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 : Proth Prime Search : Is my GPU slow?

Author Message
Profile DaveBProject donor
Avatar
Send message
Joined: 20 Jun 09
Posts: 351
ID: 42198
Credit: 11,898,570
RAC: 0
321 LLR Gold: Earned 500,000 credits (547,769)Cullen LLR Silver: Earned 100,000 credits (158,867)PPS LLR Silver: Earned 100,000 credits (162,534)PSP LLR Silver: Earned 100,000 credits (134,373)SoB LLR Gold: Earned 500,000 credits (510,012)SR5 LLR Silver: Earned 100,000 credits (266,757)SGS LLR Gold: Earned 500,000 credits (510,560)TRP LLR Silver: Earned 100,000 credits (242,439)Woodall LLR Gold: Earned 500,000 credits (508,989)321 Sieve (suspended) Silver: Earned 100,000 credits (200,123)Cullen/Woodall Sieve (suspended) Silver: Earned 100,000 credits (229,801)PPS Sieve Ruby: Earned 2,000,000 credits (4,317,221)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Bronze: Earned 10,000 credits (21,147)TRP Sieve (suspended) Silver: Earned 100,000 credits (333,494)AP 26/27 Bronze: Earned 10,000 credits (24,117)GFN Ruby: Earned 2,000,000 credits (3,730,320)
Message 42158 - Posted: 25 Oct 2011 | 14:22:04 UTC

Hi,

I have recently installed a GPU (CAL ATI Radeon HD5x00 series (Cedar) (512MB) driver: 1.4.1546) and find I am taking about 9 hours per PPS Sieve. I am running 2 x Woodall on the CPU ( http://www.primegrid.com/show_host_detail.php?hostid=171718 )

Is this normal? or is something wrong in my settings? Most reported times seem to be in minutes rather than hours for a GPU.

Any advice would be appreciated.[/url]
____________
Member team AUSTRALIA
My lucky number is 9291*2^1085585+1

Profile Pooh Bear 27Project donor
Avatar
Send message
Joined: 10 May 09
Posts: 710
ID: 39821
Credit: 408,642,319
RAC: 15,108
Discovered 1 mega primeEliminated 1 conjecture "k"Discovered 1 twin prime321 LLR Turquoise: Earned 5,000,000 credits (5,014,405)Cullen LLR Turquoise: Earned 5,000,000 credits (5,734,854)ESP LLR Turquoise: Earned 5,000,000 credits (5,421,149)Generalized Cullen/Woodall LLR Turquoise: Earned 5,000,000 credits (5,576,209)PPS LLR Jade: Earned 10,000,000 credits (17,389,837)PSP LLR Jade: Earned 10,000,000 credits (10,425,342)SoB LLR Turquoise: Earned 5,000,000 credits (8,986,628)SR5 LLR Turquoise: Earned 5,000,000 credits (5,248,659)SGS LLR Turquoise: Earned 5,000,000 credits (6,878,856)TRP LLR Turquoise: Earned 5,000,000 credits (5,416,106)Woodall LLR Turquoise: Earned 5,000,000 credits (6,400,237)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,000,576)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,000,977)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (6,843,367)PPS Sieve Emerald: Earned 50,000,000 credits (78,673,438)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Turquoise: Earned 5,000,000 credits (6,473,750)TRP Sieve (suspended) Turquoise: Earned 5,000,000 credits (8,243,608)AP 26/27 Jade: Earned 10,000,000 credits (12,531,383)WW Amethyst: Earned 1,000,000 credits (1,148,000)GFN Double Silver: Earned 200,000,000 credits (202,192,765)PSA Turquoise: Earned 5,000,000 credits (5,042,125)
Message 42159 - Posted: 25 Oct 2011 | 14:36:25 UTC - in response to Message 42158.

Time is shown as CPU time. So you will not see the GPU time currently. I believe it's being looked at but we've been crunching GPU for a couple years now and it's still not being shown.

