Folding Forum

Posted: **Mon Jun 19, 2017 2:51 pm**

Without a GPU, you have a non prime number of CPUs, and so it works.

With a GPU, it subtracts 1 CPU for use supporting the GPU, and that that leaves you with a Prime number of CPUs. F@H dislikes large primes and multiples of large primes (over 5, sometimes over 3)

24 is 2 * 2 * 2 *3 all is well but then the GPU stole one.
23 is Prime, Bad
22 is 2 * 11, Bad
21 is 3 * 7 Bad
20 is 2 * 2 * 5, Might work
19 is prime, Bad
18 is 2 * 3 * 3, Good!

Try a 20 slot CPU and a 3 slot CPU, and if you have issues with the 20, try a 18 slot CPU and a 4 slot CPU, a definite good choice.

Normally you want the greatest number of CPUs in a single slot so you get the largest Quick Return Bonus and do the most science. However, you CAN have multiple CPU slots just so long as they do not exceed your physical number of CPUs. (23 in your case as one is used by the GPU slot)

Posted: **Mon Jun 19, 2017 3:01 pm**

Set the core count to a multiple of 2, 3, and possibly 5 after pausing the CPU slot, then un-pause the slot. Multiple of higher primes than 5 will not work, and sometimes projects that will assign to a multiple of 5 will be in short supply. So with your CPU a setting of 16, 18 or 20 should get work assigned. It is important to pause the CPU slot, otherwise the WU request usually sticks to the last number set until a restart.

The other option is the beta client version, 7.4.16. It has new code that negotiates with the AS code to get WU's with the highest available CPU thread count, up to a maximum equal to the number set at the time of the WU request.

Posted: **Mon Jun 19, 2017 4:34 pm**

Aurum wrote:On the Slots/Configure folding slots tab I've tried setting the number of threads to use for folding to -1 and 22. Using -1 FAH sets it to 23.

This has nothing to do with which version of Linux you're running. FAHCore_a7 only get work if the CPU count contains no prime factors >5.

24 would work but one is needed by your GPU.
23 is prime.
22 has the factor 11
21 has the factor 7
20=5*2*2 and might work (I'm not sure about 5)
19 is prime
18=3*3*2 will certainly work if there happen to be projects that allow that many
(etc.)

When the latest version of GROMACS was incorporated into FAHCore_a7, it added a lot of restrictions which do avoid problems that earlier versions of GROMACS has always had, but FAHClient 7.4.4 doesn't know about that. FAH Development is still trying to recover.

It should be noted that setting the CPU count to -1 will let v7.4.16 choose correctly. The valid CPU count decision are made automatically in the latest beta client but not in 7.4.4 where the choice is inflexible.

Posted: **Mon Jun 19, 2017 4:48 pm**

I have six 12 & 14 core Xeons and have not seen this happen before on Win7 rigs, probably because they're all running 7.4.15 and I've never run 7.4.4. This is my first attempt at using Linux and the only option I found so far was to install 7.4.4 and when I looked I did not find the beta versions as they're not advertised. I'll look again and hopefully I can figure out how to get Linux to upgrade.

Pause and 20 or 18 did not work. So I removed the CPU and then added a 12 and a 10 and it updated the cores and DLed WUs. One has an A7 and the other an A4 and both have TPFs ~1:30 minutes.

With Win7 I would watch the Task Manager Performance and pick the number of CPU threads that kept it at 80% or less. Don't know how to do that yet in Linux.

What version of Ubuntu should I install on the next rig

Posted: **Mon Jun 19, 2017 4:52 pm**

bruce wrote:...setting the CPU count to -1 will let v7.4.16 choose correctly. The valid CPU count decision are made automatically in the latest beta client but not in 7.4.4 where the choice is inflexible.

I noticed that with Win7 and F@H 7.4.15 CPU at "-1" resulted in CPU Usage in the mid 90s and I fear this reduces GPU performance. Depending on the CPU I would select 1, 2 or 3 fewer CPU threads than "-1" would pick.

Posted: **Mon Jun 19, 2017 6:48 pm**

Aurum I would try 16 and 6, as the most CPUs in one slot should get the most points. But maybe today you just want to see it complete one, I understand.

Posted: **Mon Jun 19, 2017 7:28 pm**

Aurum wrote:
Joe_H wrote:What is the CPU setting that you are using on your CPU slot?
It's a Xeon E5-4640 v3 @ 1.90GHz with 12 cores and 24 threads. On the Slots/Configure folding slots tab I've tried setting the number of threads to use for folding to -1 and 22. Using -1 FAH sets it to 23.

