CUDA invalid value for gpu-architecture

If you think it might be a driver problem, see viewforum.php?f=79

Moderators: Site Moderators, FAHC Science Team

smurfcorpse
Posts: 16
Joined: Wed Dec 16, 2020 3:30 am

CUDA invalid value for gpu-architecture

Post by smurfcorpse »

I have searched for a solution but haven't found anything. The graphics card is an evga Gtx 580.
so far I have done a clean reinstall with the latest drivers for my card (23.21.13.9135 and made sure I do not have any type of CUDA toolkit installed.

Code: Select all

08:07:06:WU00:FS01:Started FahCore on PID 4444
08:07:06:WU00:FS01:Core PID:4400
08:07:06:WU00:FS01:FahCore 0x22 started
08:07:07:WU00:FS01:0x22:*********************** Log Started 2020-12-20T08:07:06Z ***********************
08:07:07:WU00:FS01:0x22:*************************** Core22 Folding@home Core ***************************
08:07:07:WU00:FS01:0x22:       Core: Core22
08:07:07:WU00:FS01:0x22:       Type: 0x22
08:07:07:WU00:FS01:0x22:    Version: 0.0.13
08:07:07:WU00:FS01:0x22:     Author: Joseph Coffland <joseph@cauldrondevelopment.com>
08:07:07:WU00:FS01:0x22:  Copyright: 2020 foldingathome.org
08:07:07:WU00:FS01:0x22:   Homepage: https://foldingathome.org/
08:07:07:WU00:FS01:0x22:       Date: Sep 19 2020
08:07:07:WU00:FS01:0x22:       Time: 02:35:58
08:07:07:WU00:FS01:0x22:   Revision: 571cf95de6de2c592c7c3ed48fcfb2e33e9ea7d3
08:07:07:WU00:FS01:0x22:     Branch: core22-0.0.13
08:07:07:WU00:FS01:0x22:   Compiler: Visual C++ 2015
08:07:07:WU00:FS01:0x22:    Options: /TP /nologo /EHa /wd4297 /wd4103 /O2 /Ob3 /Zc:throwingNew /MT
08:07:07:WU00:FS01:0x22:             -DOPENMM_GIT_HASH="\"189320d0\""
08:07:07:WU00:FS01:0x22:   Platform: win32 10
08:07:07:WU00:FS01:0x22:       Bits: 64
08:07:07:WU00:FS01:0x22:       Mode: Release
08:07:07:WU00:FS01:0x22:Maintainers: John Chodera <john.chodera@choderalab.org> and Peter Eastman
08:07:07:WU00:FS01:0x22:             <peastman@stanford.edu>
08:07:07:WU00:FS01:0x22:       Args: -dir 00 -suffix 01 -version 706 -lifeline 4444 -checkpoint 15
08:07:07:WU00:FS01:0x22:             -opencl-platform 1 -opencl-device 0 -cuda-device 0 -gpu-vendor
08:07:07:WU00:FS01:0x22:             nvidia -gpu 0 -gpu-usage 100
08:07:07:WU00:FS01:0x22:************************************ libFAH ************************************
08:07:07:WU00:FS01:0x22:       Date: Sep 7 2020
08:07:07:WU00:FS01:0x22:       Time: 19:09:56
08:07:07:WU00:FS01:0x22:   Revision: 44301ed97b996b63fe736bb8073f22209cb2b603
08:07:07:WU00:FS01:0x22:     Branch: HEAD
08:07:07:WU00:FS01:0x22:   Compiler: Visual C++ 2015
08:07:07:WU00:FS01:0x22:    Options: /TP /nologo /EHa /wd4297 /wd4103 /O2 /Ob3 /Zc:throwingNew /MT
08:07:07:WU00:FS01:0x22:   Platform: win32 10
08:07:07:WU00:FS01:0x22:       Bits: 64
08:07:07:WU00:FS01:0x22:       Mode: Release
08:07:07:WU00:FS01:0x22:************************************ CBang *************************************
08:07:07:WU00:FS01:0x22:       Date: Sep 7 2020
08:07:07:WU00:FS01:0x22:       Time: 19:08:30
08:07:07:WU00:FS01:0x22:   Revision: 33fcfc2b3ed2195a423606a264718e31e6b3903f
08:07:07:WU00:FS01:0x22:     Branch: HEAD
08:07:07:WU00:FS01:0x22:   Compiler: Visual C++ 2015
08:07:07:WU00:FS01:0x22:    Options: /TP /nologo /EHa /wd4297 /wd4103 /O2 /Ob3 /Zc:throwingNew /MT
08:07:07:WU00:FS01:0x22:   Platform: win32 10
08:07:07:WU00:FS01:0x22:       Bits: 64
08:07:07:WU00:FS01:0x22:       Mode: Release