Profile rroonnaallddProject donor
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
321 LLR Silver: Earned 100,000 credits (101,692)Cullen LLR Silver: Earned 100,000 credits (104,876)ESP LLR Silver: Earned 100,000 credits (101,979)PPS LLR Silver: Earned 100,000 credits (148,018)PSP LLR Silver: Earned 100,000 credits (140,441)SoB LLR Silver: Earned 100,000 credits (119,475)SR5 LLR Silver: Earned 100,000 credits (120,939)SGS LLR Silver: Earned 100,000 credits (122,783)TRP LLR Silver: Earned 100,000 credits (100,115)Woodall LLR Silver: Earned 100,000 credits (107,459)321 Sieve (suspended) Silver: Earned 100,000 credits (202,757)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (6,908,135)PPS Sieve Sapphire: Earned 20,000,000 credits (25,450,104)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Silver: Earned 100,000 credits (130,966)TRP Sieve (suspended) Silver: Earned 100,000 credits (201,525)AP 26/27 Silver: Earned 100,000 credits (100,015)GFN Silver: Earned 100,000 credits (246,369)PSA Silver: Earned 100,000 credits (226,594)
Message 42160 - Posted: 25 Oct 2011 | 14:39:07 UTC - in response to Message 42158.
Last modified: 25 Oct 2011 | 14:42:55 UTC

PPSsieve on ATI/AMD uses OpenCL and this is not a native language neither for ATI/AMD nor nVidia.
OpenCL is more like Java, runs on every platform but fast is something different.


[edit]
@PB27
Run time (sec) and CPU time (sec) are displayed for every workunit and all times are wall-clock times...
____________
Best wishes. Knowledge is power. by jjwhalen

Profile ConanProject donor
Avatar
Send message
Joined: 24 Mar 09
Posts: 219
ID: 37336
Credit: 54,051,837
RAC: 619
321 LLR Gold: Earned 500,000 credits (500,863)Cullen LLR Gold: Earned 500,000 credits (522,970)ESP LLR Gold: Earned 500,000 credits (516,922)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (543,312)PPS LLR Gold: Earned 500,000 credits (685,350)PSP LLR Amethyst: Earned 1,000,000 credits (1,019,589)SoB LLR Amethyst: Earned 1,000,000 credits (1,055,482)SR5 LLR Gold: Earned 500,000 credits (522,539)SGS LLR Gold: Earned 500,000 credits (532,524)TRP LLR Gold: Earned 500,000 credits (564,910)Woodall LLR Amethyst: Earned 1,000,000 credits (1,022,336)321 Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,398,786)Cullen/Woodall Sieve (suspended) Silver: Earned 100,000 credits (204,463)Generalized Cullen/Woodall Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,015,683)PPS Sieve Sapphire: Earned 20,000,000 credits (37,926,771)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,226,498)TRP Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,515,247)AP 26/27 Gold: Earned 500,000 credits (509,418)WW Bronze: Earned 10,000 credits (36,000)GFN Ruby: Earned 2,000,000 credits (2,506,619)PSA Silver: Earned 100,000 credits (225,555)
Message 42212 - Posted: 26 Oct 2011 | 23:44:50 UTC
Last modified: 26 Oct 2011 | 23:49:28 UTC

When I was running my ATI 5870 cards along with the last Woodall challenge I was doing a WU in around 35 or so minutes per Proth WU (a lot ran longer than this).
So with 4 cards I could only muster an maximum RAC of under 500,000 (or there abouts).

Compared to the Nvidia cards it just wasn't in the same league so I put the cards on DNETC and Moo (just Moo now as DNETC has folded).

From the listed stats it seems that 2 Nvidia GTX570 cards can easily outdo this figure. They do WUs in under 15 minutes.

So I will await till I can get some Nvidia cards for Primegrid work. ATI cards fly at Moo so they will stay there with a fling at Milkyway now and then.

So in answer to your question, Is your GPU slow? Well probably not just not as suited to this project as some others. However taking many hours on the GPU sounds wrong, are you sure that your are running a GPU WU as the CPU work units take the 9 hour run time you mentioned.
Under your preferences you just tick the GPU box if you only want the GPU work, ticking the CPU box as well you will get both types.

Conan
____________


ZydorProject donor
Avatar
Send message
Joined: 27 Nov 10
Posts: 226
ID: 74718
Credit: 25,180,844
RAC: 0
PPS LLR Gold: Earned 500,000 credits (683,896)SR5 LLR Bronze: Earned 10,000 credits (65,374)SGS LLR Silver: Earned 100,000 credits (198,823)PPS Sieve Jade: Earned 10,000,000 credits (12,095,504)PSA Jade: Earned 10,000,000 credits (12,131,472)
Message 42240 - Posted: 27 Oct 2011 | 14:04:23 UTC
Last modified: 27 Oct 2011 | 14:06:57 UTC

DaveB

The stedrr output is normal and showing reality time:
==============
<core_client_version>6.10.60</core_client_version>
<![CDATA[
<stderr_txt>
Sieve started: 58408512000000000 <= p < 58408518000000000
Thread 0 starting
Detected 32 multiprocessors (160 SPUs) on device 0.