BTW, I just picked Linux Mint because it came up first on google search and I hadn't seen the posts here saying use Ubuntu x.xx. I'm perfectly willing to replace it with a better version of Linux before I propagate it across the farm.

Either set the CPU slot manually to a number that avoids prime factors >5, (try 20, if that doesn't work 18, finally 16). Or install beta client 7.4.16 which will negotiate with the servers to avoid the problem.

Posted: **Mon Jun 19, 2017 7:39 pm**

As Jimbo Palmer suggests, configuring V7.4.4 for two slots (one set for 16 and another for 6 ... and another for 1 CPU, if you so choose) ... will increase the CPU utilization above the mid 90s. There's nothing wrong with that. The only remaining problem is that sometimes there may be no WUs that allow 16 CPUs to be used. V7.4.4 will keep trying to get a 16-CPU assignment and will remain idle until it finds one. V7.4.15 will search for another project which will assign the maximum number of CPUs, less than the maximum you've set.

In my book, running a WU at less that 100% is better than not running a WU at all.

Posted: **Tue Jun 20, 2017 1:47 am**

New mystery, I installed F@H 7.4.16 for Linux Mint and now the CPUs work fine but the GPU stopped working. I tried Removing it and Adding. I tried rebooting.
Driver Manager says I have nvidia 375.66 installed (worked fine with F@Hv7.4.4) and is now offering me an upgrade to 381.22.
What can I do? This pattern just keeps repeating:

Code: Select all

01:18:47:WU01:FS01:Download 71.13%
01:18:52:WU01:FS01:Download complete
01:18:52:WU01:FS01:Received Unit: id:01 state:DOWNLOAD error:NO_ERROR project:10496 run:106 clone:59 gen:2 core:0x21 unit:0x000000068ca304f558897334530bc677
01:18:52:WU01:FS01:Starting
01:18:52:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:18:52:WU01:FS01:Starting
01:18:52:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:19:52:WU01:FS01:Starting
01:19:52:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:21:30:WU01:FS01:Starting
01:21:30:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:24:07:WU01:FS01:Starting
01:24:07:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:28:21:WU01:FS01:Starting
01:28:21:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:28:21:WU01:FS01:Sending unit results: id:01 state:SEND error:FAILED project:10496 run:106 clone:59 gen:2 core:0x21 unit:0x000000068ca304f558897334530bc677
01:28:21:WU01:FS01:Connecting to 140.163.4.245:8080
01:28:22:WU01:FS01:Server responded WORK_ACK (400)
01:28:22:WU01:FS01:Cleaning up
01:28:22:WU03:FS01:Connecting to 171.67.108.45:80
01:28:22:WU03:FS01:Assigned to work server 140.163.4.245
01:28:22:WU03:FS01:Requesting new work unit for slot 01: READY gpu:0:GP102 [GeForce GTX 1080 Ti] 11380 from 140.163.4.245
01:28:22:WU03:FS01:Connecting to 140.163.4.245:8080
01:28:22:WU03:FS01:Downloading 14.50MiB
01:28:28:WU03:FS01:Download 38.37%
01:28:34:WU03:FS01:Download 74.16%
01:28:39:WU03:FS01:Download complete
01:28:39:WU03:FS01:Received Unit: id:03 state:DOWNLOAD error:NO_ERROR project:10496 run:6 clone:23 gen:99 core:0x21 unit:0x0000007e8ca304f55889580aded96963
01:28:39:WU03:FS01:Starting
01:28:39:ERROR:WU03:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:28:40:WU03:FS01:Starting
01:28:40:ERROR:WU03:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:29:40:WU03:FS01:Starting
01:29:40:ERROR:WU03:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:31:17:WU03:FS01:Starting
01:31:17:ERROR:WU03:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:33:54:WU03:FS01:Starting
01:33:54:ERROR:WU03:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:38:09:WU03:FS01:Starting
01:38:09:ERROR:WU03:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:38:09:WU03:FS01:Sending unit results: id:03 state:SEND error:FAILED project:10496 run:6 clone:23 gen:99 core:0x21 unit:0x0000007e8ca304f55889580aded96963
01:38:09:WU03:FS01:Connecting to 140.163.4.245:8080
01:38:14:WU01:FS01:Connecting to 171.67.108.45:80
01:38:14:WU03:FS01:Server responded WORK_ACK (400)
01:38:14:WU03:FS01:Cleaning up
01:38:16:WU01:FS01:Assigned to work server 140.163.4.245
01:38:16:WU01:FS01:Requesting new work unit for slot 01: READY gpu:0:GP102 [GeForce GTX 1080 Ti] 11380 from 140.163.4.245
01:38:16:WU01:FS01:Connecting to 140.163.4.245:8080
01:38:17:WU01:FS01:Downloading 14.50MiB
01:38:23:WU01:FS01:Download 33.19%
01:38:29:WU01:FS01:Download 59.48%
01:38:35:WU01:FS01:Download 81.46%
01:38:38:WU01:FS01:Download complete
01:38:38:WU01:FS01:Received Unit: id:01 state:DOWNLOAD error:NO_ERROR project:10496 run:104 clone:74 gen:13 core:0x21 unit:0x000000178ca304f5588972ba200e8030
01:38:38:WU01:FS01:Starting
01:38:38:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:38:38:WU01:FS01:Starting
01:38:38:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually
01:39:38:WU01:FS01:Starting
01:39:38:ERROR:WU01:FS01:Failed to start core: OpenCL device matching slot 1 not found, try setting 'opencl-index' manually

