Low / Inconsistent PPD with Radeon 7950 & Core 17

It seems that a lot of GPU problems revolve around specific versions of drivers. Though AMD has their own support structure, you can often learn from information reported by others who fold.

Moderators: Site Moderators, FAHC Science Team

bmwman91
Posts: 20
Joined: Tue Feb 17, 2009 1:33 am

Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by bmwman91 »

I have a couple of new (old) machines to add to my collection of number crunchers, and I am hoping that someone can point me in the right direction to get them cranking out research at as high a rate as possible. Both machines use Opteron 6272 CPUs and Radeon HD 7950 GPUs on stock clocks (for now). One has 8GB of RAM and the other 12GB. Both have slow 5400RPM freebie HDDs (500GB and 1TB). Both were wiped and had a fresh Win8.1x64 install done on them, and based on what I read on here I installed Catalyst 14.9 drivers. Also, both are connected to the same monitor, one via DVI and the other via a DVI=>VGA adapter.

When I first fired up the F@H client and let it run for a couple of hours, I was seeing ~16K PPD from both CPUs, 290K PPD from one of the GPUs (project 13000) and 13K PDD from the other GPU (project 13001). TPF was 2:47 on the one machine with a fast-rising completion percentage to match, and 22:50 on the other. I checked the Catalyst performance panel and both seemed to be getting 100% GPU utilization. I then swapped the monitor cables between the 2 machines (again, both use DVI output but one has a DVI=>VGA adapter since the old monitor has one of each type of input), and I paused/restarted the F@H clients. After that, both were seeing ~13K PPD after an hour. So, I uninstalled the F@H client (I know it is bad practice and I very rarely do it) and fully removed the Catalyst drivers, and reinstalled both (sticking with Catalyst 14.9). After a few hours, both machines were sitting at ~12% completion on 13000 and 13001 WUs, with ~27K PPD and ~22:50 TPF.

SO...was the 290K PPD and 2:47 TPD that I was seeing on project 13000 a fluke of some sort (didn't seem like it since the progress bar was going quite fast)? From looking at some of the GPU PPD databases, it seems like a 7950 should be seeing 70K - 120K PPD, although I am not sure which core version those numbers were with. Anyway, can anyone that has been folding with an HD 7950 chime in about what sort of performance I should be expecting from Core 17 WUs? I will have access to the machines again tomorrow, so if the good folks here need to see logs I can pull them then. Is it still the case that Catalyst 14.9 is preferred for older GPUs rather than the latest version? Are there any secret driver settings hidden somewhere that could be hindering performance? This is probably a totally redundant question at this point too, but is there any dependence on what is connected to the GPU output (I saw mention of dummy plugs being needed on an older folding client)?

Thanks!
Image
Mstenholm
Posts: 84
Joined: Fri Oct 22, 2010 10:17 pm
Hardware configuration: 4 x GTX 970. Win 7.

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by Mstenholm »

That was a fluke. You are right about the 70 - 120k PPD.
bmwman91
Posts: 20
Joined: Tue Feb 17, 2009 1:33 am

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by bmwman91 »

Thanks for the confirmation. So, ~26K PDD is not right, and I guess I will have to mess around with drivers and whatnot to see if I can rectify the situation. The performance of Core 17 units on AMD cards relies on OpenCL drivers and such, correct?

Here is a log from one of the machines (both look about the same).

