GPU stuck on READY

[ajm]

Code: Select all

sudo apt install ocl-icd-opencl-dev

and reboot.

[HugoNotte]

So, this was happening before, and still is -- why does my CPU still crank up to 100% on Medium? I thought it was supposed to be about 75%?

i5 8600k OC'd @ 5GHz

I think I do need a rear exhaust fan but until then, I'm hovering around 80-90C on Medium; meanwhile my GPU just shrugs, barely breaks a sweat. I'm afraid to leave it on Full for too long, for fear of overheating.

The high temperatures would be a result of 2 things:
1) FAH uses AVX instructions, which make the CPU work harder than most other work your CPU would usually encounter. This creates more heat than keeping the CPU at 100% with a non AVX task.
2) your overclock of the CPU. It creates more heat than standard. If your cooling solution is good enough for gaming & a 5 GHz overclock, that doesn't mean it would be sufficient for FAH at 5 GHz.

Both factors together result in the high temperatures you are experiencing. You might get more work done on the CPU if you clock the CPU at standard speeds rather than reducing the amount of cores available to FAH until you have upgraded your cooling system.

It is true, that compared to GPU folding, CPU WUs earn a lot less points. Hence a lot of people only fold on their GPUs. However, the CPU can do tasks that the GPU can't do and therefore certain projects can only run on CPUs. CPUs are important to FAH.
Depending on your motivation, be it that you want to support science to the best of your abilities or be it that you just want to accumulate points, decide whether you want to use your CPU and dedicate 4 cores to FAH.

[JimboPalmer]

So, this was happening before, and still is -- why does my CPU still crank up to 100% on Medium? I thought it was supposed to be about 75%?

i5 8600k OC'd @ 5GHz.

F@H scheme would work out to that way with 4 threads, yes. The actual choices are Light = 50% of threads, Medium = all threads but one, and Full = all threads

With 6 threads that is over 83% used by Core_a7.

https://ark.intel.com/content/www/us/en ... 0-ghz.html

[bruce]

This has been reported to work:
> sudo apt update
> sudo apt install ocl-icd-opencl-dev

If the library is already present, it may be as simple as adding a link between the actual library (that has some revision numbers in it's name) and the basic name libOpenCL.so so FAHClient can find it.

Please report back.

[SteffenCBC]

ocl-icd-opencl-dev did the trick - now the GPU folds.

Thank you for your quick help!

[ajm]

Great, thank you for reporting back!

Come to think of it, it is such a common issue that the line

Code: Select all

sudo apt install ocl-icd-opencl-dev

should be added in https://foldingathome.org/support/faq/i ... -advanced/
and/or even in https://foldingathome.org/support/faq/i ... uirements/

[SteffenCBC]

In addition I am trying to get the GPU folding in my notebook as well - and guess what, I get other errors. The OS is again an Ubuntu 18.4, the GPU is from AMD (Manhattan [Mobility Radeon HD 5000 Series]).

1. Should folding work with that GPU at all? It's not from Nvidia.
2. I installed ocl-icd-opencl-dev, but I suppose there are other packages missing, again. Please give me a hint.

Here's the log for more details.

