4 Prime95 threads plus GPU doing another project

RickC · 2010-02-11, 01:59

I have a machine with an AMD Phenom II X4, 4GB of DDR3 1600 RAM, and integrated graphics. Then I decided rather than add more RAM I would add a video card with 1GB DDR3 RAM so that I get to use the full 4GB system RAM. I added a Radeon HD 4670 that has a Zalman heatsink and supposedly can sustain half a TFLOPS. I have plenty of cooling with a 200mm top fan and 120mm rear fan, and a 700W power supply.

Has anyone successfully tried running 4 Prime95 threads along with another project that uses the GPU at the same time? Any details or advice?

Thanks

cheesehead · 2010-02-11, 04:00

This isn't a direct response, but ...

Have you looked at the Hardware subforum (http://www.mersenneforum.org/forumdisplay.php?f=9) threads about using GPUs for GIMPS-related work?

Its sticky thread "The prime-crunching on dedicated hardware FAQ (II)" (http://www.mersenneforum.org/showthread.php?t=12720) summarizes the situation as of a few months ago.

Some folks are doing trial division factoring (not an under-resourced area of GIMPS, however) on their graphics cards. See "Trial division with CUDA" at http://www.mersenneforum.org/showthread.php?t=12827

There's also "The ATI GPU Computing thread" (yes, I see you're not using an ATI) at http://www.mersenneforum.org/showthread.php?t=12875

But perhaps you're asking about some more well-established GPU-using project, and that would presently be outside GIMPS. :-)

RickC · 2010-02-11, 05:00

I understand that GIMPS is highly optimized for CPUs but I always heard that there are other projects like folding proteins that can use GPUs which have a much higher GFLOPS for other types of math.

I was just wondering if the two co-exist well on the same machine.

mdettweiler · 2010-02-11, 05:10

Quote:

Originally Posted by RickC

I understand that GIMPS is highly optimized for CPUs but I always heard that there are other projects like folding proteins that can use GPUs which have a much higher GFLOPS for other types of math.

I was just wondering if the two co-exist well on the same machine.

I'm not an expert on this stuff, but from what I've heard GIMPS will co-exist reasonably well with GPU applications running simultaneously. Some GPU applications use more of the CPU than others; if the application is particularly CPU-intensive, it may cut into GIMPS throughput on one core somewhat. (The reverse, GIMPS cutting into the CPU side of the GPU application, may also occur, though from what I've heard it seems to be the other way around.) I think Folding@Home specifically (which from your comments I'm guessing you're using) has been reported to play pretty well with Prime95.

BTW, even though GIMPS has been highly optimized for CPUs, development efforts for it on GPUs has advanced enough by now that both LL and TF actually run faster now on some GPUs than a state-of-the-art CPU (the only caveat being that the LL application must run on one of the newer double-precision GPUs). Probably the main drawback of this is that the GPU applications are not nearly as automated as Prime95--you have to reserve, run, and report results for exponents manually.

RickC · 2010-02-11, 07:02

I haven't tried any distributed projects other than GIMPS but it seems to me like I might have some wasted resources lying idle.

Two of my motherboards are this below that have built-in graphics and two graphics card slots:
http://www.newegg.com/Product/Produc...82E16813128398
The GPUs sit idle. I have a Radeon HD 4670 with 512MB RAM, the new Radeon HD 4670 with 1GB RAM in another machine, and a GeForce GT 220 with 1GB RAM.

I figured I should do something with them. They are all in well cooled cases with good 80 plus certified power supplies. One is in a 55F cold basement. Another is in a 63F basement. I'm sure they are cold and bored and could do a lot of work. Weird how one Radeon GPU has a higher GLOPS than a 4 core CPU. The Radeon can do about 480 GFLOPS while the 4 core processor can only do about 64 GFLOPS if using SSE3 extensions. I know they do different types of math though.

I was hoping someone else could tell me what to do with the GPUs because I'm only familiar with GIMPS right now but I temporarily have tons of time on my hands to configure stuff and do my taxes.