Code: Select all

*********************** Log Started 2015-01-07T23:15:50Z ***********************
23:15:50:************************* Folding@home Client *************************
23:15:50:      Website: http://folding.stanford.edu/
23:15:50:    Copyright: (c) 2009-2014 Stanford University
23:15:50:       Author: Joseph Coffland <joseph@cauldrondevelopment.com>
23:15:50:         Args: --open-web-control
23:15:50:       Config: C:/Users/Administrator/AppData/Roaming/FAHClient/config.xml
23:15:50:******************************** Build ********************************
23:15:50:      Version: 7.4.4
23:15:50:         Date: Mar 4 2014
23:15:50:         Time: 20:26:54
23:15:50:      SVN Rev: 4130
23:15:50:       Branch: fah/trunk/client
23:15:50:     Compiler: Intel(R) C++ MSVC 1500 mode 1200
23:15:50:      Options: /TP /nologo /EHa /Qdiag-disable:4297,4103,1786,279 /Ox -arch:SSE
23:15:50:               /QaxSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 /Qopenmp /Qrestrict /MT /Qmkl
23:15:50:     Platform: win32 XP
23:15:50:         Bits: 32
23:15:50:         Mode: Release
23:15:50:******************************* System ********************************
23:15:50:          CPU: AMD Opteron(TM) Processor 6272
23:15:50:       CPU ID: AuthenticAMD Family 21 Model 1 Stepping 2
23:15:50:         CPUs: 16
23:15:50:       Memory: 7.98GiB
23:15:50:  Free Memory: 6.40GiB
23:15:50:      Threads: WINDOWS_THREADS
23:15:50:   OS Version: 6.2
23:15:50:  Has Battery: false
23:15:50:   On Battery: false
23:15:50:   UTC Offset: -8
23:15:50:          PID: 4696
23:15:50:          CWD: C:/Users/Administrator/AppData/Roaming/FAHClient
23:15:50:           OS: Windows 8.1 Pro
23:15:50:      OS Arch: AMD64
23:15:50:         GPUs: 1
23:15:50:        GPU 0: ATI:5 Tahiti PRO [Radeon R9 280/HD 7900/8950]
23:15:50:         CUDA: Not detected
23:15:50:Win32 Service: false
23:15:50:***********************************************************************
23:15:50:<config>
23:15:50:  <!-- Slot Control -->
23:15:50:  <power v='FULL'/>
23:15:50:
23:15:50:  <!-- User Information -->
23:15:50:  <passkey v='********************************'/>
23:15:50:  <team v='#####'/>
23:15:50:  <user v='XXXXX'/>
23:15:50:
23:15:50:  <!-- Folding Slots -->
23:15:50:  <slot id='0' type='CPU'/>
23:15:50:  <slot id='1' type='GPU'/>
23:15:50:</config>
23:15:50:Trying to access database...
23:15:51:Successfully acquired database lock
23:15:51:Enabled folding slot 00: READY cpu:15
23:15:51:Enabled folding slot 01: READY gpu:0:Tahiti PRO [Radeon R9 280/HD 7900/8950]
23:15:51:WU00:FS00:Starting
23:15:51:WU00:FS00:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:/Users/Administrator/AppData/Roaming/FAHClient/cores/web.stanford.edu/~pande/Win32/AMD64/Core_a3.fah/FahCore_a3.exe -dir 00 -suffix 01 -version 704 -lifeline 4696 -checkpoint 15 -np 15
23:15:53:WU00:FS00:Started FahCore on PID 4648
23:15:54:WU00:FS00:Core PID:5244
23:15:54:WU00:FS00:FahCore 0xa3 started
23:15:55:WU01:FS01:Starting
23:15:55:WU01:FS01:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:/Users/Administrator/AppData/Roaming/FAHClient/cores/web.stanford.edu/~pande/Win32/AMD64/ATI/R600/Core_17.fah/FahCore_17.exe -dir 01 -suffix 01 -version 704 -lifeline 4696 -checkpoint 15 -gpu 0 -gpu-vendor ati
23:15:55:WU01:FS01:Started FahCore on PID 4924
23:15:55:WU00:FS00:0xa3:
23:15:55:WU00:FS00:0xa3:*------------------------------*
23:15:55:WU00:FS00:0xa3:Folding@Home Gromacs SMP Core
23:15:55:WU00:FS00:0xa3:Version 2.27 (Dec. 15, 2010)
23:15:55:WU00:FS00:0xa3:
23:15:55:WU00:FS00:0xa3:Preparing to commence simulation
23:15:55:WU00:FS00:0xa3:- Ensuring status. Please wait.
23:15:58:WU01:FS01:Core PID:5312
23:15:58:WU01:FS01:FahCore 0x17 started
23:15:59:WU01:FS01:0x17:*********************** Log Started 2015-01-07T23:15:58Z ***********************
23:15:59:WU01:FS01:0x17:Project: 13000 (Run 461, Clone 0, Gen 101)
23:15:59:WU01:FS01:0x17:Unit: 0x000000cc538b3db753101df7640bb293
23:15:59:WU01:FS01:0x17:CPU: 0x00000000000000000000000000000000
23:15:59:WU01:FS01:0x17:Machine: 1
23:15:59:WU01:FS01:0x17:Digital signatures verified
23:15:59:WU01:FS01:0x17:Folding@home GPU core17
23:15:59:WU01:FS01:0x17:Version 0.0.52
23:16:04:WU00:FS00:0xa3:- Looking at optimizations...
23:16:04:WU00:FS00:0xa3:- Working with standard loops on this execution.
23:16:04:WU00:FS00:0xa3:- Previous termination of core was improper.
23:16:04:WU00:FS00:0xa3:- Files status OK
23:16:05:WU00:FS00:0xa3:- Expanded 3792565 -> 4163444 (decompressed 109.7 percent)
23:16:05:WU00:FS00:0xa3:Called DecompressByteArray: compressed_data_size=3792565 data_size=4163444, decompressed_data_size=4163444 diff=0
23:16:05:WU00:FS00:0xa3:- Digital signature verified
23:16:05:WU00:FS00:0xa3:
23:16:05:WU00:FS00:0xa3:Project: 8557 (Run 1, Clone 2, Gen 417)
23:16:05:WU00:FS00:0xa3:
23:16:05:WU00:FS00:0xa3:Entering M.D.
23:16:11:WU00:FS00:0xa3:Using Gromacs checkpoints
23:16:13:WU00:FS00:0xa3:Mapping NT from 15 to 15 
23:16:17:WU00:FS00:0xa3:Resuming from checkpoint
23:16:17:WU00:FS00:0xa3:Verified 00/wudata_01.log