08:07:07:WU00:FS01:0x22:************************************ System ************************************
08:07:07:WU00:FS01:0x22:        CPU: AMD FX(tm)-6120 Six-Core Processor
08:07:07:WU00:FS01:0x22:     CPU ID: AuthenticAMD Family 21 Model 1 Stepping 2
08:07:07:WU00:FS01:0x22:       CPUs: 6
08:07:07:WU00:FS01:0x22:     Memory: 9.96GiB
08:07:07:WU00:FS01:0x22:Free Memory: 7.81GiB
08:07:07:WU00:FS01:0x22:    Threads: WINDOWS_THREADS
08:07:07:WU00:FS01:0x22: OS Version: 6.1
08:07:07:WU00:FS01:0x22:Has Battery: false
08:07:07:WU00:FS01:0x22: On Battery: false
08:07:07:WU00:FS01:0x22: UTC Offset: -5
08:07:07:WU00:FS01:0x22:        PID: 4400
08:07:07:WU00:FS01:0x22:        CWD: C:\Users\Dad\AppData\Roaming\FAHClient\work
08:07:07:WU00:FS01:0x22:************************************ OpenMM ************************************
08:07:07:WU00:FS01:0x22:   Revision: 189320d0
08:07:07:WU00:FS01:0x22:********************************************************************************
08:07:07:WU00:FS01:0x22:Project: 17420 (Run 0, Clone 804, Gen 221)
08:07:07:WU00:FS01:0x22:Unit: 0x00000000000000000000000000000000
08:07:07:WU00:FS01:0x22:Digital signatures verified
08:07:07:WU00:FS01:0x22:Folding@home GPU Core22 Folding@home Core
08:07:07:WU00:FS01:0x22:Version 0.0.13
08:07:07:WU00:FS01:0x22:  Checkpoint write interval: 25000 steps (2%) [50 total]
08:07:07:WU00:FS01:0x22:  JSON viewer frame write interval: 12500 steps (1%) [100 total]
08:07:07:WU00:FS01:0x22:  XTC frame write interval: 10000 steps (0.8%) [125 total]
08:07:07:WU00:FS01:0x22:  Global context and integrator variables write interval: disabled
08:07:07:WU00:FS01:0x22:There are 4 platforms available.
08:07:07:WU00:FS01:0x22:Platform 0: Reference
08:07:07:WU00:FS01:0x22:Platform 1: CPU
08:07:07:WU00:FS01:0x22:Platform 2: OpenCL
08:07:07:WU00:FS01:0x22:  opencl-device 0 specified
08:07:07:WU00:FS01:0x22:Platform 3: CUDA
08:07:07:WU00:FS01:0x22:  cuda-device 0 specified
08:07:18:Removing old file 'configs/config-20201216-025308.xml'
08:07:18:Saving configuration to config.xml
08:07:18:<config>
08:07:18:  <!-- Folding Core -->
08:07:18:  <core-priority v='low'/>
08:07:18:
08:07:18:  <!-- Network -->
08:07:18:  <proxy v=':8080'/>
08:07:18:
08:07:18:  <!-- Slot Control -->
08:07:18:  <pause-on-battery v='false'/>
08:07:18:  <power v='full'/>
08:07:18:
08:07:18:  <!-- User Information -->
08:07:18:  <passkey v='*****'/>
08:07:18:  <team v='33258'/>
08:07:18:  <user v='smurfcorpse'/>
08:07:18:
08:07:18:  <!-- Folding Slots -->
08:07:18:  <slot id='0' type='CPU'/>
08:07:18:  <slot id='1' type='GPU'>
08:07:18:    <pci-bus v='1'/>
08:07:18:    <pci-slot v='0'/>
08:07:18:  </slot>
08:07:18:</config>
08:07:25:WU00:FS01:0x22:Attempting to create CUDA context:
08:07:25:WU00:FS01:0x22:  Configuring platform CUDA
08:07:26:WU00:FS01:0x22:Failed to create CUDA context:
08:07:26:WU00:FS01:0x22:Error compiling program: nvrtc: error: invalid value for --gpu-architecture (-arch)
08:07:26:WU00:FS01:0x22:
08:07:26:WU00:FS01:0x22:Compilation flags: -arch=compute_20 --use_fast_math
08:07:26:WU00:FS01:0x22:Device compute capability: 2
08:07:26:WU00:FS01:0x22:nvrtc version: 9.2
08:07:26:WU00:FS01:0x22:Attempting to create OpenCL context:
08:07:26:WU00:FS01:0x22:  Configuring platform OpenCL
Any thoughts?
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: CUDA invalid value for gpu-architecture

