Nvidia hotfix driver 376.48 that fix folding

[foldy]

Nvidia hotfix driver 376.48 that fix folding.
http://nvidia.custhelp.com/app/answers/detail/a_id/4288

[cruswi]

... works for me

[ChristianVirtual]

... works for me

Impact on performance, positive or negative ? Can you compare TFP before and after ?

[cruswi]

... works for me

Impact on performance, positive or negative ? Can you compare TFP before and after ?

No, no noticeable changes in performance. Tested both drivers with two GTX 980 working on PRCG 9179 and 13112 at about 90% of done WUs: Deviations of Estimated TPF are marginal, not evident (1-3 seconds).

[foldy]

Same here, work unit 11407 (9, 28, 22)
372.90 TPF 3:15
376.48 TPF 3:19
So only a marginal difference.

[foldy]

After several work units finished successful with this driver, now a work unit failed with clEnqueueReadBuffer (-5) at 84%

Code: Select all

*********************** Log Started 2017-01-06T12:28:13Z ***********************
12:28:13:************************* Folding@home Client *************************
12:28:13:        Website: http://folding.stanford.edu/
12:28:13:      Copyright: (c) 2009-2016 Stanford University
12:28:13:         Author: Joseph Coffland <joseph@cauldrondevelopment.com>
12:28:13:           Args: --open-web-control
12:28:13:         Config: C:\ProgramData\FAHClient\config.xml
12:28:13:******************************** Build ********************************
12:28:13:        Version: 7.4.16
12:28:13:           Date: Jan 6 2017
12:28:13:           Time: 00:25:14
12:28:13:     Repository: Git
12:28:13:       Revision: a9e9e27dc2ee6ff01398c439677bc27f6cb74032
12:28:13:         Branch: master
12:28:13:       Compiler: Visual C++ 2008
12:28:13:        Options: /TP /nologo /EHa /wd4297 /wd4103 /Ox -arch:SSE /MT
12:28:13:       Platform: win32 10
12:28:13:           Bits: 32
12:28:13:           Mode: Release
12:28:13:******************************* System ********************************
12:28:13:            CPU: Intel(R) Core(TM) i5-2500K CPU @ 3.30GHz
12:28:13:         CPU ID: GenuineIntel Family 6 Model 42 Stepping 7
12:28:13:           CPUs: 4
12:28:13:         Memory: 15.60GiB
12:28:13:    Free Memory: 10.16GiB
12:28:13:        Threads: WINDOWS_THREADS
12:28:13:     OS Version: 6.1
12:28:13:    Has Battery: false
12:28:13:     On Battery: false
12:28:13:     UTC Offset: 1
12:28:13:            PID: 4344
12:28:13:            CWD: C:\ProgramData\FAHClient
12:28:13:             OS: Windows 7 Professional
12:28:13:        OS Arch: AMD64
12:28:13:           GPUs: 2
12:28:13:          GPU 0: Bus:1 Slot:0 Func:0 ATI:5 Tahiti PRO [Radeon R9 280/HD
12:28:13:                 7900/8950]
12:28:13:          GPU 1: Bus:2 Slot:0 Func:0 NVIDIA:5 GM204 [GeForce GTX 970]
12:28:13:  CUDA Device 0: Platform:0 Device:0 Bus:2 Slot:0 Compute:5.2 Driver:8.0
12:28:13:OpenCL Device 0: Platform:0 Device:0 Bus:1 Slot:0 Compute:1.2 Driver:2117.13
12:28:13:OpenCL Device 2: Platform:1 Device:0 Bus:2 Slot:0 Compute:1.2 Driver:376.48
12:28:13:  Win32 Service: false
12:28:13:***********************************************************************
12:28:13:<config>
12:28:13:  <!-- Client Control -->
12:28:13:  <disable-sleep-when-active v='false'/>
12:28:13:
12:28:13:  <!-- Folding Core -->
12:28:13:  <checkpoint v='5'/>
12:28:13:
12:28:13:  <!-- Folding Slot Configuration -->
12:28:13:  <client-type v=''/>
12:28:13:  <max-packet-size v='small'/>
12:28:13:
12:28:13:  <!-- Network -->
12:28:13:  <proxy v=':8080'/>
12:28:13:
12:28:13:  <!-- Slot Control -->
12:28:13:  <power v='FULL'/>
12:28:13:
12:28:13:  <!-- User Information -->
12:28:13:  <passkey v='********************************'/>
12:28:13:  <team v='224497'/>
12:28:13:  <user v='xxx'/>
12:28:13:
12:28:13:  <!-- Work Unit Control -->
12:28:13:  <next-unit-percentage v='100'/>
12:28:13:
12:28:13:  <!-- Folding Slots -->
12:28:13:  <slot id='1' type='GPU'>
12:28:13:    <gpu-index v='0'/>
12:28:13:  </slot>
12:28:13:  <slot id='0' type='GPU'>
12:28:13:    <gpu-index v='1'/>
12:28:13:  </slot>
12:28:13:</config>
12:28:13:Trying to access database...
12:28:13:Successfully acquired database lock
12:28:13:Enabled folding slot 01: READY gpu:0:Tahiti PRO [Radeon R9 280/HD 7900/8950]
12:28:13:Enabled folding slot 00: READY gpu:1:GM204 [GeForce GTX 970]
...
17:06:48:WU00:FS00:0x21:Completed 2100000 out of 2500000 steps (84%)
17:07:41:WU00:FS00:0x21:ERROR:exception: Error downloading array energyBuffer: clEnqueueReadBuffer (-5)
17:07:41:WU00:FS00:0x21:Saving result file logfile_01.txt
17:07:41:WU00:FS00:0x21:Saving result file log.txt
17:07:41:WU00:FS00:0x21:Folding@home Core Shutdown: BAD_WORK_UNIT
17:07:42:WARNING:WU00:FS00:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
17:07:42:WU00:FS00:Sending unit results: id:00 state:SEND error:FAULTY project:9191 run:2 clone:24 gen:151 core:0x21 unit:0x000000dfab40415457cb2ce1fb67e874

