Low GPU usage with GTX 970 Windows7

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

Moderators: Site Moderators, FAHC Science Team

Post Reply
Kebast
Posts: 386
Joined: Thu Aug 06, 2015 5:21 pm

Low GPU usage with GTX 970 Windows7

Post by Kebast »

I got a new GPU installed recently, MSI GTX970 100MillionEdition. PPD seems ok, generally around 240k, but the GPU usage only averages around 90%. I've got some logs and screenshots below for various WUs. Doesn't matter if I overclock or not, GPU usage never gets close to 100%. I'm running a 750ti in Ubuntu and that one stays capped at 99%, so curious why this machine is not the same.

Config:

Code: Select all

*********************** Log Started 2015-11-11T08:24:51Z ***********************
08:24:51:************************* Folding@home Client *************************
08:24:51:      Website: http://folding.stanford.edu/
08:24:51:    Copyright: (c) 2009-2014 Stanford University
08:24:51:       Author: Joseph Coffland <joseph@cauldrondevelopment.com>
08:24:51:         Args: 
08:24:51:       Config: C:/ProgramData/FAHClient/config.xml
08:24:51:******************************** Build ********************************
08:24:51:      Version: 7.4.4
08:24:51:         Date: Mar 4 2014
08:24:51:         Time: 20:26:54
08:24:51:      SVN Rev: 4130
08:24:51:       Branch: fah/trunk/client
08:24:51:     Compiler: Intel(R) C++ MSVC 1500 mode 1200
08:24:51:      Options: /TP /nologo /EHa /Qdiag-disable:4297,4103,1786,279 /Ox -arch:SSE
08:24:51:               /QaxSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 /Qopenmp /Qrestrict /MT /Qmkl
08:24:51:     Platform: win32 XP
08:24:51:         Bits: 32
08:24:51:         Mode: Release
08:24:51:******************************* System ********************************
08:24:51:          CPU: AMD FX(tm)-6300 Six-Core Processor
08:24:51:       CPU ID: AuthenticAMD Family 21 Model 2 Stepping 0
08:24:51:         CPUs: 6
08:24:51:       Memory: 7.97GiB
08:24:51:  Free Memory: 6.81GiB
08:24:51:      Threads: WINDOWS_THREADS
08:24:51:   OS Version: 6.1
08:24:51:  Has Battery: false
08:24:51:   On Battery: false
08:24:51:   UTC Offset: -5
08:24:51:          PID: 3080
08:24:51:          CWD: C:/ProgramData/FAHClient
08:24:51:           OS: Windows 7 Home Premium
08:24:51:      OS Arch: AMD64
08:24:51:         GPUs: 1
08:24:51:        GPU 0: NVIDIA:5 GM204 [GeForce GTX 970]
08:24:51:         CUDA: 5.2
08:24:51:  CUDA Driver: 7050
08:24:51:Win32 Service: false
08:24:51:***********************************************************************
08:24:51:<config>
08:24:51:  <!-- HTTP Server -->
08:24:51:  <allow v='127.0.0.1,192.168.66.95-192.168.66.151'/>
08:24:51:
08:24:51:  <!-- Network -->
08:24:51:  <proxy v=':8080'/>
08:24:51:
08:24:51:  <!-- Remote Command Server -->
08:24:51:  <command-allow-no-pass v='127.0.0.1,192.168.66.95-192.168.66.151'/>
08:24:51:
08:24:51:  <!-- Slot Control -->
08:24:51:  <power v='full'/>
08:24:51:
08:24:51:  <!-- User Information -->
08:24:51:  <passkey v='********************************'/>
08:24:51:  <team v='229226'/>
08:24:51:  <user v='Kebast'/>
08:24:51:
08:24:51:  <!-- Work Unit Control -->
08:24:51:  <next-unit-percentage v='100'/>
08:24:51:
08:24:51:  <!-- Folding Slots -->
08:24:51:  <slot id='1' type='GPU'/>
08:24:51:</config>
For project:10484 run:0 clone:156 gen:8:
Image
http://pasteboard.co/23SnXpxC.png