RMAC9.5 · 2010-02-11, 08:09

RickC,
Like you I didn't want to have some wasted resources lying idle and fold for Ars Technica Team EggRoll (team number 14). I have 3 PCs with "decent" integrated MB graphics and 2 PCs with Sapphire ATI HD3850 AGP video cards that run Folding@home and Prime95 on Windows XP (32 & 64 bit). They play nicely together when the "right" environmental variables are set properly as documented in one of the ATI GPU Folding@Home threads. For example, the Phenom II 940 BE in my Gigabyte GA-MA78GM-US2H motherboard is currently running 2 M43 LLs and 2 M22 DCs. Iteration times are 0.052, 0.053, 0.024, and 0.024 sec. BROOK_YIELD = 2 is the Windows environmental variable that I added to this PC to reduce the amount of CPU used to feed data to the Folding GPU software running on the MB IGP. Your ATI HD4670 cards are well thought of by the Folding@Home enthusiasts and your GeForce GT 220 isn't bad either. While I am not as expert as many of the folks on the Folding@Home forums, I can and would be happy to provide some help if this is how you decide to use your GPUs.

Brain · 2010-02-11, 18:31

I'm running 4 prime95 threads on a core i5-750 and BOINC "Collatz Conjecture" (http://boinc.thesonntags.com/collatz/) on my ATI HD5770: no measurable slowdown for Prime95 when BOINC running on GPU.

RickC · 2010-02-12, 17:33

Thanks. I found a good list here:

http://boinc.berkeley.edu/wiki/GPU_computing

db597 · 2010-02-14, 17:54

I'm running Prime95 on my Q6600 CPU and F@H on my GPU. The impact depends on what graphics card you have. Previously when I was using a HD3870, there was a large load on the CPU when running the F@H client (easily 50%+ load on 1 core). But after switching to a GTX260, there is minimal CPU impact (practically zero).

The difference in CPU load is an issue of how the F@H client is implemented - ATI cards simply have a higher CPU load, and also produce less points per day. This may change when F@H moves to the v3 GPU client.

RMAC9.5 · 2010-02-15, 07:14

ATI cards can be tweaked by the use of some ATI specific Window's environmental settings to use a much smaller amount of CPU. The Folding@Home ATI forum has a thread which explains how to apply these settings. The CPU usage percentage can also be adjusted via a slider setting located on the Advanced tab of the Folding@Home Control Panel.

2010-02-11, 01:59	#1
RickC Mar 2003 55₈ Posts	4 Prime95 threads plus GPU doing another project I have a machine with an AMD Phenom II X4, 4GB of DDR3 1600 RAM, and integrated graphics. Then I decided rather than add more RAM I would add a video card with 1GB DDR3 RAM so that I get to use the full 4GB system RAM. I added a Radeon HD 4670 that has a Zalman heatsink and supposedly can sustain half a TFLOPS. I have plenty of cooling with a 200mm top fan and 120mm rear fan, and a 700W power supply. Has anyone successfully tried running 4 Prime95 threads along with another project that uses the GPU at the same time? Any details or advice? Thanks

2010-02-11, 04:00	#2
cheesehead "Richard B. Woods" Aug 2002 Wisconsin USA 2²·3·641 Posts	This isn't a direct response, but ... Have you looked at the Hardware subforum (http://www.mersenneforum.org/forumdisplay.php?f=9) threads about using GPUs for GIMPS-related work? Its sticky thread "The prime-crunching on dedicated hardware FAQ (II)" (http://www.mersenneforum.org/showthread.php?t=12720) summarizes the situation as of a few months ago. Some folks are doing trial division factoring (not an under-resourced area of GIMPS, however) on their graphics cards. See "Trial division with CUDA" at http://www.mersenneforum.org/showthread.php?t=12827 There's also "The ATI GPU Computing thread" (yes, I see you're not using an ATI) at http://www.mersenneforum.org/showthread.php?t=12875 But perhaps you're asking about some more well-established GPU-using project, and that would presently be outside GIMPS. :-) Last fiddled with by cheesehead on 2010-02-11 at 04:10