Posted: **Tue Jun 20, 2017 2:27 am**

Ok now it's running, but...
I removed the single GPU and rebooted and got same thing. I then tried setting gpu-index=0 and opencl-index=0 from both being "-1" and it works. Now I have no faith that when I add the next 3 GPUs this thing is going to behave.

Posted: **Tue Jun 20, 2017 2:59 am**

The beta client does not work well with multi gpus. You have to reset each slot after reboot or shutdown. I am running 2 6 core xeons and to gtx1070s on Mint 18.1. So I have 10 cores available for folding. I never have a problem with getting work. Also, I am running the 7.4.16 client. I would suggest you go back to 7.4.16 and assign 2 10 core cpu clients. This should keep all 4 gpus and remaining cpu cores at 100%. My system runs 24/7 with few issues. Nvidia cards usually get 10% better production on Linux.

Posted: **Tue Jun 20, 2017 3:53 am**

Aurum wrote:Ok now it's running, but...
I removed the single GPU and rebooted and got same thing. I then tried setting gpu-index=0 and opencl-index=0 from both being "-1" and it works. Now I have no faith that when I add the next 3 GPUs this thing is going to behave.

FAHClient has always had problems with multiple GPUs. I've written several tickets and with each successive beta, the problem MOVES (some things are fixed and others are broken.

The last time I move GPUs around, I was running 7.4.16 and I got the same message .... try setting 'opencl-index' manually ... which I did. Maybe it's a coincidence, but I ended up just like you, with opencl-index needing to be set to 0 for the remaining GPU. I have rebooted and that setting continues to work.

Somebody recommended that if you're going to remove a GPU, you should removed the slot before shutting down to change the hardware.

To minimize the trial-and-error, I've discovered the following:

Code: Select all

Let's say you have 3 GPU, two of which are working and one which is producing that error message.

The two that are running will have a long message containing
Running FahCore: ... Core_21.fah/FahCore_21.exe ... -opencl-device 1 ....

The two that are working will have two numbers from the set {0, 1, 2} and the other GPU needs the missing number.

Posted: **Mon Jun 26, 2017 6:16 am**

Mint vs Ubuntu isn't going to be a big difference.
They are both Debian offshoots and pretty close.

Ubuntu is one of the "officially supported" distributions for almost anything that runs on LINUX - but as a general rule anything that runs on Ubuntu also runs on Debian or Mint without issues.

Posted: **Mon Jun 26, 2017 12:49 pm**

What works for me is to set the gpu-index and opencl-index both to the number of the gpu. That is 0 and 0, 1 and 1, etc.

Posted: **Mon Jun 26, 2017 10:32 pm**

I finally got my 1080 and 970 folding together. It was a long and not very pretty struggle. Some of the problems were unrelated to Folding, and some were user error. I still have a hardware issue related to my cpu that I need to deal with, and some minor errors with the desktop display. Temps on the 1080 are holding at 55 C. The 970 is running around 70 C with 55% fan. That's a bit warmer than before and is probaly related to it's new location in the case. I have some kind of heat issue on the CPU that's been developing over the past month or so, and it won't run with acceptable temps while doing CPU folding any more. Just serving the video cards, it's running around 68 - 70 C. That's the remaining hardware problem that I have to resolve. I'll have to deal with that later.

I'm very excited to see the PPD hovering around 1.3 million.

I turned 68 a couple weeks ago. My goal is to join the billionaire's club by the time I'm 70. I'm 1/4th of the way there after 10 years of folding. With this new equipment, I think I have a shot at it.

Folding Forum

Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint

Re: Adding a 1080 Card to Linux Mint