23:16:17:WU00:FS00:0xa3:Verified 00/wudata_01.trr
23:16:17:WU00:FS00:0xa3:Verified 00/wudata_01.edr
23:16:17:WU00:FS00:0xa3:Completed 6365 out of 500000 steps  (1%)
23:24:06:WU01:FS01:0x17:Completed 0 out of 5000000 steps (0%)
23:24:06:WU01:FS01:0x17:Temperature control disabled. Requirements: single Nvidia GPU, tmax must be < 110 and twait >= 900
23:24:59:WU00:FS00:0xa3:Completed 10000 out of 500000 steps  (2%)
23:36:35:WU00:FS00:0xa3:Completed 15000 out of 500000 steps  (3%)
23:47:02:WU01:FS01:0x17:Completed 50000 out of 5000000 steps (1%)
23:48:09:WU00:FS00:0xa3:Completed 20000 out of 500000 steps  (4%)
23:59:41:WU00:FS00:0xa3:Completed 25000 out of 500000 steps  (5%)
00:09:26:WU01:FS01:0x17:Completed 100000 out of 5000000 steps (2%)
00:11:19:WU00:FS00:0xa3:Completed 30000 out of 500000 steps  (6%)
00:22:57:WU00:FS00:0xa3:Completed 35000 out of 500000 steps  (7%)
00:32:29:WU01:FS01:0x17:Completed 150000 out of 5000000 steps (3%)
00:34:30:WU00:FS00:0xa3:Completed 40000 out of 500000 steps  (8%)
00:46:03:WU00:FS00:0xa3:Completed 45000 out of 500000 steps  (9%)
00:54:53:WU01:FS01:0x17:Completed 200000 out of 5000000 steps (4%)
00:57:37:WU00:FS00:0xa3:Completed 50000 out of 500000 steps  (10%)
01:09:09:WU00:FS00:0xa3:Completed 55000 out of 500000 steps  (11%)
01:17:19:WU01:FS01:0x17:Completed 250000 out of 5000000 steps (5%)
01:20:45:WU00:FS00:0xa3:Completed 60000 out of 500000 steps  (12%)
01:32:16:WU00:FS00:0xa3:Completed 65000 out of 500000 steps  (13%)
01:40:24:WU01:FS01:0x17:Completed 300000 out of 5000000 steps (6%)
01:43:44:WU00:FS00:0xa3:Completed 70000 out of 500000 steps  (14%)
01:55:13:WU00:FS00:0xa3:Completed 75000 out of 500000 steps  (15%)
02:02:48:WU01:FS01:0x17:Completed 350000 out of 5000000 steps (7%)
02:06:44:WU00:FS00:0xa3:Completed 80000 out of 500000 steps  (16%)
02:18:21:WU00:FS00:0xa3:Completed 85000 out of 500000 steps  (17%)
02:25:49:WU01:FS01:0x17:Completed 400000 out of 5000000 steps (8%)
02:29:53:WU00:FS00:0xa3:Completed 90000 out of 500000 steps  (18%)
02:41:28:WU00:FS00:0xa3:Completed 95000 out of 500000 steps  (19%)
02:48:10:WU01:FS01:0x17:Completed 450000 out of 5000000 steps (9%)
02:52:59:WU00:FS00:0xa3:Completed 100000 out of 500000 steps  (20%)
03:04:30:WU00:FS00:0xa3:Completed 105000 out of 500000 steps  (21%)
03:10:31:WU01:FS01:0x17:Completed 500000 out of 5000000 steps (10%)
03:16:09:WU00:FS00:0xa3:Completed 110000 out of 500000 steps  (22%)
03:27:44:WU00:FS00:0xa3:Completed 115000 out of 500000 steps  (23%)
03:33:33:WU01:FS01:0x17:Completed 550000 out of 5000000 steps (11%)
03:39:13:WU00:FS00:0xa3:Completed 120000 out of 500000 steps  (24%)
03:50:43:WU00:FS00:0xa3:Completed 125000 out of 500000 steps  (25%)
03:55:55:WU01:FS01:0x17:Completed 600000 out of 5000000 steps (12%)
04:02:12:WU00:FS00:0xa3:Completed 130000 out of 500000 steps  (26%)
04:13:47:WU00:FS00:0xa3:Completed 135000 out of 500000 steps  (27%)
04:18:58:WU01:FS01:0x17:Completed 650000 out of 5000000 steps (13%)
04:25:17:WU00:FS00:0xa3:Completed 140000 out of 500000 steps  (28%)
04:36:48:WU00:FS00:0xa3:Completed 145000 out of 500000 steps  (29%)
04:41:22:WU01:FS01:0x17:Completed 700000 out of 5000000 steps (14%)
04:48:18:WU00:FS00:0xa3:Completed 150000 out of 500000 steps  (30%)
04:59:47:WU00:FS00:0xa3:Completed 155000 out of 500000 steps  (31%)
05:03:45:WU01:FS01:0x17:Completed 750000 out of 5000000 steps (15%)
05:11:18:WU00:FS00:0xa3:Completed 160000 out of 500000 steps  (32%)
******************************* Date: 2015-01-08 *******************************
05:22:47:WU00:FS00:0xa3:Completed 165000 out of 500000 steps  (33%)
05:26:49:WU01:FS01:0x17:Completed 800000 out of 5000000 steps (16%)
05:34:15:WU00:FS00:0xa3:Completed 170000 out of 500000 steps  (34%)
05:45:38:WU00:FS00:0xa3:Completed 175000 out of 500000 steps  (35%)
05:49:12:WU01:FS01:0x17:Completed 850000 out of 5000000 steps (17%)
05:57:05:WU00:FS00:0xa3:Completed 180000 out of 500000 steps  (36%)
06:08:38:WU00:FS00:0xa3:Completed 185000 out of 500000 steps  (37%)
06:12:14:WU01:FS01:0x17:Completed 900000 out of 5000000 steps (18%)
06:20:06:WU00:FS00:0xa3:Completed 190000 out of 500000 steps  (38%)
06:31:35:WU00:FS00:0xa3:Completed 195000 out of 500000 steps  (39%)
06:34:36:WU01:FS01:0x17:Completed 950000 out of 5000000 steps (19%)
06:43:01:WU00:FS00:0xa3:Completed 200000 out of 500000 steps  (40%)
06:54:28:WU00:FS00:0xa3:Completed 205000 out of 500000 steps  (41%)
06:56:58:WU01:FS01:0x17:Completed 1000000 out of 5000000 steps (20%)
07:06:02:WU00:FS00:0xa3:Completed 210000 out of 500000 steps  (42%)
07:17:33:WU00:FS00:0xa3:Completed 215000 out of 500000 steps  (43%)
07:20:00:WU01:FS01:0x17:Completed 1050000 out of 5000000 steps (21%)
07:29:02:WU00:FS00:0xa3:Completed 220000 out of 500000 steps  (44%)
07:40:35:WU00:FS00:0xa3:Completed 225000 out of 500000 steps  (45%)
07:42:23:WU01:FS01:0x17:Completed 1100000 out of 5000000 steps (22%)
07:52:07:WU00:FS00:0xa3:Completed 230000 out of 500000 steps  (46%)
08:03:41:WU00:FS00:0xa3:Completed 235000 out of 500000 steps  (47%)
08:05:25:WU01:FS01:0x17:Completed 1150000 out of 5000000 steps (23%)
