Driver 337.88 and 600 (and earlier) series GPUs

[b-morgan]

There isn't a topic on version 337.88 of the driver and it folding performance (or lack thereof) on 600 and older series GPUs. Is this driver finally a replacement for the recommended 327.23 driver for these older GPUs? Has NVidia ever acknowledged that there is a problem? Have they ever said that it will be fixed?

[PantherX]

The slowdown of folding on non GK-110 GPUs might potentially be fixed in FahCore_17 version 0.0.55 which is being beta tested (viewtopic.php?f=66&t=26190). I haven't heard any news regarding the Maxwell issue which is caused by a bug in OpenCL. Nvidia is aware of it.

[Jimboc]

Hi b-morgan,

In addition to the link provided by PantherX (above) the following EVGA thread discusses the results when using this beta version of Core 17:

http://forums.evga.com/tm.aspx?m=2149461

I hope this helps. Thank you.

[7im]

It's well discussed here also.

There are at least six threads in the NVidia Drivers section talking about performance with different versions. For example:

viewtopic.php?f=80&t=25850

They all end the same way. -> Use the 327.xx drivers on all hardware except the GK-110s until NV fixes their drivers.

There is a kluge patch in the v55 beta fahcore_17 to workaround this driver issue, but the best speed will come when NV puts the fix in the drivers.

[b-morgan]

They all end the same way. -> Use the 327.xx drivers on all hardware except the GK-110s until NV fixes their drivers.

There is a kluge patch in the v55 beta fahcore_17 to workaround this driver issue, but the best speed will come when NV puts the fix in the drivers.

So my questions remain unanswered...

1) Has NV acknowledged that there is a problem?

2) If so, has NV said they will fix the problem?

If the answers to both of these questions is no, then "kluge" or not, the V55 beta fahcore_17 is the best solution offered so far. How much longer will the beta last (i.e. when will this version be released)?

I have both a GTX 660 (GK106) and a GTX 660 Ti (GK104) and I'd be happy to test the beta, but the instructions aren't "simple" enough for me to follow.

[7im]

Good questions. Wrong forum. Please ask NV in their forum.

[PantherX]

...How much longer will the beta last (i.e. when will this version be released)?...

That can only be answered by a PG Member. They decided when FahCore_17 is stable enough to be released to the public.

[codysluder]

Proteneer (the Stanford developer of FahCore_17) mentioned that NV isn't handling a particular OpenCL call correctly. I presume he reported it to NV. We have to wait for them to release new drivers that contain that fix.

[folding_hoomer]

b-morgan - nVidia has been informed by folders since this issue occurs - with driver-version 331.
But - nothing has happend (until now) to patch this issue.
It seems that there are much more important things to do for them - eg. to force Cuda or to get better FPS in games . . .

[Ghetto_Child]

Well I'm not sure where my issue lands. I'm using old tech vastly outbalanced, compare my graphics adaptor to the rest of the system. I've been switching back and forth between a GeForce GTX 285 and a GeForce GTX 295, always using GeForce 337.88 drivers since I first put either card into this PC. I'm trying to make sure the 295 is working properly. First time last week I tried the GTX 295 in F@H client 7.3.6 and each GPU it has received a work unit. Then I swapped to the GTX 285 and also got a single work unit for its single GPU. By the next day I swapped back to the GTX 295 and was not getting any GPU WUs for days. I tried upgrading to client 7.4.4 to see if that would fix not receiving GPU WUs, no change. I just started receiving GPU WUs yesterday (YAY I can connect with GPU assignment servers).

So here's the issue:, the GPU slot(s) will not run unless set to medium or higher. When set to light the GPU slot(s) status is "waiting for idle". When I do set a single GPU slot to medium or higher, it completes its WUs in about an hour but the whole system because very sluggish to respond and the CPU load in Task Manager stays between 75%-90%+. So basically running GPU processing seems super efficient for WUs but my system is otherwise unuseable during GPU processing. I also noticed that I never get any points for the GPU WUs. There are project numbers and details attached to the GPU WUs but the Base Credit, Estimated Credit, Estimated PPD all say unknown. Estimated TPF usually shows 30 seconds - 3 minutes. How can I reduce the strain of GPU processing and keep the system useable without issue as it is when processing CPU WUs?

