Going to swap GPU - how does this affect my folding config?

[Crawdaddy79]

Hello Folding Forum,

I know this is kind of a dumb question, but when swapping GPU, will it require a reinstallation of F@H?

I do plan on finishing the current WU before doing the deed, but am hoping that the next time the client is run, it will pick up my other GPU and not try to download WUs that are incompatible (but compatible with my current GPU).

I'll be RMAing my Vega 64 in the hopes of fixing my crashing issues and temporarily replacing it with my Radeon HD 7970.

[Joe_H]

From various reports, it will depend a bit on which OS. On Linux, once the drivers are installed and the system restarted the client should be able to pick up the new GPU. You may need to manually add the GPU folding slot.

On Windows people have reported better success with doing a reinstall after installing the GPU and its drivers. Changes in the most recent version of the client may make that less necessary, but no track record yet from peoples' reports.

[JimboPalmer]

The current AMD driver should work for the Radeon HD 7970. So Removing the GPU slot then re Adding it, should be enough.

[Crawdaddy79]

Thanks - I wanted to avoid a scenario where I just permanently didn't have a GPU slot. I'm really curious to see how this card will perform.

I'm using Windows 10.

In deciding to be a guinea pig, I did not uninstall F@H. Using client 7.6.13.

My steps: finished the WU, removed the GPU slot from the client. Closed it. Uninstalled drivers. Swapped cards. Reinstalled drivers (with Factory reset option). Started F@H client. Added GPU slot. It immediately picked up one of my Vega's problem child WUs, a 16435 Project. 2m 10s TPF is now 6m 45s

Hopefully it doesn't crash. Also I forgot how loud this card was.

All in all, relatively painless. Thanks for the help.

[BobWilliams757]

My steps: finished the WU, removed the GPU slot from the client. Closed it. Uninstalled drivers. Swapped cards. Reinstalled drivers (with Factory reset option). Started F@H client. Added GPU slot. It immediately picked up one of my Vega's problem child WUs, a 16435 Project. 2m 10s TPF is now 6m 45s

I hope you get the issues with your Vega sorted out. At least the swap out was painless, and possibly helpful for others in the same boat.

As for 16435 I remember the trouble it gave you with the Vega, it will be interesting to see how it does with your "new" older card.

As for me 16435 is like groundhog day. It accounts for some 20% of the WU's I have done since i set up logging with HFM.

And just to give you some perspective of how painful it can be......

With a slight memory OC (3200) on my 2400G TPF is in the 13m 42s range.

Now if bump the graphics core to 1550 it really blazes, bringing TPF all the way down to the 12m 15s range.


With bumping both with the OC I'm still in for a 20+ hour fold.

[Crawdaddy79]

Thanks Mr. BobWilliams for the encouragement.

Unfortunately, after letting it run all night, one of the three fan bearings is very angry with me. I did look at my history and saw that I got 10k points for a 13415 WU, which was only 5k less than the Vega. I even saved it from being faulty (four others returned Faulty 2). https://apps.foldingathome.org/wu?p=134 ... ne=6&gen=0

Gave the card a two hour break to cool down, but the fan rattle did not reduce. I've got 11 hours left on a 16444 WU, and once it finishes, I think I'm done GPU folding until my Vega gets back. I may not even let it finish.

[BobWilliams757]

It's amazing how the lesser GPUs can eat up those small atom count WU's almost as well as the power cards. I guess anything above a certain level is like trying to drive a Ferrari on ice.... it can be done, just slowly.

My 2400G with Vega 11 just hit a record new high PPD return on a 13415 yesterday when it grabbed 3 of them in in row.


Bummer about the fan. But the pain of long folds I do understand.

[Endgame124]

My R7 250X has been cruising along at about 85,000 points per day on these work units, so I figured why not try the integrated R250 on my Kaveri. Yeah... that one is only getting around 20K per day, and it cuts the cpu speed. Guess you can’t win them all.

[JimboPalmer]

My R7 250X has been cruising along at about 85,000 points per day on these work units, so I figured why not try the integrated R250 on my Kaveri. Yeah... that one is only getting around 20K per day, and it cuts the cpu speed. Guess you can’t win them all.

Every GPU takes away one CPU from the CPU slot, fortunately most GPUs produce enough points that you get more points than you lose.

There are high end CPUs that rivals GPUs.
There are low end GPUs that rival CPUs. (As F@H begins to work on Intel iGPUs, watch to see that your overall points don't go down with GPU folding)

A GTS 450 may make so few points that your CPU with 4 threads is faster than your CPU with 3 threads and a GTS 450, as an example.

[Crawdaddy79]

This may be helpful to someone out there.

Got the Vega 64 back. The HD 7970 was at a 9 hour ETA for its current work unit, and I didn't wait for it to finish.

Paused both CPU/GPU slot.
Uninstalled Drivers. Rebooted. Shut down.
Swapped video card.
Powered on PC.
Factory Reset option checked, installed new drivers.
Did some initial tests to see if the video card worked at least as good as it did before I sent it in.

Started F@H - GPU WU got dumped. CPU WU continued on.

