Second GPU query

[wkdwill]

Hi all,

A question to the wise,

I have an Asus Prime Z270-P MB and the specs state the following for PCIe slots.

2 x M.2, 2242/2260/2280, PCIe/SATA
4 x PCIe 3.0 x1
1 x PCIe 3.0 x16
1 x PCIe 3.0 x16 (x4 Bandwidth)

My question relates to the second x16 slot,

As this slot is only x4 Bandwidth, is it worth me getting a decent second graphics card? or will the x4 limit it severely?

This second card would be used solely for crunching, so I am not bothered about gaming performance from it, maybe get a better card for gaming to put in the full bandwidth slot and move the GTX1070 into the x4 for crunching.

Thanks,

Will

[UBT - Mikee]

Hi

Nice m/board - same as mine!

I'm running two 2 GTX 1060's and they both play nicely. They are different makes so they run slightly different - if I had both the same I could steer you better. However, I have no problems with them.

Mike

[chriscambridge]

What I found is on BOINC (rather than Folding@home), on the GPU projects I ran was that you were fine even on x1 risers (same as a x1 slot) as long as you were on Gen3. If you ran on Gen2 then you took a small hit on work done.

I was running GTX 1080's on x1 slots (risers), so because you are running at x4, you should be fine and not really notice any difference.

Normally if you go into the BIOS when you have GPUs connected it will show you what Gen the PCIE slot is running at; also on some boards, you can choose an option to run at Gen3 if it happens to be running at Gen2.

I agree on the Mobo, we run some Asus and these some of the best mobos I have used; things like the q-connector, the LCD status LED, on/off button on the mobo, were very useful.

I had a quick look in the manual and I think it must be using some kind of PCIE switcher as it says that if you run 2 GPUs it will run at x16 and x4 (which adds up to 20).

You can get the manual here:
https://www.asus.com/uk/Motherboards/PR ... sk_Manual/

[wkdwill]

Thanks guys,

Think I’ll put the 1070 in the x4 slot and get a 1080 or 1080ti for the gaming, need to upgrade the PSU first though.

Thanks again,

Will

[wkdwill]

Decided to go with a 1080ti and a nice 1000W PSU, with the power I'll be using it'll be right in the sweet spot for efficiency
both due to arrive Monday, so fingers crossed they arrive and I can get them installed that evening, should increase my output nicely

Will

[chriscambridge]

The 1080Ti's do wonders on PrimeGrid and GPUGrid; on the other projects I would take a look at the GPU stats, as often they just cannot be utilised by some projects, and therefore you would probably do better to run the 1070 or a 1080 on them.

This is an example on Asteroids:

http://asteroidsathome.net/boinc/gpu_list.php

This is for Windows GPUs; as you can see you would probably be better running a 1070 on this project instead of a 1080 or 1080Ti, as it does not really do that much more..

(1.000) GeForce GTX 1080 Ti
(0.947) GeForce GTX 980 Ti
(0.849) GeForce GTX 1080
(0.838) GeForce GTX 1070
(0.827) GeForce GTX 980
(0.670) GeForce GTX 1060 3GB
(0.667) GeForce GTX 970
(0.653) GeForce GTX 1060 6GB
(0.568) GeForce GTX 780
(0.422) GeForce GTX 960
(0.353) GeForce GTX 1050 Ti
(0.308) GeForce GTX 680
(0.263) GeForce GTX 750 Ti
(0.230) GeForce GTX 950M
(0.225) GeForce GTX 750

What Ti did you go for in the end? I have a Gigabyte Gaming OC and a MSI GamingX; the Gigabyte is slight better than the MSI for crunching.

[wkdwill]

Hi Chris,

Thanks for the info

I went for an EVGA https://www.evga.com/products/product.a ... P4-6593-KR as I already have an EVGA card and would like to use the same control panel, the 1080ti will mostly be used for gaming and the 1070 for crunching, although whilst I'm not using the PC both will be crunching

As for what I can get out of it, I'm not sure, my 1070 is listed as 1784 boost clock but it is currently running at 1974.

Will

[chriscambridge]

I heard that the SC2 Gaming was a pretty Good Ti.