Code: Select all

*********************** Log Started 2014-06-26T00:34:45Z ***********************
00:34:45:************************* Folding@home Client *************************
00:34:45:      Website: http://folding.stanford.edu/
00:34:45:    Copyright: (c) 2009-2014 Stanford University
00:34:45:       Author: Joseph Coffland <joseph@cauldrondevelopment.com>
00:34:45:         Args: 
00:34:45:       Config: C:/ProgramData/FAHClient/config.xml
00:34:45:******************************** Build ********************************
00:34:45:      Version: 7.4.4
00:34:45:         Date: Mar 4 2014
00:34:45:         Time: 20:26:54
00:34:45:      SVN Rev: 4130
00:34:45:       Branch: fah/trunk/client
00:34:45:     Compiler: Intel(R) C++ MSVC 1500 mode 1200
00:34:45:      Options: /TP /nologo /EHa /Qdiag-disable:4297,4103,1786,279 /Ox -arch:SSE
00:34:45:               /QaxSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 /Qopenmp /Qrestrict /MT /Qmkl
00:34:45:     Platform: win32 XP
00:34:45:         Bits: 32
00:34:45:         Mode: Release
00:34:45:******************************* System ********************************
00:34:45:          CPU: Intel(R) Pentium(R) 4 CPU 2.66GHz
00:34:45:       CPU ID: GenuineIntel Family 15 Model 4 Stepping 9
00:34:45:         CPUs: 1
00:34:45:       Memory: 2.00GiB
00:34:45:  Free Memory: 897.61MiB
00:34:45:      Threads: WINDOWS_THREADS
00:34:45:   OS Version: 6.1
00:34:45:  Has Battery: false
00:34:45:   On Battery: false
00:34:45:   UTC Offset: -4
00:34:45:          PID: 7740
00:34:45:          CWD: C:/ProgramData/FAHClient
00:34:45:           OS: Windows 7 Ultimate
00:34:45:      OS Arch: AMD64
00:34:45:         GPUs: 2
00:34:45:        GPU 0: NVIDIA:1 GT200b [GeForce GTX 295]
00:34:45:        GPU 1: NVIDIA:1 GT200b [GeForce GTX 295]
00:34:45:         CUDA: 1.3
00:34:45:  CUDA Driver: 6000
00:34:45:Win32 Service: false
00:34:45:***********************************************************************
00:34:45:<config>
00:34:45:  <service-description v='Folding@home Client'/>
00:34:45:  <service-restart v='true'/>
00:34:45:  <service-restart-delay v='5000'/>
00:34:45:
00:34:45:  <!-- Client Control -->
00:34:45:  <client-threads v='6'/>
00:34:45:  <cycle-rate v='4'/>
00:34:45:  <cycles v='-1'/>
00:34:45:  <data-directory v='.'/>
00:34:45:  <disable-sleep-when-active v='true'/>
00:34:45:  <exec-directory v='C:\Program Files (x86)\FAHClient'/>
00:34:45:  <exit-when-done v='false'/>
00:34:45:  <fold-anon v='false'/>
00:34:45:  <open-web-control v='false'/>
00:34:45:
00:34:45:  <!-- Configuration -->
00:34:45:  <config-rotate v='true'/>
00:34:45:  <config-rotate-dir v='configs'/>
00:34:45:  <config-rotate-max v='16'/>
00:34:45:
00:34:45:  <!-- Debugging -->
00:34:45:  <assignment-servers>
00:34:45:    assign3.stanford.edu:8080 assign4.stanford.edu:80
00:34:45:  </assignment-servers>
00:34:45:  <auth-as v='true'/>
00:34:45:  <capture-directory v='capture'/>
00:34:45:  <capture-on-error v='false'/>
00:34:45:  <capture-packets v='false'/>
00:34:45:  <capture-requests v='false'/>
00:34:45:  <capture-responses v='false'/>
00:34:45:  <capture-sockets v='false'/>
00:34:45:  <core-exec v='FahCore_$type'/>
00:34:45:  <core-wrapper-exec v='FAHCoreWrapper'/>
00:34:45:  <debug-sockets v='false'/>
00:34:45:  <exception-locations v='true'/>
00:34:45:  <gpu-assignment-servers>
00:34:45:    assign-GPU.stanford.edu:80 assign-GPU2.stanford.edu:80
00:34:45:  </gpu-assignment-servers>
00:34:45:  <stack-traces v='false'/>
00:34:45:
00:34:45:  <!-- Error Handling -->
00:34:45:  <max-slot-errors v='10'/>
00:34:45:  <max-unit-errors v='5'/>
00:34:45:
00:34:45:  <!-- Folding Core -->
00:34:45:  <checkpoint v='30'/>
00:34:45:  <core-dir v='cores'/>
00:34:45:  <core-priority v='idle'/>
00:34:45:  <cpu-affinity v='false'/>
00:34:45:  <cpu-usage v='100'/>
00:34:45:  <gpu-usage v='100'/>
00:34:45:  <no-assembly v='false'/>
00:34:45:
00:34:45:  <!-- Folding Slot Configuration -->
00:34:45:  <cause v='ANY'/>
00:34:45:  <client-subtype v='STDCLI'/>
00:34:45:  <client-type v='normal'/>
00:34:45:  <cpu-species v='UNKNOWN'/>
00:34:45:  <cpu-type v='AMD64'/>
00:34:45:  <cpus v='-1'/>
00:34:45:  <gpu v='true'/>
00:34:45:  <max-packet-size v='normal'/>
00:34:45:  <os-species v='UNKNOWN'/>
00:34:45:  <os-type v='WIN32'/>
00:34:45:  <project-key v='0'/>
00:34:45:  <smp v='true'/>
00:34:45:
00:34:45:  <!-- GUI -->
00:34:45:  <gui-enabled v='true'/>
00:34:45:
00:34:45:  <!-- HTTP Server -->
00:34:45:  <allow v='*.*.*.*'/>
00:34:45:  <connection-timeout v='60'/>
00:34:45:  <deny v='0/0'/>
00:34:45:  <http-addresses v='*:*'/>
00:34:45:  <https-addresses v=''/>
00:34:45:  <max-connect-time v='900'/>
00:34:45:  <max-connections v='800'/>
00:34:45:  <max-request-length v='52428800'/>
00:34:45:  <min-connect-time v='300'/>
00:34:45:  <threads v='1'/>
00:34:45:
00:34:45:  <!-- Logging -->
00:34:45:  <log v='log.txt'/>
00:34:45:  <log-color v='false'/>
00:34:45:  <log-crlf v='true'/>
00:34:45:  <log-date v='false'/>
00:34:45:  <log-date-periodically v='21600'/>
00:34:45:  <log-debug v='true'/>
00:34:45:  <log-domain v='false'/>
00:34:45:  <log-header v='true'/>
00:34:45:  <log-level v='true'/>
00:34:45:  <log-no-info-header v='true'/>
00:34:45:  <log-redirect v='false'/>
00:34:45:  <log-rotate v='true'/>
00:34:45:  <log-rotate-dir v='logs'/>
00:34:45:  <log-rotate-max v='16'/>
00:34:45:  <log-short-level v='false'/>
00:34:45:  <log-simple-domains v='true'/>
00:34:45:  <log-thread-id v='false'/>
00:34:45:  <log-thread-prefix v='true'/>
00:34:45:  <log-time v='true'/>
00:34:45:  <log-to-screen v='true'/>
00:34:45:  <log-truncate v='false'/>
00:34:45:  <verbosity v='5'/>

