mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing

Reply
 
Thread Tools
Old 2020-07-24, 22:55   #3279
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

22·13·31 Posts
Default

There is something I feel I need to pass on. I received a new GPU today. A GTX 1650. The older CUDA 10 version of mfaktc will not run with it. The newer 2047 runs very well.

Off-topic: I suspect gpuOwl will run with it. I have my doubts about CUDALucas and CUDAPm1. I will have to try each.

It is running around 950 GHz-d/day. The temperature is 64°C. The clock speeds are slightly less than my 1080. Everything on the back of the machine is relatively cool to the touch. So, it appears everything will be fine. If my 1080 decides to die, I know what I will replace it with. For the cost, I have no complaints at all.
storm5510 is offline   Reply With Quote
Old 2020-07-28, 19:46   #3280
TheJudger
 
TheJudger's Avatar
 
"Oliver"
Mar 2005
Germany

33·41 Posts
Default

Hi,

seems like mfaktc runs fine with CUDA 11 on Ampere (no specific changes for Ampere except Makefile).

Code:
mfaktc v0.22-pre8 (64bit built)
[...]
CUDA version info
  binary compiled for CUDA  11.0
  CUDA runtime version      11.0
  CUDA driver version       11.0

CUDA device info
  name                      A100-SXM4-40GB
  compute capability        8.0
  max threads per block     1024
  max shared memory per MP  167936 byte
  number of multiprocessors 108
  clock rate (CUDA cores)   1410MHz
  memory clock rate:        1215MHz
  memory bus width:         5120 bit
[...]
Starting trial factoring M66362159 from 2^74 to 2^75 (57.65 GHz-days)
 k_min =  142321062303420
 k_max =  284642124610180
Using GPU kernel "barrett76_mul32_gs"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
Jul 19 21:19 |    0   0.1% |  0.829  13m15s |   6259.18    82485    n.a.%
Jul 19 21:19 |    4   0.2% |  0.779  12m26s |   6660.92    82485    n.a.%
Jul 19 21:19 |    9   0.3% |  0.780  12m26s |   6652.38    82485    n.a.%
[...]
Jul 19 21:31 | 4617 100.0% |  0.780   0m00s |   6652.38    82485    n.a.%
no factor for M66362159 from 2^74 to 2^75 [mfaktc 0.22-pre8 barrett76_mul32_gs CUDA 11.0 arch 8.0] 51D74917
tf(): total time spent: 12m 32.323s
New absolute performance champion and I guess best performance per watt, too!


Older benchmark data for Turing (RTX 2080 Ti): https://mersenneforum.org/showpost.p...postcount=2912

Oliver
TheJudger is offline   Reply With Quote
Old 2020-07-28, 20:37   #3281
James Heinrich
 
James Heinrich's Avatar
 
"James Heinrich"
May 2004
ex-Northern Ontario

3·1,039 Posts
Default

Quote:
Originally Posted by TheJudger View Post
New absolute performance champion and I guess best performance per watt, too!
Oooh, juicy new benchmark! It is indeed the current performance champion, but (according to my numbers) the Telsa T4 is still the efficiency champion (70W vs 400W is a big difference).

If you have access to this GPU again and can run a quick CUDAlucas and/or GPUowl benchmark I would appreciate it.
James Heinrich is offline   Reply With Quote
Old 2020-07-28, 20:46   #3282
TheJudger
 
TheJudger's Avatar
 
"Oliver"
Mar 2005
Germany

33·41 Posts
Default

Hi James,

I can't remember a T4 hitting 2500 GHz-d/d. And while that A100 has a TDP of 400 W it reports during mfaktc "just" 290 to 300 W. But you still might be correct that a T4 has a better performance per watt ratio when looking just at the power consumption of the GPU itself. I tend to ignore those smaller GPUs, sorry.

Oliver

Last fiddled with by TheJudger on 2020-07-28 at 21:15
TheJudger is offline   Reply With Quote
Old 2020-07-28, 21:00   #3283
James Heinrich
 
James Heinrich's Avatar
 
"James Heinrich"
May 2004
ex-Northern Ontario

3·1,039 Posts
Default

Quote:
Originally Posted by TheJudger View Post
I tend to ignore those smaller GPUs, sorry.
Those "smaller" GPUs that still have nearly 5x the TF performance of my RX 480
James Heinrich is offline   Reply With Quote
Old 2020-07-28, 21:12   #3284
James Heinrich
 
James Heinrich's Avatar
 
"James Heinrich"
May 2004
ex-Northern Ontario

3×1,039 Posts
Default

Quote:
Originally Posted by TheJudger View Post
I can't remember a T4 hitting 2500 THz-d/d.
Neither can I. Not even 2500 GHz-d/d

