about relative timings.

1) You don't have to download LLR2 from PrimeGrid if it already ran in BOINC.
It's on your computer in the primegrid subdirectory of BOINC projects.
For Linux: /var/lib/boinc/projects/www.primegrid.com/sllr2_1.1.0_linux64_201114

2) Do not multiply the ms per bit by the number of threads.
The time per bit already shows the effect of using multiple threads.
You'll get to 40,000 bits with 4 threads a lot sooner than you would with 1 thread.

3) Don't use the first time you see (at 10,000 bits).
It includes extra real time related to initially loading cache lines and rebalancing CPU processes.
At higher thread counts ~ 10, bit time doesn't stabilize to +/-5% until around 40,000 bits @ 3.3 GHz.
Stabilization seems to take a fixed amount of time,
so you must compensate for processing more bits in that fixed amount of time
when using more threads, by using the sample at a higher bit count.

4) Regarding your depiction of accuracy, even dropping the outlier at 10,000 bits (up to 12% higher than the stable value),
these timings are repeatable to just +/-5% because of other things happening in the computer.
You could get very accurate timing of program performance if the boot loader directly started the LLR2
executable statically linked with everything it needs to run without an operating system.
I saw this much difference with N=24 samples on Linux.
For reference, 5% of 24 hours is 72 minutes.

I tried a single Fermat Divisor task whose FFT size is 512k (so 4 MB) which fits well within in my CPU's L3 cache of 12 MB.
The first ms per bit report (at 10,000 bits) is always higher than the second report (at 20,000 bits) no matter how many threads I use.
The difference is around 2% with 1 thread, to over 6% with 11 threads.
So there's some startup time in the 10,000 bit ms report that should be excluded
when dealing with tasks whose FFT fits in cache. Ignoring the 10,000 bit report would be prudent for any task.

I created a new subdirectory (in my home directory) and copied the LLR executable to it, along with this script.
The script runs 3 trials of 50-second tests for each thread setting from 1 to 12 (to include HT on a 6-core CPU).
!/bin/bash rm -f times.txt for ((t=1;t<13;t++)); do for ((i=1;i<4;i++)); do rm -f llr.ini r* z* sync # flush the disk buffers sleep 10 # give the computer lots of time to settle down ./sllr2_1.1.0_linux64_201114 -t$t -d -q17*2^8983247+1 | tee -a times.txt & sleep 50 kill %1 wait done done # you'll want to convert <ctrl-M> to newlines in times.txt by your favorite method before loading into a spreadsheet or some other mathematical tool.

Post to thread

Message boards : Number crunching : a variety of CPU generations, overwelmed by all the subprojects in a good way

[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 3.59, 2.56, 2.13
Generated 21 Sep 2021 | 21:19:13 UTC