Code: Select all

21:25:44:WU00:FS01:Received Unit: id:00 state:DOWNLOAD error:NO_ERROR project:10484 run:0 clone:156 gen:8 core:0x18 unit:0x0000000c538b3dbb54ac2dd9c9b2eb01
21:26:05:WU00:FS01:Starting
21:26:05:WU00:FS01:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:/ProgramData/FAHClient/cores/web.stanford.edu/~pande/Win32/AMD64/NVIDIA/Fermi/Core_18.fah/FahCore_18.exe -dir 00 -suffix 01 -version 704 -lifeline 2720 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
21:26:05:WU00:FS01:Started FahCore on PID 4628
21:26:05:WU00:FS01:Core PID:492
21:26:05:WU00:FS01:FahCore 0x18 started
21:26:06:WU00:FS01:0x18:*********************** Log Started 2015-11-09T21:26:05Z ***********************
21:26:06:WU00:FS01:0x18:Project: 10484 (Run 0, Clone 156, Gen 8)
21:26:06:WU00:FS01:0x18:Unit: 0x0000000c538b3dbb54ac2dd9c9b2eb01
21:26:06:WU00:FS01:0x18:CPU: 0x00000000000000000000000000000000
21:26:06:WU00:FS01:0x18:Machine: 1
21:26:06:WU00:FS01:0x18:Reading tar file state.xml
21:26:06:WU00:FS01:0x18:Reading tar file system.xml
21:26:07:WU00:FS01:0x18:Reading tar file integrator.xml
21:26:07:WU00:FS01:0x18:Reading tar file core.xml
21:26:07:WU00:FS01:0x18:Digital signatures verified
21:26:07:WU00:FS01:0x18:Folding@home GPU core18
21:26:07:WU00:FS01:0x18:Version 0.0.4
For project: 10478 (Run 0, Clone 146, Gen 195)

Image
http://pasteboard.co/23SeGMOw.png

Code: Select all

08:24:51:Enabled folding slot 01: READY gpu:0:GM204 [GeForce GTX 970]
08:24:51:WU00:FS01:Starting
08:24:51:WU00:FS01:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:/ProgramData/FAHClient/cores/web.stanford.edu/~pande/Win32/AMD64/NVIDIA/Fermi/Core_18.fah/FahCore_18.exe -dir 00 -suffix 01 -version 704 -lifeline 3080 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
08:24:51:WU00:FS01:Started FahCore on PID 4080
08:24:51:WU00:FS01:Core PID:2904
08:24:51:WU00:FS01:FahCore 0x18 started
08:24:52:WU00:FS01:0x18:*********************** Log Started 2015-11-11T08:24:51Z ***********************
08:24:52:WU00:FS01:0x18:Project: 10478 (Run 0, Clone 146, Gen 195)
08:24:52:WU00:FS01:0x18:Unit: 0x000000f9538b3dba548f672491d81acc
08:24:52:WU00:FS01:0x18:CPU: 0x00000000000000000000000000000000
08:24:52:WU00:FS01:0x18:Machine: 1
08:24:52:WU00:FS01:0x18:Digital signatures verified
08:24:52:WU00:FS01:0x18:Folding@home GPU core18
08:24:52:WU00:FS01:0x18:Version 0.0.4
08:24:52:WU00:FS01:0x18:  Found a checkpoint file
08:25:14:WU00:FS01:0x18:Completed 500000 out of 5000000 steps (10%)
For Project: 9413 (Run 45, Clone 6, Gen 53)

Image
http://pasteboard.co/23SxeSVE.png

Code: Select all

13:56:50:WU01:FS01:Starting
13:56:50:WU01:FS01:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:/ProgramData/FAHClient/cores/web.stanford.edu/~pande/Win32/AMD64/NVIDIA/Fermi/Core_18.fah/FahCore_18.exe -dir 01 -suffix 01 -version 704 -lifeline 3080 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
13:56:50:WU01:FS01:Started FahCore on PID 2004
13:56:50:WU01:FS01:Core PID:1156
13:56:50:WU01:FS01:FahCore 0x18 started
13:56:50:WU01:FS01:0x18:*********************** Log Started 2015-11-11T13:56:50Z ***********************
13:56:50:WU01:FS01:0x18:Project: 9413 (Run 45, Clone 6, Gen 53)
13:56:50:WU01:FS01:0x18:Unit: 0x00000047ab40413a55410d78b736d473
13:56:50:WU01:FS01:0x18:CPU: 0x00000000000000000000000000000000
13:56:50:WU01:FS01:0x18:Machine: 1
13:56:50:WU01:FS01:0x18:Reading tar file state.xml
13:56:50:WU01:FS01:0x18:Reading tar file system.xml
13:56:50:WU01:FS01:0x18:Reading tar file integrator.xml
13:56:50:WU01:FS01:0x18:Reading tar file core.xml
13:56:50:WU01:FS01:0x18:Digital signatures verified
13:56:50:WU01:FS01:0x18:Folding@home GPU core18
13:56:50:WU01:FS01:0x18:Version 0.0.4
Also, what am I doing wrong so that the image isn't showing directly in the post?
Image
Ryzen 5900x 12T - RTX 4070 TI
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Low GPU usage with GTX 970 Windows7