Loggy woof woof:

Code: Select all

*********************** Log Started 2020-07-28T14:01:57Z ***********************
14:01:57:Trying to access database...
14:01:57:Successfully acquired database lock
14:01:57:Read GPUs.txt
14:01:57:Enabled folding slot 00: PAUSED cpu:14 (by user)
14:01:59:Enabled folding slot 01: PAUSED gpu:0:Vega 10 XL/XT [Radeon RX Vega 56/64] (by user)
14:01:59:WARNING:WU02:No longer matches Slot 1's configuration and there are no other matching slots, dumping
14:01:59:****************************** FAHClient ******************************
14:01:59:        Version: 7.6.13
14:01:59:         Author: Joseph Coffland <joseph@cauldrondevelopment.com>
14:01:59:      Copyright: 2020 foldingathome.org
14:01:59:       Homepage: https://foldingathome.org/
14:01:59:           Date: Apr 27 2020
14:01:59:           Time: 21:21:01
14:01:59:       Revision: 5a652817f46116b6e135503af97f18e094414e3b
14:01:59:         Branch: master
14:01:59:       Compiler: Visual C++ 2008
14:01:59:        Options: /TP /nologo /EHa /wd4297 /wd4103 /Ox /MT
14:01:59:       Platform: win32 10
14:01:59:           Bits: 32
14:01:59:           Mode: Release
14:01:59:         Config: G:\FoldingData\config.xml
14:01:59:******************************** CBang ********************************
14:01:59:           Date: Apr 24 2020
14:01:59:           Time: 17:07:55
14:01:59:       Revision: ea081a3b3b0f4a37c4d0440b4f1bc184197c7797
14:01:59:         Branch: master
14:01:59:       Compiler: Visual C++ 2008
14:01:59:        Options: /TP /nologo /EHa /wd4297 /wd4103 /Ox /MT
14:01:59:       Platform: win32 10
14:01:59:           Bits: 32
14:01:59:           Mode: Release
14:01:59:******************************* System ********************************
14:01:59:            CPU: AMD Ryzen 7 2700X Eight-Core Processor
14:01:59:         CPU ID: AuthenticAMD Family 23 Model 8 Stepping 2
14:01:59:           CPUs: 16
14:01:59:         Memory: 31.95GiB
14:01:59:    Free Memory: 23.75GiB
14:01:59:        Threads: WINDOWS_THREADS
14:01:59:     OS Version: 6.2
14:01:59:    Has Battery: false
14:01:59:     On Battery: false
14:01:59:     UTC Offset: -4
14:01:59:            PID: 17360
14:01:59:            CWD: G:\FoldingData
14:01:59:  Win32 Service: false
14:01:59:             OS: Windows 10 Home
14:01:59:        OS Arch: AMD64
14:01:59:           GPUs: 1
14:01:59:          GPU 0: Bus:31 Slot:0 Func:0 AMD:5 Vega 10 XL/XT [Radeon RX Vega 56/64]
14:01:59:           CUDA: Not detected: Failed to open dynamic library 'nvcuda.dll': The
14:01:59:                 specified module could not be found.
14:01:59:
14:01:59:OpenCL Device 0: Platform:0 Device:0 Bus:31 Slot:0 Compute:1.2 Driver:3110.7
14:01:59:******************************* libFAH ********************************
14:01:59:           Date: Apr 15 2020
14:01:59:           Time: 14:53:14
14:01:59:       Revision: 216968bc7025029c841ed6e36e81a03a316890d3
14:01:59:         Branch: master
14:01:59:       Compiler: Visual C++ 2008
14:01:59:        Options: /TP /nologo /EHa /wd4297 /wd4103 /Ox /MT
14:01:59:       Platform: win32 10
14:01:59:           Bits: 32
14:01:59:           Mode: Release
14:01:59:***********************************************************************
14:01:59:<config>
14:01:59:  <!-- Network -->
14:01:59:  <proxy v=':8080'/>
14:01:59:
14:01:59:  <!-- User Information -->
14:01:59:  <passkey v='*****'/>
14:01:59:  <team v='64'/>
14:01:59:  <user v='Crawdaddy79'/>
14:01:59:
14:01:59:  <!-- Folding Slots -->
14:01:59:  <slot id='0' type='CPU'>
14:01:59:    <paused v='true'/>
14:01:59:  </slot>
14:01:59:  <slot id='1' type='GPU'>
14:01:59:    <paused v='true'/>
14:01:59:  </slot>
14:01:59:</config>
14:01:59:WU02:FS01:Sending unit results: id:02 state:SEND error:DUMPED project:16909 run:5 clone:44 gen:55 core:0x22 unit:0x0000004f0002894c5ef4e617f80ded82
14:01:59:WU02:FS01:Connecting to 155.247.166.220:8080
14:02:00:WU02:FS01:Server responded WORK_ACK (400)
14:02:00:WU02:FS01:Cleaning up

[HaloJones]

not unexpected that a unit started on one card won't continue on another one.

has the Vega picked up a new unit?

[Crawdaddy79]

It did, almost immediately. Everything works. Client detected and displayed Vega. The only abnormality is the WU dump. I looked up that particular one and sadly it looked like mine would have been a good return after four faulty ones.

[Neil-B]

I wouldn't call it an "abnormality" - it is doing exactly what would be expected tbh !!