08:15:11:WU00:FS00:0xa3:Completed 240000 out of 500000 steps  (48%)
08:26:41:WU00:FS00:0xa3:Completed 245000 out of 500000 steps  (49%)
08:27:49:WU01:FS01:0x17:Completed 1200000 out of 5000000 steps (24%)
08:38:10:WU00:FS00:0xa3:Completed 250000 out of 500000 steps  (50%)
08:49:37:WU00:FS00:0xa3:Completed 255000 out of 500000 steps  (51%)
08:50:12:WU01:FS01:0x17:Completed 1250000 out of 5000000 steps (25%)
09:01:08:WU00:FS00:0xa3:Completed 260000 out of 500000 steps  (52%)
09:12:42:WU00:FS00:0xa3:Completed 265000 out of 500000 steps  (53%)
09:13:19:WU01:FS01:0x17:Completed 1300000 out of 5000000 steps (26%)
09:24:09:WU00:FS00:0xa3:Completed 270000 out of 500000 steps  (54%)
09:35:36:WU00:FS00:0xa3:Completed 275000 out of 500000 steps  (55%)
09:35:44:WU01:FS01:0x17:Completed 1350000 out of 5000000 steps (27%)
09:47:02:WU00:FS00:0xa3:Completed 280000 out of 500000 steps  (56%)
09:58:30:WU00:FS00:0xa3:Completed 285000 out of 500000 steps  (57%)
09:58:48:WU01:FS01:0x17:Completed 1400000 out of 5000000 steps (28%)
10:09:56:WU00:FS00:0xa3:Completed 290000 out of 500000 steps  (58%)
10:21:12:WU01:FS01:0x17:Completed 1450000 out of 5000000 steps (29%)
10:21:22:WU00:FS00:0xa3:Completed 295000 out of 500000 steps  (59%)
10:32:50:WU00:FS00:0xa3:Completed 300000 out of 500000 steps  (60%)
10:43:41:WU01:FS01:0x17:Completed 1500000 out of 5000000 steps (30%)
10:44:24:WU00:FS00:0xa3:Completed 305000 out of 500000 steps  (61%)
10:55:50:WU00:FS00:0xa3:Completed 310000 out of 500000 steps  (62%)
11:06:46:WU01:FS01:0x17:Completed 1550000 out of 5000000 steps (31%)
11:07:21:WU00:FS00:0xa3:Completed 315000 out of 500000 steps  (63%)
11:18:47:WU00:FS00:0xa3:Completed 320000 out of 500000 steps  (64%)
******************************* Date: 2015-01-08 *******************************
11:29:09:WU01:FS01:0x17:Completed 1600000 out of 5000000 steps (32%)
11:30:11:WU00:FS00:0xa3:Completed 325000 out of 500000 steps  (65%)
11:41:41:WU00:FS00:0xa3:Completed 330000 out of 500000 steps  (66%)
11:52:15:WU01:FS01:0x17:Completed 1650000 out of 5000000 steps (33%)
11:53:09:WU00:FS00:0xa3:Completed 335000 out of 500000 steps  (67%)
12:04:33:WU00:FS00:0xa3:Completed 340000 out of 500000 steps  (68%)
12:14:40:WU01:FS01:0x17:Completed 1700000 out of 5000000 steps (34%)
12:15:58:WU00:FS00:0xa3:Completed 345000 out of 500000 steps  (69%)
12:27:23:WU00:FS00:0xa3:Completed 350000 out of 500000 steps  (70%)
12:37:04:WU01:FS01:0x17:Completed 1750000 out of 5000000 steps (35%)
12:38:53:WU00:FS00:0xa3:Completed 355000 out of 500000 steps  (71%)
12:50:19:WU00:FS00:0xa3:Completed 360000 out of 500000 steps  (72%)
13:00:10:WU01:FS01:0x17:Completed 1800000 out of 5000000 steps (36%)
13:01:46:WU00:FS00:0xa3:Completed 365000 out of 500000 steps  (73%)
13:13:10:WU00:FS00:0xa3:Completed 370000 out of 500000 steps  (74%)
13:22:34:WU01:FS01:0x17:Completed 1850000 out of 5000000 steps (37%)
13:24:37:WU00:FS00:0xa3:Completed 375000 out of 500000 steps  (75%)
13:36:06:WU00:FS00:0xa3:Completed 380000 out of 500000 steps  (76%)
13:45:37:WU01:FS01:0x17:Completed 1900000 out of 5000000 steps (38%)
13:47:33:WU00:FS00:0xa3:Completed 385000 out of 500000 steps  (77%)
13:58:55:WU00:FS00:0xa3:Completed 390000 out of 500000 steps  (78%)
14:08:00:WU01:FS01:0x17:Completed 1950000 out of 5000000 steps (39%)
14:10:23:WU00:FS00:0xa3:Completed 395000 out of 500000 steps  (79%)
14:21:49:WU00:FS00:0xa3:Completed 400000 out of 500000 steps  (80%)
14:30:22:WU01:FS01:0x17:Completed 2000000 out of 5000000 steps (40%)
14:33:17:WU00:FS00:0xa3:Completed 405000 out of 500000 steps  (81%)
14:44:42:WU00:FS00:0xa3:Completed 410000 out of 500000 steps  (82%)
14:53:24:WU01:FS01:0x17:Completed 2050000 out of 5000000 steps (41%)
14:56:11:WU00:FS00:0xa3:Completed 415000 out of 500000 steps  (83%)
15:07:39:WU00:FS00:0xa3:Completed 420000 out of 500000 steps  (84%)
15:15:47:WU01:FS01:0x17:Completed 2100000 out of 5000000 steps (42%)
15:19:08:WU00:FS00:0xa3:Completed 425000 out of 500000 steps  (85%)
15:30:38:WU00:FS00:0xa3:Completed 430000 out of 500000 steps  (86%)
15:38:51:WU01:FS01:0x17:Completed 2150000 out of 5000000 steps (43%)
15:42:09:WU00:FS00:0xa3:Completed 435000 out of 500000 steps  (87%)
15:53:40:WU00:FS00:0xa3:Completed 440000 out of 500000 steps  (88%)
16:01:15:WU01:FS01:0x17:Completed 2200000 out of 5000000 steps (44%)
16:05:07:WU00:FS00:0xa3:Completed 445000 out of 500000 steps  (89%)
16:16:38:WU00:FS00:0xa3:Completed 450000 out of 500000 steps  (90%)
16:23:39:WU01:FS01:0x17:Completed 2250000 out of 5000000 steps (45%)
16:28:09:WU00:FS00:0xa3:Completed 455000 out of 500000 steps  (91%)
16:39:40:WU00:FS00:0xa3:Completed 460000 out of 500000 steps  (92%)
16:46:45:WU01:FS01:0x17:Completed 2300000 out of 5000000 steps (46%)
Image
PS3EdOlkkola
Posts: 184
Joined: Tue Aug 26, 2014 9:48 pm
Hardware configuration: 10 SMP folding slots on Intel Phi "Knights Landing" system, configured as 24 CPUs/slot
9 AMD GPU folding slots
31 Nvidia GPU folding slots
50 total folding slots
Average PPD/slot = 459,500
Location: Dallas, TX