[silverpulser]

Elsewhere it has been said that:-

But Folders are noticing a 100K-300K drop in Folding points per card with these drivers.


thats because theres a bug in the fah_core that needs resolving, the nvidia side fix was a work around that reduces performance.

[7im]

Elsewhere it has been said that:-

But Folders are noticing a 100K-300K drop in Folding points per card with these drivers.

thats because theres a bug in the fah_core that needs resolving, the nvidia side fix was a work around that reduces performance.

Because the points curve is so steep in the faster GPUs, even a few seconds per frame, which is truly marginal in reality, creates an unrealistic difference in points, aka the 100K-300K difference you saw mentioned elsewhere.

[Aurum]

My Nvidia GPUs do not fold as fast as some have reported using driver 376.48, e.g. GTX 1070 at ~500,000 versus 703,238. Can someone please clarify something for me? Would going back to an earlier driver fold at full rate or is hotfix as fast as Nvidia GPUs go

[bruce]

I've seen several reports that the hotfix driver does, in fact, slow down folding, compared to earlier versions of 375.xx/376.xx. When a hot-fix is developed, the objective is to make it work, even if performance is impacted. I suspect that somebody at NV is still working on the issue and eventually there will be a working version of the OpenCL drivers that is more optimized. [I hope my assumption pans out eventually.]

I don't have solid data regarding the differences you're asking about, but I'd probably try it myself. Personally, the "game ready" feature isn't important to me so I'm still on the older version (and my GPUs are still pre 10xx -- though I'll probably upgrade the next time prices come down ... "soon."

NV has said that the FAHCore_21 has a bug. FAH-Development says that the drivers have a bug. Whether either is true (or maybe even both), the hotfix drivers are a temporary solution and FAH-Development is .