Post by bruce »

Any GPU task consists of a collection of different steps. The major pieces are (A) Transferring data from RAM to VRAM, (B) processing the data that's in VRAM, (C) Returning results to RAM. (D) Checking the data for consistency, and (E) Writing the data to a checkpoint file ... and repeat until done.

E: Generally takes a second or so, depending on the speed of you disk, and that could be what you're seeing in the image.
D: May take some time, too, depending mostly on the speed of your CPU.
A & C: Depends mostly on the speed of you PCIe bus.
B: is more or less what you see in the GPU-busy number.

Well written code can overlap A&C with B provided there's room in VRAM and providing the speeds allow that to happen. If you speed up a 750 TI to be a 970 {AND NOTHIN ELSE CHANGES} part B gets faster but parts A&C probably don't change so the processing may need to wait for new data to be delivered ... i.e.- the processing is less able to overlap with the I/O. If that's what is going on, putting the GPU in a faster slot will help.
Kebast
Posts: 386
Joined: Thu Aug 06, 2015 5:21 pm

Re: Low GPU usage with GTX 970 Windows7

Post by Kebast »

Here's the motherboard: Gigabyte GA-990FXA-UD3
http://www.gigabyte.com/products/produc ... id=3894#ov
It has PCI-e 2.0 x16 slots. The only thing I can think of is that I had to install the card in the #2 x16 slot instead of the primary. The bus utilization is usually only 40% though. 970 card is LONG and my HDD bracket is in the way of the primary slot. OS drive is a SSD, quite fast so I don't think that would be the issue.

Of note, I had the CPU folding using 5 of the 6 threads. Removing that slot didn't affect the GPU utilization. Google didn't find much on my first attempt, but of this usage is normal I'll leave it be.
Thanks for the reply.
Image
Ryzen 5900x 12T - RTX 4070 TI
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Low GPU usage with GTX 970 Windows7

Post by bruce »

It's not valid to assume the system cannot overlap I/O with calculations, but let's do it anyway. Using your 40% number of potentially the time spent waiting on the bus -- the GPU core might be processing data as little as 60% of the time. It would never be that bad, but the fact is that as GPUs get faster without a corresponding increase in the data supply rate, the calculations finish sooner so they wait a larger and larger percent of the total time.
Kebast
Posts: 386
Joined: Thu Aug 06, 2015 5:21 pm

Re: Low GPU usage with GTX 970 Windows7

Post by Kebast »

I had the first core 21 this morning on that machine. GPU usage was bouncing around from 95 to 99% on that one. What you said above might be exactly the case.
Image
Ryzen 5900x 12T - RTX 4070 TI
Kebast
Posts: 386
Joined: Thu Aug 06, 2015 5:21 pm

Re: Low GPU usage with GTX 970 Windows7

Post by Kebast »

Another core 21 this afternoon, 9206(0,2123,1). GPU usage is again 96-99% for this one. I didn't notice before, but oddly enough, the BUS usage is slightly higher, at around 48%, than on the core 18 WUs. Anyway, I'm getting ~250k ppd on average, think it's good.
Image
Ryzen 5900x 12T - RTX 4070 TI
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Low GPU usage with GTX 970 Windows7

Post by bruce »

The new proteins that are coming on-line are expected to have much hidher atom-counts that what has typically been assigned. Folding@home is gradually finishing studies of small proteins and graduating to larger and larger proteins. That ups the ante in terms of the amount of data that must be moved and makes WUs use more wall-clock-time per unit simulated-time. -- but we will be working on more challenging scientific research.
Post Reply