Code: Select all

*********************** Log Started 2020-05-25T08:24:59Z ***********************
08:24:59:****************************** FAHClient ******************************
08:24:59:    Version: 7.6.9
08:24:59:     Author: Joseph Coffland <joseph@cauldrondevelopment.com>
08:24:59:  Copyright: 2020 foldingathome.org
08:24:59:   Homepage: https://foldingathome.org/
08:24:59:       Date: Apr 17 2020
08:24:59:       Time: 18:11:26
08:24:59:   Revision: 398c2b17fa535e0cc6c9d10856b2154c32771646
08:24:59:     Branch: master
08:24:59:   Compiler: GNU 8.3.0
08:24:59:    Options: -std=c++11 -ffunction-sections -fdata-sections -O3 -funroll-loops
08:24:59:             -fno-pie
08:24:59:   Platform: linux2 4.19.0-5-amd64
08:24:59:       Bits: 64
08:24:59:       Mode: Release
08:24:59:       Args: --child /etc/fahclient/config.xml --run-as fahclient
08:24:59:             --pid-file=/var/run/fahclient.pid --daemon
08:24:59:     Config: /etc/fahclient/config.xml
08:24:59:******************************** CBang ********************************
08:24:59:       Date: Apr 17 2020
08:24:59:       Time: 18:10:13
08:24:59:   Revision: 2fb0be7809c5e45287a122ca5fbc15b5ae859a3b
08:24:59:     Branch: master
08:24:59:   Compiler: GNU 8.3.0
08:24:59:    Options: -std=c++11 -ffunction-sections -fdata-sections -O3 -funroll-loops
08:24:59:             -fno-pie -fPIC
08:24:59:   Platform: linux2 4.19.0-5-amd64
08:24:59:       Bits: 64
08:24:59:       Mode: Release
08:24:59:******************************* System ********************************
08:24:59:        CPU: Intel(R) Core(TM) i3 CPU M 370 @ 2.40GHz
08:24:59:     CPU ID: GenuineIntel Family 6 Model 37 Stepping 5
08:24:59:       CPUs: 4
08:24:59:     Memory: 3.71GiB
08:24:59:Free Memory: 2.96GiB
08:24:59:    Threads: POSIX_THREADS
08:24:59: OS Version: 5.3
08:24:59:Has Battery: true
08:24:59: On Battery: false
08:24:59: UTC Offset: 2
08:24:59:        PID: 1172
08:24:59:        CWD: /var/lib/fahclient
08:24:59:         OS: Linux 5.3.0-53-generic x86_64
08:24:59:    OS Arch: AMD64
08:24:59:       GPUs: 1
08:24:59:      GPU 0: Bus:2 Slot:0 Func:0 AMD:0 Manhattan [Mobility Radeon HD 5000
08:24:59:             Series]
08:24:59:       CUDA: Not detected: Failed to open dynamic library 'libcuda.so':
08:24:59:             libcuda.so: cannot open shared object file: No such file or
08:24:59:             directory
08:24:59:     OpenCL: Not detected: clGetPlatformIDs() returned -1001
08:24:59:******************************* libFAH ********************************
08:24:59:       Date: Apr 15 2020
08:24:59:       Time: 21:43:24
08:24:59:   Revision: 216968bc7025029c841ed6e36e81a03a316890d3
08:24:59:     Branch: master
08:24:59:   Compiler: GNU 8.3.0
08:24:59:    Options: -std=c++11 -ffunction-sections -fdata-sections -O3 -funroll-loops
08:24:59:             -fno-pie
08:24:59:   Platform: linux2 4.19.0-5-amd64
08:24:59:       Bits: 64
08:24:59:       Mode: Release
08:24:59:***********************************************************************

If folding does not work with that GPU, it's ok, I am just trying to give my hardware's best.

[ajm]

Code: Select all

08:24:59:     OpenCL: Not detected: clGetPlatformIDs() returned -1001

For FAH to work with AMD cards, you have to download the card specific drivers from https://www.amd.com/en/support and then the least problematic is to install the legacy headless version of the Pro Variant: https://amdgpu-install.readthedocs.io/e ... ro-variant

Code: Select all

./amdgpu-pro-install -y --opencl=pal,legacy --headless

But it's a bit of a game of chance. It can work right out of the box or demand endless tweaking, and some just don't get there.

[JimboPalmer]

I do not believe any of the HD 5000 mobility GPUs support Double Precision floating point math, (FP64) a show stopper.

Even if you get OpenCL to load it won't fold successfully.

https://en.wikipedia.org/wiki/List_of_A ... 000_series

https://www.techpowerup.com/gpu-specs/m ... -5870.c338

[Joe_H]

The HD 5000 series GPUs for the desktop and mobile applications were the last that shared chip designs. You would need a GPU from the HD 5800M to have even a chance for Double Precision support, but from what I have found so far AMD did not include that in the Mobility GPUs. Do you happen to know the exact model? The device ID that the F@h client detected is sort of generic, AMD/ATI used the same ID number for several different models.

But the bigger problem under Linux will be drivers. AMD dropped full support for the Terascale and GCN 1 based GPUs about 2015 on Linux. So getting older proprietary drivers and OpenCL support is officially limited to Ubuntu 14.04 LTS or equivalent. People have hacked the drivers to work on later versions of Linux, but that appears to be a bit involved from what I have read so far.

Around 2015 is also when AMD/ATI moved driver support for many of these older cards to legacy support, and some of the most recent drivers that include video drivers for Terascale and early GCN 1 based GPUs may not include working OpenCL support for them.

[SteffenCBC]

That sounds like a quite complicated puzzle - I will not try. Thank very much you for your advice!