00:34:45:  <!-- Process Control -->
00:34:45:  <child v='false'/>
00:34:45:  <daemon v='false'/>
00:34:45:  <pid v='false'/>
00:34:45:  <pid-file v='Folding@home Client.pid'/>
00:34:45:  <respawn v='false'/>
00:34:45:  <service v='false'/>

00:34:45:  <!-- Slot Control -->
00:34:45:  <idle v='false'/>
00:34:45:  <max-shutdown-wait v='60'/>
00:34:45:  <pause-on-battery v='true'/>
00:34:45:  <pause-on-start v='false'/>
00:34:45:  <paused v='false'/>
00:34:45:  <power v='full'/>

00:34:45:  <!-- Work Unit Control -->
00:34:45:  <dump-after-deadline v='true'/>
00:34:45:  <max-queue v='16'/>
00:34:45:  <max-units v='0'/>
00:34:45:  <next-unit-percentage v='99'/>
00:34:45:  <stall-detection-enabled v='false'/>
00:34:45:  <stall-percent v='5'/>
00:34:45:  <stall-timeout v='1800'/>
00:34:45:
00:34:45:  <!-- Folding Slots -->
00:34:45:  <slot id='0' type='CPU'>
00:34:45:    <paused v='true'/>
00:34:45:  </slot>
00:34:45:</config>

[7im]

1. Returned back to the v327.23 drivers for better performance.
2. Install both cards, fold on one, use the other for your display.
3. Not related, but set the verbosity setting back to 3. Don't need that extra info, just unnecessary data to have to sort through.

[PantherX]

...By the next day I swapped back to the GTX 295 and was not getting any GPU WUs for days. I tried upgrading to client 7.4.4 to see if that would fix not receiving GPU WUs, no change. I just started receiving GPU WUs yesterday (YAY I can connect with GPU assignment servers)...