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by PS3EdOlkkola »

Using Catalyst 14.9 and 7970's running at stock frequencies (1GHz) on a PCIe-16 3.0 interface yields a consistent 7:40 frame time @ 137,000 PPD for p13000 in Win7 x64 with a dedicated core for the GPU. Since the 7950 is slightly less powerful, 70K to 120K PPD sounds about right.
Image
Hardware config viewtopic.php?f=66&t=17997&p=277235#p277235
bmwman91
Posts: 20
Joined: Tue Feb 17, 2009 1:33 am

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by bmwman91 »

Thanks. I just did a complete AMD driver sweep-out and am going to try going with Catalyst 14.12 to see how things go. Some people seem to see a ~10% loss in PPD versus 14.9, but I would rather have 90% of 100K PPD than the 26K PDD I was getting!

EDIT: I have Catalyst 14.12 installed and the F@H client back up and going. I did not delete the WU's (13000 and 13001) that were in progress, and they are picking back up from 46% completion. So, I will check back in after a few hours to see where the TPF is at.

One other question...the motherboards have basic display adapters built in to them. Would I get any better GPU folding performance if I drove the monitor from that, rather than from the HD 7950's?
Image
bmwman91
Posts: 20
Joined: Tue Feb 17, 2009 1:33 am

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by bmwman91 »

PS3EdOlkkola wrote:Using Catalyst 14.9 and 7970's running at stock frequencies (1GHz) on a PCIe-16 3.0 interface yields a consistent 7:40 frame time @ 137,000 PPD for p13000 in Win7 x64 with a dedicated core for the GPU. Since the 7950 is slightly less powerful, 70K to 120K PPD sounds about right.
I am using the default Folding client setup...should I try to assign one CPU core to the GPU (if it isn't doing this by default)? How would I go about doing that?