Post by bruce »

Did you install the drivers from nvidia.com or from some other source?
smurfcorpse
Posts: 16
Joined: Wed Dec 16, 2020 3:30 am

Re: CUDA invalid value for gpu-architecture

Post by smurfcorpse »

Directly from Nvidia
smurfcorpse
Posts: 16
Joined: Wed Dec 16, 2020 3:30 am

Re: CUDA invalid value for gpu-architecture

Post by smurfcorpse »

Code: Select all

03:38:15:WU00:FS01:0x22:        CWD: C:\Users\Dad\AppData\Roaming\FAHClient\work
03:38:15:WU00:FS01:0x22:************************************ OpenMM ************************************
03:38:15:WU00:FS01:0x22:   Revision: 189320d0
03:38:15:WU00:FS01:0x22:********************************************************************************
03:38:15:WU00:FS01:0x22:Project: 17420 (Run 0, Clone 804, Gen 221)
03:38:15:WU00:FS01:0x22:Unit: 0x00000000000000000000000000000000
03:38:15:WU00:FS01:0x22:Digital signatures verified
03:38:15:WU00:FS01:0x22:Folding@home GPU Core22 Folding@home Core
03:38:15:WU00:FS01:0x22:Version 0.0.13
03:38:15:WU00:FS01:0x22:  Checkpoint write interval: 25000 steps (2%) [50 total]
03:38:15:WU00:FS01:0x22:  JSON viewer frame write interval: 12500 steps (1%) [100 total]
03:38:15:WU00:FS01:0x22:  XTC frame write interval: 10000 steps (0.8%) [125 total]
03:38:15:WU00:FS01:0x22:  Global context and integrator variables write interval: disabled
03:38:15:WU00:FS01:0x22:There are 4 platforms available.
03:38:15:WU00:FS01:0x22:Platform 0: Reference
03:38:15:WU00:FS01:0x22:Platform 1: CPU
03:38:15:WU00:FS01:0x22:Platform 2: OpenCL
03:38:15:WU00:FS01:0x22:  opencl-device 0 specified
03:38:15:WU00:FS01:0x22:Platform 3: CUDA
03:38:15:WU00:FS01:0x22:  cuda-device 0 specified
03:38:32:WU00:FS01:0x22:Attempting to create CUDA context:
03:38:32:WU00:FS01:0x22:  Configuring platform CUDA
03:38:32:WU00:FS01:0x22:Failed to create CUDA context:
03:38:32:WU00:FS01:0x22:Error compiling program: nvrtc: error: invalid value for --gpu-architecture (-arch)
03:38:32:WU00:FS01:0x22:
03:38:32:WU00:FS01:0x22:Compilation flags: -arch=compute_20 --use_fast_math
03:38:32:WU00:FS01:0x22:Device compute capability: 2
03:38:32:WU00:FS01:0x22:nvrtc version: 9.2
03:38:32:WU00:FS01:0x22:Attempting to create OpenCL context:
03:38:32:WU00:FS01:0x22:  Configuring platform OpenCL
03:38:39:WU00:FS01:0x22:  Using OpenCL on platformId 1 and gpu 0
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: CUDA invalid value for gpu-architecture

Post by JimboPalmer »

https://www.techpowerup.com/gpu-specs/g ... x-580.c270

The GTX 580 is a Fermi based card, F@H does not support CUDA until Kepler.

If it folds, it will be OpenCL.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
psaam0001
Posts: 383
Joined: Mon May 18, 2020 2:02 am
Location: Ruckersville, Virginia, USA

Re: CUDA invalid value for gpu-architecture

Post by psaam0001 »

I would suggest upgrading your card (if possible), as there are two things to consider:

1) Your card is not supported by the current "latest" drivers from Nvidia (460.89);
2) To use CUDA with F@H 7.6.21, you will need at least v457.30 of the NVidia drivers, and a Kepler based GPU--or newer (Turing; Maxwell; Pascal series; as others have suggested above).

Paul
Last edited by psaam0001 on Sun Dec 20, 2020 6:37 pm, edited 1 time in total.
smurfcorpse
Posts: 16
Joined: Wed Dec 16, 2020 3:30 am

Re: CUDA invalid value for gpu-architecture

Post by smurfcorpse »

A GPU card update is not possible for me without a different motherboard so I'm stuck with what I have for now.