Thread 0 completed
Sieve complete: 58408512000000000 <= p < 58408518000000000
count=155401791,sum=0xa7a10b0445914621
Elapsed time: 32991.39 sec. (1.43 init + 32989.96 sieve) at 181880 p/sec.
Processor time: 1084.49 sec. (1.40 init + 1083.08 sieve) at 5539935 p/sec.
Average processor utilization: 0.99 (init), 0.03 (sieve)
04:43:30 (4944): called boinc_finish

</stderr_txt>
]]>
================
It does however show a lower end 5xxx card ("Detected 32 multiprocessors (160 SPUs) on device 0."), and that together with the fact that the AMD application is on OpenCL, not any native ATI language means it will be markedly slower than the CUDA cards - fact of life with this application and current stage of development of OpenCL. Slower 5xxx + OpanCL is going to mean the timing is about right.

(Very much worth updating your 6.10.60 BOINC Client to the current 6.12.33 if multiple GPU on a card, and 6.12.34 if single GPU on a card. A lot of work was done between 6.10.XX and 6.12.XX on working with GPUs)

What is the actual card you are using ?

Profile rroonnaallddProject donor
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
321 LLR Silver: Earned 100,000 credits (101,692)Cullen LLR Silver: Earned 100,000 credits (104,876)ESP LLR Silver: Earned 100,000 credits (101,979)PPS LLR Silver: Earned 100,000 credits (148,018)PSP LLR Silver: Earned 100,000 credits (140,441)SoB LLR Silver: Earned 100,000 credits (119,475)SR5 LLR Silver: Earned 100,000 credits (120,939)SGS LLR Silver: Earned 100,000 credits (122,783)TRP LLR Silver: Earned 100,000 credits (100,115)Woodall LLR Silver: Earned 100,000 credits (107,459)321 Sieve (suspended) Silver: Earned 100,000 credits (202,757)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (6,908,135)PPS Sieve Sapphire: Earned 20,000,000 credits (25,450,104)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Silver: Earned 100,000 credits (130,966)TRP Sieve (suspended) Silver: Earned 100,000 credits (201,525)AP 26/27 Silver: Earned 100,000 credits (100,015)GFN Silver: Earned 100,000 credits (246,369)PSA Silver: Earned 100,000 credits (226,594)
Message 42241 - Posted: 27 Oct 2011 | 14:20:59 UTC

All PPS-workunits from DaveB are listed as "PPS (Sieve) v1.38 (ati13ati)" and the string "(ati13ati)" is used by the ATI-app.
The "CPU time (sec)" is reported by the boinc-client for calculating a unit on one cpu-core. This time can but must not correspond to the "Run time (sec)" because the "Run time" is reported by the app. In most cases the times for "CPU time" and "Run time" varies only a few seconds. This will change if you calculate multi-threaded apps like Aqua@Home or use a GPU.
In Aqua the "Run time" was ever "CPU time" multiplicated with the count of available cores for the boinc client or the count of cores reglemented by an app_info-file. The same for running a GPU-app, only the point of view is different. "CPU time" is ever the time, the boinc-client has registered for calculating a workunit on the cpu and this should be only a few seconds in optimal cases while using a GPU-app. "Run time" means time reported by the GPU-app for calculating a WU.

PPSsieve on Cuda needs only some CPU time to load the workunit on the GPU or at the end to copy all files from the slot directory back to the PG-folder. In dependance of the used driver-version, the CPU time can be increased. Some drivers are known to cause a higher CPU-usage while doing GPGPU-work than needed. Fastest driver in most cases was version 256.X, if your card is supported by this.

PPSsieve with OpenCL (ATI/AMD and Mac OS X) is a different story. OpenCL has either an instruction limitation and some calulations can only be done by the CPU or can not be optimized like Cuda because of the missing documentation from ATI/AMD.

Ken Brazier (nVidia, ATI) and Iain Bethune (Mac OS X) have invited every user to optimize their gpu-code. Maybe an other user is able to write a better app or can write a CAL-version of the PPSsieve-app. A CAL-version would speed up all ATI-GPUs on Linux and Windows...
____________
Best wishes. Knowledge is power. by jjwhalen

Message boards : Proth Prime Search : Is my GPU slow?

[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 2.83, 3.14, 3.34
Generated 27 Feb 2021 | 17:07:35 UTC