Thanks.
Image
davidcoton
Posts: 1102
Joined: Wed Nov 05, 2008 3:19 pm
Location: Cambridge, UK

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by davidcoton »

It should happen by default. In Advanced Control, check your total number of CPU cores (near top of log file) and then the number assigned to the CPU slot (a bit further down, when the CPU slot starts ... look for "Mapping"). If necessary, adjust in the CPU slot config tab CPU slot config, where the default "-1" means "leave one for each GPU slot".

Example, slightly abbreviated:
*********************** Log Started 2014-12-19T18:46:30Z ***********************
18:46:30:************************* Folding@home Client *************************
18:46:30: Website: http://folding.stanford.edu/
18:46:30: Copyright: (c) 2009-2014 Stanford University
18:46:30: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
18:46:30: Args: --child --lifeline 1245 /etc/fahclient/config.xml --run-as
18:46:30: fahclient --pid-file=/var/run/fahclient.pid --daemon
18:46:30: Config: /etc/fahclient/config.xml
18:46:30:******************************** Build ********************************

...

18:46:30:******************************* System ********************************
18:46:30: CPU: AMD Phenom(tm) II X6 1055T Processor
18:46:30: CPU ID: AuthenticAMD Family 16 Model 10 Stepping 0
18:46:30: CPUs: 6
18:46:30: Memory: 3.86GiB
18:46:30:Free Memory: 3.48GiB

...

05:10:27:WU00:FS00:0xa3:Mapping NT from 5 to 5
Image
PS3EdOlkkola
Posts: 184
Joined: Tue Aug 26, 2014 9:48 pm
Hardware configuration: 10 SMP folding slots on Intel Phi "Knights Landing" system, configured as 24 CPUs/slot
9 AMD GPU folding slots
31 Nvidia GPU folding slots
50 total folding slots
Average PPD/slot = 459,500
Location: Dallas, TX

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by PS3EdOlkkola »

One other question...the motherboards have basic display adapters built in to them. Would I get any better GPU folding performance if I drove the monitor from that, rather than from the HD 7950's?
If you routinely use your computer for tasks other than FAH, using the motherboard VGA/DVI display connection instead of the direct connection to the 7950 GPU will yield better folding performance for the GPU, particularly if you're gaming or use other GPU intensive tasks other than FAH. Using the motherboard display interface will yield you an even better benefit, though, in reduced screen lag compared to the screen refresh performance when running FAH on the GPU attached to the display. On the other hand, if your rig is used primarily for FAH and occasional light web browsing or applications not using the GPU all that much, then there will be very little improved folding performance when running off the motherboard display adapter.
Image
Hardware config viewtopic.php?f=66&t=17997&p=277235#p277235
bmwman91
Posts: 20
Joined: Tue Feb 17, 2009 1:33 am

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by bmwman91 »

davidcoton wrote:It should happen by default. In Advanced Control, check your total number of CPU cores (near top of log file) and then the number assigned to the CPU slot (a bit further down, when the CPU slot starts ... look for "Mapping"). If necessary, adjust in the CPU slot config tab CPU slot config, where the default "-1" means "leave one for each GPU slot".

Example, slightly abbreviated:
*********************** Log Started 2014-12-19T18:46:30Z ***********************
18:46:30:************************* Folding@home Client *************************
18:46:30: Website: http://folding.stanford.edu/
18:46:30: Copyright: (c) 2009-2014 Stanford University
18:46:30: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
18:46:30: Args: --child --lifeline 1245 /etc/fahclient/config.xml --run-as
18:46:30: fahclient --pid-file=/var/run/fahclient.pid --daemon
18:46:30: Config: /etc/fahclient/config.xml
18:46:30:******************************** Build ********************************

...

18:46:30:******************************* System ********************************
18:46:30: CPU: AMD Phenom(tm) II X6 1055T Processor
18:46:30: CPU ID: AuthenticAMD Family 16 Model 10 Stepping 0
18:46:30: CPUs: 6
18:46:30: Memory: 3.86GiB
18:46:30:Free Memory: 3.48GiB

...

05:10:27:WU00:FS00:0xa3:Mapping NT from 5 to 5
Alrighty, I have 16 cores and 15 to 15 are mapped to the CPU slot. So, it looks like the GPU has its dedicated core, as it should.
PS3EdOlkkola wrote:
One other question...the motherboards have basic display adapters built in to them. Would I get any better GPU folding performance if I drove the monitor from that, rather than from the HD 7950's?
If you routinely use your computer for tasks other than FAH, using the motherboard VGA/DVI display connection instead of the direct connection to the 7950 GPU will yield better folding performance for the GPU, particularly if you're gaming or use other GPU intensive tasks other than FAH. Using the motherboard display interface will yield you an even better benefit, though, in reduced screen lag compared to the screen refresh performance when running FAH on the GPU attached to the display. On the other hand, if your rig is used primarily for FAH and occasional light web browsing or applications not using the GPU all that much, then there will be very little improved folding performance when running off the motherboard display adapter.
The machine is going to sit and do almost nothing BUT fold, aside from some visits to the Folding Forum for assistance!