2010-02-11, 08:09	#6
RMAC9.5 Jun 2003 3²·17 Posts	RickC, Like you I didn't want to have some wasted resources lying idle and fold for Ars Technica Team EggRoll (team number 14). I have 3 PCs with "decent" integrated MB graphics and 2 PCs with Sapphire ATI HD3850 AGP video cards that run Folding@home and Prime95 on Windows XP (32 & 64 bit). They play nicely together when the "right" environmental variables are set properly as documented in one of the ATI GPU Folding@Home threads. For example, the Phenom II 940 BE in my Gigabyte GA-MA78GM-US2H motherboard is currently running 2 M43 LLs and 2 M22 DCs. Iteration times are 0.052, 0.053, 0.024, and 0.024 sec. BROOK_YIELD = 2 is the Windows environmental variable that I added to this PC to reduce the amount of CPU used to feed data to the Folding GPU software running on the MB IGP. Your ATI HD4670 cards are well thought of by the Folding@Home enthusiasts and your GeForce GT 220 isn't bad either. While I am not as expert as many of the folks on the Folding@Home forums, I can and would be happy to provide some help if this is how you decide to use your GPUs. Last fiddled with by RMAC9.5 on 2010-02-11 at 08:17

2010-02-11, 18:31	#7
Brain Dec 2009 Peine, Germany 14B₁₆ Posts	Collatz Conjecture for ATI GPU I'm running 4 prime95 threads on a core i5-750 and BOINC "Collatz Conjecture" (http://boinc.thesonntags.com/collatz/) on my ATI HD5770: no measurable slowdown for Prime95 when BOINC running on GPU.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Prime95 minor bug, "TortureThreads" not checked for maximum threads	Explorer09	Software	1	2014-08-19 04:13
[Prime95] How to choose number of worker threads?	Sohjin	Information & Answers	1	2014-03-09 14:14
Maximum number of threads for Prime95	treesong	Information & Answers	11	2012-06-18 13:35
Workers, Threads, Helper Threads, Cores, Affinity.	lorgix	Information & Answers	12	2011-01-13 22:31
P-1 with 2 threads?	lycorn	Software	7	2009-09-13 03:19

2010-02-11, 05:00	#3
RickC Mar 2003 45₁₀ Posts	I understand that GIMPS is highly optimized for CPUs but I always heard that there are other projects like folding proteins that can use GPUs which have a much higher GFLOPS for other types of math. I was just wondering if the two co-exist well on the same machine.

2010-02-11, 07:02	#5
RickC Mar 2003 45₁₀ Posts	I haven't tried any distributed projects other than GIMPS but it seems to me like I might have some wasted resources lying idle. Two of my motherboards are this below that have built-in graphics and two graphics card slots: http://www.newegg.com/Product/Produc...82E16813128398 The GPUs sit idle. I have a Radeon HD 4670 with 512MB RAM, the new Radeon HD 4670 with 1GB RAM in another machine, and a GeForce GT 220 with 1GB RAM. I figured I should do something with them. They are all in well cooled cases with good 80 plus certified power supplies. One is in a 55F cold basement. Another is in a 63F basement. I'm sure they are cold and bored and could do a lot of work. Weird how one Radeon GPU has a higher GLOPS than a 4 core CPU. The Radeon can do about 480 GFLOPS while the 4 core processor can only do about 64 GFLOPS if using SSE3 extensions. I know they do different types of math though. I was hoping someone else could tell me what to do with the GPUs because I'm only familiar with GIMPS right now but I temporarily have tons of time on my hands to configure stuff and do my taxes.

2010-02-12, 17:33	#8
RickC Mar 2003 3²×5 Posts	Thanks. I found a good list here: http://boinc.berkeley.edu/wiki/GPU_computing

2010-02-14, 17:54	#9
db597 Jan 2003 2×103 Posts	I'm running Prime95 on my Q6600 CPU and F@H on my GPU. The impact depends on what graphics card you have. Previously when I was using a HD3870, there was a large load on the CPU when running the F@H client (easily 50%+ load on 1 core). But after switching to a GTX260, there is minimal CPU impact (practically zero). The difference in CPU load is an issue of how the F@H client is implemented - ATI cards simply have a higher CPU load, and also produce less points per day. This may change when F@H moves to the v3 GPU client.

2010-02-15, 07:14	#10
RMAC9.5 Jun 2003 3²·17 Posts	ATI cards can be tweaked by the use of some ATI specific Window's environmental settings to use a much smaller amount of CPU. The Folding@Home ATI forum has a thread which explains how to apply these settings. The CPU usage percentage can also be adjusted via a slider setting located on the Advanced tab of the Folding@Home Control Panel.