I was worried the card's age was the reason. Thanks for the answer. Not what I wanted to hear but at least I can stop worrying about it.
psaam0001
Posts: 383
Joined: Mon May 18, 2020 2:02 am
Location: Ruckersville, Virginia, USA

Re: CUDA invalid value for gpu-architecture

Post by psaam0001 »

Does your motherboard have at least 1 PCIe x1 slot? If so, try the ASUS GT 710 (model GT710-4H-SL-2GD5). I've got them running on 3 different systems.

Paul
smurfcorpse
Posts: 16
Joined: Wed Dec 16, 2020 3:30 am

Re: CUDA invalid value for gpu-architecture

Post by smurfcorpse »

It's an HP MB https://support.hp.com/us-en/document/c03353629 and finding a graphics card to work with it was a pain. I'll research it a bit to see if it's possible to get a newer gpu with my MB.
psaam0001
Posts: 383
Joined: Mon May 18, 2020 2:02 am
Location: Ruckersville, Virginia, USA

Re: CUDA invalid value for gpu-architecture

Post by psaam0001 »

You might need to go into the BIOS/UEFI settings, before you insert the new card. Look for a setting to get the motherboard to recognize the PCIe graphics card on boot up.

I do see that you can also use something like a low profile x16 length PCIe card. I'd suggest the GT 1030 from Gigabyte (model GV-N1030D5-2GL).

Paul
Last edited by psaam0001 on Thu Dec 24, 2020 2:42 pm, edited 1 time in total.
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: CUDA invalid value for gpu-architecture

Post by bruce »

I have a couple of Kepler GT710s (NOT FERMI) and they meet the deadlines with some elbow room using CUDA. I like them because they fit in a half-height case with a single slot. I think they only use x4 in a x16 slot so if you're forced to truncate PCIe to less that x4, you won't lose as much performance as if you were expecting it to use the full slot width for data. They're great for keeping old PC's folding. The 1030 is a good choice, too, if you have room in the case.
psaam0001
Posts: 383
Joined: Mon May 18, 2020 2:02 am
Location: Ruckersville, Virginia, USA

Re: CUDA invalid value for gpu-architecture

Post by psaam0001 »

I was looking at the specs of his motherboard.

BTW: I had to put my GT 1030 into the x16 slot (where I plan on adding a newer RTX series card in the future), as the other x16 length (electrical x4) slot is too close for comfort to place it in--due to it being a 1.5 slot width card.

That is what has me thinking of the low profile version of that GT 1030 to go in that slot on my re-built system, as the slot I refer to is near the bottom of the micro-ATX case.

Paul
psaam0001
Posts: 383
Joined: Mon May 18, 2020 2:02 am
Location: Ruckersville, Virginia, USA

Re: CUDA invalid value for gpu-architecture

Post by psaam0001 »

bruce wrote:I have a couple of Kepler GT710s (NOT FERMI) and they meet the deadlines with some elbow room using CUDA. I like them because they fit in a half-height case with a single slot. I think they only use x4 in a x16 slot so if you're forced to truncate PCIe to less that x4, you won't lose as much performance as if you were expecting it to use the full slot width for data. They're great for keeping old PC's folding. The 1030 is a good choice, too, if you have room in the case.
3 of the Kepler based GT 710's I'm using, are from ASUS and support the single lane PCIe slot quite nicely. They have 2GB on each card.

However, I do have one 2GB Kepler based GT 710 from EVGA, that still supports the DB-15 VGA output (as well as DVI and HDMI) in my PowerEdge T105. It supports the PCIe x8 slot, but I'm thinking it is only electrically using 4 lanes max. I will likely have to get a few more when I can, as spares (along with a couple 1TB SATA HD's; and a modular 450 watt PSU). As I don't anticipate retiring that system anytime soon--unless it takes a power spike.

I would suggest that smurfcorpse check either Newegg.com, or OutletPC.com if he is in the USA, once the holidays are over (and inventories are replenished...) ::: crossing fingers on one hand :::

Paul
smurfcorpse
Posts: 16
Joined: Wed Dec 16, 2020 3:30 am

Re: CUDA invalid value for gpu-architecture

Post by smurfcorpse »

just to report back.. I did update my GPU. The GTX 1030 couldn't beat my current GTX 580 in overall performance. I didn't want to take a step backwards in performance so I went with the Gigabyte GTX 1050 Ti OC Low Profile 4G. Kinda weird buying the GTX 1050. I canceled it a few times due to back order dates and looked at an MSI version as well. NewEgg finally had it in stock. Got it for 170.00 but prices jumped all over the place along with availability. I really didn't want to spend that much on an old system but so far it looks like I'll see a big boost in my PPD.

Thanks all for the help!
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: CUDA invalid value for gpu-architecture

Post by JimboPalmer »

Thanks for reporting back!
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
Post Reply