So it sounds like there is something messed up on my rigs, although it seems a little odd that they both perform exactly this way. I have another machine, a Lenovo W520 with 3720QM and Quadro 2000M GPU, and it was seeing 1:20:00 or so TPF on project 13000. Maybe it is the 13000 series WUs that just kill the GPU performance? I guess I'll give the AMD GPUs another day or so to finish these and see what they get assigned next. Anyone have any helpful tips about why I am seeing an 60-80% loss in GPU folding performance on a clean installation?
Image
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by bruce »

PS3EdOlkkola wrote:
One other question...the motherboards have basic display adapters built in to them. Would I get any better GPU folding performance if I drove the monitor from that, rather than from the HD 7950's?
If you routinely use your computer for tasks other than FAH, using the motherboard VGA/DVI display connection instead of the direct connection to the 7950 GPU will yield better folding performance for the GPU, particularly if you're gaming or use other GPU intensive tasks other than FAH. Using the motherboard display interface will yield you an even better benefit, though, in reduced screen lag compared to the screen refresh performance when running FAH on the GPU attached to the display. On the other hand, if your rig is used primarily for FAH and occasional light web browsing or applications not using the GPU all that much, then there will be very little improved folding performance when running off the motherboard display adapter.
Many motherboards disable the built-in VGA device when they find another GPU. Others do not. Hopefully you're one of the lucky ones which permit both devices to operate simultaneously.
PS3EdOlkkola
Posts: 184
Joined: Tue Aug 26, 2014 9:48 pm
Hardware configuration: 10 SMP folding slots on Intel Phi "Knights Landing" system, configured as 24 CPUs/slot
9 AMD GPU folding slots
31 Nvidia GPU folding slots
50 total folding slots
Average PPD/slot = 459,500
Location: Dallas, TX

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by PS3EdOlkkola »

a Lenovo W520 with 3720QM and Quadro 2000M GPU
A Quadro 2000M is based on the couple-generations-old Fermi architecture. Here is a reference: http://www.notebookcheck.net/NVIDIA-Qua ... 316.0.html

Since it has 192 CUDA cores/shaders, I suspect it will be fast enough to complete a p13000 work unit in time, assuming it folds 24 x 7 in your Lenovo notebook workstation and the heat/noise don't become too unbearable. GPU folding in a notebook is certainly possible, but it does put a lot of stress on the hardware due to limitations of the physical package for cooling. You'll want to make sure you use compressed air to blow out all dust/gunk that might be blocking the cooling ports and heat sinks and keep them clean and maintained over time.
Many motherboards disable the built-in VGA device when they find another GPU. Others do not. Hopefully you're one of the lucky ones which permit both devices to operate simultaneously.
Excellent clarification Bruce.

If your system supports dual-VGA use, it can typically be found in the BIOS. At system startup, enter the BIOS to determine if your motherboard supports the simultaneous use of the motherboard display and an add-in GPU. You should also check to see if there is a BIOS update to your motherboard that provides this functionality as a feature.
Image
Hardware config viewtopic.php?f=66&t=17997&p=277235#p277235
bmwman91
Posts: 20
Joined: Tue Feb 17, 2009 1:33 am

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by bmwman91 »

Good info, thanks. There is a BIOS option to enable the onboard VGA and I would assume that the GPU should also work as long as the PCIe slot it is in is enabled. I'll give it a try, although it sounds like folding performance is not affected if the GPU is used for video output but unused by anything else. These are more or less "dedicated" folding workstations.

The W530 has been folding for a couple of weeks now. It is used for a lot of things on a daily basis, but it has been cranking out ~25K PPD pretty steadily. As far as I can remember, this is the first P13000 WU it has encountered, and it is definitely one of the tougher ones for the GPU.

For the 2 machines with 7950's performing abnormally slowly, I am thinking of maybe installing Linux Mint and trying F@H again in there (once the current WUs finish). It sounds like the Linux Catalyst drivers are OK now...or is AMD GPU folding in Linux still going to way underperform versus Windows?

Any pointers as to why I am having such poor GPU folding performance in Windows, on a WU that others are seeing 4x the performance on with the same/similar GPUs? It is sort of bumming me out, especially since these cards are fairly power hungry!
Image
foldy
Posts: 2061
Joined: Sat Dec 01, 2012 3:43 pm
Hardware configuration: Folding@Home Client 7.6.13 (1 GPU slots)
Windows 7 64bit
Intel Core i5 2500k@4Ghz
Nvidia gtx 1080ti driver 441

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by foldy »

I also have a HD7950 and for P13000 work units it takes some 10min per frame.
I dont know if it helps but beside Amd graphics driver are there any drivers missing
in Windows device manager? Such as chipset drivers, especially PCI-Express?
PS3EdOlkkola
Posts: 184
Joined: Tue Aug 26, 2014 9:48 pm
Hardware configuration: 10 SMP folding slots on Intel Phi "Knights Landing" system, configured as 24 CPUs/slot
9 AMD GPU folding slots
31 Nvidia GPU folding slots
50 total folding slots
Average PPD/slot = 459,500
Location: Dallas, TX

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by PS3EdOlkkola »

Any pointers as to why I am having such poor GPU folding performance in Windows, on a WU that others are seeing 4x the performance on with the same/similar GPUs? It is sort of bumming me out, especially since these cards are fairly power hungry!
Here are some test processes, procedures and tools I use to diagnose GPU issues. They may help you isolate what's going on with your 7950's:

1. Run the display driver uninstaller (Catalyst Control) and remove the driver from the system.
2. Reboot - very critical!
3. Run the installer for Catalyst 14.9
4. Reboot

That process should produce a nice clean installation. You can then start FAH and check performance. Don't shortcut the process by skipping the reboot on Step 2.