Please note that there was an issue with the FahCore_11 Work Servers which was resolved recently (viewtopic.php?f=18&t=26481).

...So here's the issue:, the GPU slot(s) will not run unless set to medium or higher. When set to light the GPU slot(s) status is "waiting for idle"...

That is the default setting. It means that the system will fold on the GPU once the system is detected to be in an idle state (http://folding.stanford.edu/home/faq/fa ... ion/#ntoc3).

...When I do set a single GPU slot to medium or higher, it completes its WUs in about an hour but the whole system because very sluggish to respond and the CPU load in Task Manager stays between 75%-90%+. So basically running GPU processing seems super efficient for WUs but my system is otherwise unuseable during GPU processing...

That is called screen lag and there is nothing that can be done about it unless the GPU vendors decide to implement a scheduling system. While the CPU can prioritize the tasks given to it, a GPU will work on a first-in-first-out basis which is why you encounter the lag. Depending on the hardware and drivers, some donors experience lag while others don't.

...I also noticed that I never get any points for the GPU WUs. There are project numbers and details attached to the GPU WUs but the Base Credit, Estimated Credit, Estimated PPD all say unknown. Estimated TPF usually shows 30 seconds - 3 minutes...

Please note that you do get the points (you can check the official Stats). The reason that those values are displayed as unknown is that the Work Server is running an older version of the code so it can't send that information. Considering that FahCore_11 WUs have gone end-of-life (http://folding.stanford.edu/home/remind ... -core78-2/), I doubt that the code would be updated (since it isn't a trivial task).

...How can I reduce the strain of GPU processing and keep the system useable without issue as it is when processing CPU WUs?...

Is there a specific reason why you are swapping the GPUs so often? As 7im stated, you can install both GPUs (to get 3 GPUs in total) and you can pause (or set the Slot to fold only when idle) the primary GPU (the one connected to your monitor) while the other GPUs can fold. This would prevent the screen lag that you are encountering. Do make sure that your system has enough wattage for the GPUs and that the GPUs are operating within a normal temperature range.

[Ghetto_Child]

I'm using an eVGA nForce 630i mATX motherboard so it has only 1 PCI-e x16 slot. I cannot install both the GTX285 & the GTX295 cards at the same time. I've been swapping cards to test if either card is damaged or not. I recently aquired them 2nd hand. The GTX295 itself is a dual GPU card so F@H allows me to have two GPU folding slots when it is in my system. The screen lag occurs even when I have only configured 1 of 2 F@H GPU slots with the GTX295. I would have thought using half the card's power would make the system balance its responsiveness but for some reason I'm not getting that kind of performance. I would like to add that I run Prime95 24/7. That is configured with a single worker for my single core CPU and lowest priority. I don't know if this is the reason but my PC is never idle, how can I get F@H GPU slots to balance/share processing time with Prime95? My PC doesn't seem to ever be idle.

[PantherX]

...The GTX295 itself is a dual GPU card so F@H allows me to have two GPU folding slots when it is in my system. The screen lag occurs even when I have only configured 1 of 2 F@H GPU slots with the GTX295. I would have thought using half the card's power would make the system balance its responsiveness but for some reason I'm not getting that kind of performance...

You have to ensure that the "correct" half is idle, i.e. if you pause the secondary GPU, you will still encounter the lag since the primary GPU is folding. Use GPU-Z (http://www.techpowerup.com/downloads/SysInfo/GPU-Z/) to ensure that both GPUs are functioning. Then pause one of them and see if the screen lag disappears or not. If it doesn't, then pause the other GPU and screen lag should disappear.

...I would like to add that I run Prime95 24/7. That is configured with a single worker for my single core CPU and lowest priority...

It is possible that your system resources are being over-used. A simple test would be to temporarily stop Prime95 and see if there is any effect on the screen lag or not while the primary GPU is paused. Do note that F@H too uses the lowest priority by default. Thus, you may want to increase the priority of F@H to ensure that the GPU has a constant supply of data. I am not sure how that would effect Prime95 so you may want to test a few different cases and use the one which you find the most usable.

[P5-133XL]

When using the 295 with two folding slots, the GPU/slot associated with the primary display is the one that will cause the most lag while the other one will typically have little lag. The lag you observe is from folding using the entire GPU's capabilities so when the display requests some resources like needing to move the mouse on the screen it has to wait till the current folding chunk is finished.

The video subsystem is not designed to share. There is no way to interrupt an operation, nor a way to prioritize it. If you assign the primary display to one GPU and just fold on the other, there is no conflict and thereby little lag. So the answer to eliminating lag is as simple as switching folding slots.