I actually looked back and found one actual benchmark for a T4 that put it at around 1700/day. On my chart it was being lumped in with the other CUDA 7.5 cards, I have adjusted my data so the numbers should be a little more accurate. Still more efficient than the A100, but only 50% better not 100% better.
James Heinrich is offline   Reply With Quote
Old 2020-07-28, 21:17   #3285
TheJudger
 
TheJudger's Avatar
 
"Oliver"
Mar 2005
Germany

33×41 Posts
Default

I guess for many of us a RX 480 is far more enjoyable than a T4 for home usage (PC games )
TheJudger is offline   Reply With Quote
Old 2020-07-29, 01:19   #3286
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

22·13·31 Posts
Default

I only use the two I have now for this project. I don't play games with them. I watch the power consumption. 200W for one and 75W for the other. 6.6 kWh in 24 hours, if I keep them running continuously, which I do not. In simpler terms, about $25 USD a month for both at the current rate. That's no back-breaker.

Utility costs here are quite steady. An average would be around $0.13 USD per kWh. There are members here to pay far more. If I had to pay what some are, there wold be a lot of oil lamps and candles used. I don't see how they manage.
storm5510 is offline   Reply With Quote
Old 2020-07-29, 12:40   #3287
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

23×3×191 Posts
Default

Quote:
Originally Posted by storm5510 View Post
Utility costs here are quite steady. An average would be around $0.13 USD per kWh. There are members here to pay far more. If I had to pay what some are, there wold be a lot of oil lamps and candles used. I don't see how they manage.
Price paraffin candles on a $/Btu and $/lumen-hour basis, and electricity even at $1/kw-hr won't look so bad. The same goes for kerosene lamps.
Have you tried using nvidia-smi to reduce power use on your gpus? I find running RTX20xx or GTX1650 gpus at 50% power still provides 80% of TF throughput. It's a lot easier on the air conditioner too, so power and cost savings are considerable, even at my ~$.12/kwhr.
Code:
:d0 gtx1080 90 to 291 W
"c:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe" -i 0 -pl 90

:d1 gtx1650 45 to 75 w
"c:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe" -i 1 -pl 45

:d2 gtx1650 45 to 90 w
"c:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe" -i 2 -pl 45

:d3 rtx2080 125 to 258 w
"c:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe" -i 3 -pl 125

:d4 gtx1080ti 125 to 300 w
"c:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe" -i 4 -pl 125

Last fiddled with by kriesel on 2020-07-29 at 12:50
kriesel is online now   Reply With Quote
Old 2020-07-29, 15:54   #3288
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

22·13·31 Posts
Default

Quote:
Originally Posted by kriesel View Post
Price paraffin candles on a $/Btu and $/lumen-hour basis, and electricity even at $1/kw-hr won't look so bad. The same goes for kerosene lamps.
Have you tried using nvidia-smi to reduce power use on your gpus? I find running RTX20xx or GTX1650 gpus at 50% power still provides 80% of TF throughput. It's a lot easier on the air conditioner too, so power and cost savings are considerable, even at my ~$.12/kwhr.
Two years ago when I started using this 1080, my UPS would begin to "whistle" after running for a few seconds. I used MSI Afterburner to throttle the GPU back to 85% of capacity. It was at, or beyond, the capacity of the UPS. It is really not large enough having a maximum capacity of 300W. Now, I no longer need to throttle. The GPU's overall performance has dropped around 10%. It still runs above a thousand gigahertz days per day running mfaktc.

My average monthly utility cost is $115 USD during the warm weather months. In the winter is where I have to be cautious. My furnace has two 20 ampere heating elements. It is 240 VAC. It can eat a decent size hole in my pocket if I fail to pay attention.
storm5510 is offline   Reply With Quote
Old 2020-07-29, 16:03   #3289
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2×4,643 Posts
Default

Quote:
Originally Posted by storm5510 View Post
In the winter is where I have to be cautious. My furnace has two 20 ampere heating elements. It is 240 VAC. It can eat a decent size hole in my pocket if I fail to pay attention.
Isn't that what GPUs are really for? Space heaters???
chalsall is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
mfakto: an OpenCL program for Mersenne prefactoring Bdot GPU Computing 1656 2020-10-13 14:21
The P-1 factoring CUDA program firejuggler GPU Computing 752 2020-09-08 16:15
"CUDA runtime version 0.0" when running mfaktc.exe froderik GPU Computing 4 2016-10-30 15:29
World's second-dumbest CUDA program fivemack Programming 112 2015-02-12 22:51
World's dumbest CUDA program? xilman Programming 1 2009-11-16 10:26

All times are UTC. The time now is 09:19.

Fri Oct 23 09:19:38 UTC 2020 up 43 days, 6:30, 0 users, load averages: 1.27, 1.44, 1.63

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.