If it performance is still an issue:

1. Pause FAH, exit FAH
2. Download and install GPU-Z here: http://www.techpowerup.com/gpuz/
3. Reboot
4. Start FAH, unpause FAH
5. Run GPU-Z
6. Select the "Sensors" tab and check the first entry, which is "GPU Clock"
7. It may take 2 to 4 minutes for the GPU to get fully engaged, but once it does, you should see the GPU Clock increase from appx 300 MHz to something close to the rated clock rate of your 7950, which is probably between 900MHZ and 1GHz.
8. If you have more than 1 GPU in your system, you can monitor each in the drop-down window at the bottom of GPU-Z.
9. Monitor GPU performance, noting that the GPU Clock should stay right at the 900MHz+ rate. For a p13000 WU, monitor for at least one full frame.
10. Take note of the GPU Temperature. If the GPU Temperature starts climbing above 90 deg C, then the GPU may start throttling performance to save itself from overheating. You should see that in fluctuating GPU Clock (cycling up and down from 900MHz+). That will confirm that you have a heat evacuation problem in your case and will need additional fans, or re-engineer the fans you have to move more air through the case. Overheating not only dramatically shortens component life, but can be a source of substantial degradation in folding performance. I try to keep my GPUs under 75 deg C on air at stock speeds (w/ambient 74 deg F)
11. Assuming GPU temperatures stay under 90 deg C, then continue to watch the GPU Clock to see if it drops back down to something close to or under 300 MHz. If it does, that could mean a couple of things: You likely have a power supply that is unable to push enough amperage through the 12V rail and need to upgrade your power supply to support the power draw of the GPUs, and/or you have a defective GPU. Shut everything down and begin running one GPU, to confirm temperature and clock rates, then the other, but both separately (physically remove one of the 7950's from your case, test the one that remains, remove that one, install the second one, test it). You can then isolate either a power supply issue or a defective GPU.

If GPU-Z shows that temperatures are the likely culprit in reduced performance, you'll have to either increase case ventilation (as above) or slow down GPU clock rate. If you choose to slow the GPU clock rate, then use this procedure:

1. Pause FAH, exit FAH
2. Download and install MSI Afterburner http://event.msi.com/vga/afterburner/download.htm
3. Reboot
4. Start Afterburner
5. Start FAH, unpause FAH
6. Drop the GPU Clock rate by about 100 MHz, and continue slowing down the GPU clock rate so the GPU temperature is not going above 90 deg C.

The other possible issue is not having enough CPU time allocated to the GPU to allow the GPU to stay fully engaged. It was described earlier to set the number of CPU you use to fold with equal to total CPUs less the number of GPUs in your system. With an Opteron 6272, a 16 core CPU, be conservative and set the number of CPUs to 12, leaving 4 to support the GPUs. That setting also allows some headroom for system overhead. I like to be more precise and use ProcessLasso https://bitsum.com/processlasso/to specifically allocate CPUs to running processes and move all processes off the CPUs supporting the GPUs to avoid any contention. Latency is a big performance killer, so keeping the cores supporting the GPUs as "quiet" as possible has a positive impact on PPD.

One last note: If you don't see the readings you expect to see in either GPU-Z and/or Afterburner, then the GPU driver installation is seriously hosed. That has happened to me on a couple of occasions when my folding rigs blue-screened for no apparent reason. The systems would restart and the GPUs would fold, but performance was awful, and GPU-Z and Afterburner could not get any performance data out of the GPU (specifically for AMD GPUs). The only fix was a full re-installation of Windows. Hopefully that's not your issue.
Image
Hardware config viewtopic.php?f=66&t=17997&p=277235#p277235
bmwman91
Posts: 20
Joined: Tue Feb 17, 2009 1:33 am

Re: Low / Inconsistent PPD with Radeon 7950 & Core 17

Post by bmwman91 »

foldy wrote:I also have a HD7950 and for P13000 work units it takes some 10min per frame.
I dont know if it helps but beside Amd graphics driver are there any drivers missing
in Windows device manager? Such as chipset drivers, especially PCI-Express?
So, I have not been able to find chipset drivers. The machines have Supermicro H8DG6 motherboards (Dual AMD chipset SR5690/SP5100). Supermiro doesn't have anything that looks like chipset drivers, and I cannot find anything specific for this chipset on AMD's site. I tried downloading the 14.12 Catalyst chipset drivers, and it only seems to show USB ports and other peripheral stuff as the hardware it will install drivers for. The SATA/AHCI/chipset drivers in Device Manager seem to be the generic Microsoft ones.

So, maybe you are on to something if the PCIe slots won't properly work with devices without specific chipset drivers. Looks like it is time for more internet searching!
Image
Post Reply