If you want to max out on credits, then run Collatz with the 1070 and 1080 Ti speed up code.

The 1080 Ti code is:

verbose=1
kernels/reduction=48
threads=9
lut_size=18
sieve_size=30
sleep=1
cache_sieve=1
reducecpu=0

You will have to ask, probably, Mark or Tim for the 1070 code.

On Windows you want to put the code in a file within the hidden folder ProgramData; in Windows 10 this is:

C:\ProgramData\BOINC\projects\boinc.thesonntags.com_collatz\

Other version of Windows no doubt you can work it out; if you are running Linux then I would ask Mark where this folder resides.

You want to put the speed up code in the config file named:

collatz_sieve_1.21_windows_x86_64__opencl_nvidia_gpu.config

(The above file name is for 64 bit WIndows)

The speed up code drops a 10 minute work unit on a 1080 to 6 minutes, and on a 1080 Ti to 4 minutes! I do not run 1070 so I am not sure of them but it should definitely reduce the runtime to completion.

admin edit: I've added the correct suffix to the end of the "config" filename above - as it's not "xml", like other BOINC config files.

[wkdwill]

Thanks for the code Chris,

I am already running the speed up code for my 1070, I cannot remember the WU time but I'll check when I turn on Collatz again, running 100% Moo at the moment.

Don't know if this is a stupid question, but I'll ask anyway.

Is there a way to use speed up code for both GPU's at the same time?

Just wondered as the code is different and there is only one file to put it in.

Thanks again,

Will

[UBT - Timbo]

Hi Will,

Woodles gave a "chapter and verse" on Collatz speed up code here, along with some examples:

viewtopic.php?f=93&t=6001#p111354

regards
Tim

[chriscambridge]

I did try looking for that link but could not find it, so thought I would repeat it.

Unfortunately not, it's just one speed up code per GPU model; Sorry that never crossed my mind.. That is why it is a bit of a pain having both 1080s and 1080 Tis, rather than one or the other.

[wkdwill]

Thanks for all the info guys,

I'll do a bit of playing around with the speed up code, see what I can get to work best.

Will

[Woodles]

Hi Will,

The only way that you can get two versions of the speed up code for different GPUs is to run two copies of Boinc with one GPU disabled in each (different ones of course)

Then you'll have two different data directories and two copies of the collatz_sieve_1.21_ ... .config file.

Either run one copy in a VM or run two independent copies of Boinc. Both possibly but not straightforward. Probably easier just to run with a compromised file for both of them.

Or I could swap you a 1070 for your 1080Ti then you wouldn't have a problem?

Mark

[chriscambridge]

that seems fair

[Woodles]

I'm prepared to make the sacrifice

[wkdwill]

Hi Mark,

Soooooo kind of you to offer the swap

but I must respectfully decline

I'm undecided which route to take, the two copies of Boinc sounds easier to me, I have played around with VM's in the past but always seemed to run into problems with them.

The hardware has arrived and is currently sat acclimatising, this is going to test my patience.

IMG_1306.png

Will

[wkdwill]

Got the 1080 up and running, nice improvement

moo.JPG

Will

[Woodles]

Hi Will,

Shame you feel that way about the swap

It's even better than you think

I was a little surprised to begin with as my dual 1070s finish a task in 620 - 650 seconds so I was expecting your 1080Ti/1070 hybrid to be quicker than that. Then I saw you have the same problem that Mikee had that the times reported aren't accurate.

Moo reckons it took 13:53 (or 833 seconds) for one of your tasks but the output file has a start time of 22:19:35 and an end time of 22:27:18 which I make 7:43 or (463 seconds)

So slightly less than half the time of your single 1070 which is more like it

Mark

[wkdwill]

Thanks Mark, that makes me feel even better about my purchase

A quick question if I may,
I have read that you can define which GPU's run which projects but I am not sure which card is which,
I see the devices listed as

GPU.JPG

How do I know which is device 0 and device 1?

Thanks,

Will

[Woodles]

Hi Will,

A very good purchase for crunching

In most cases, device 0 is the more capable slot and the one nearer to the CPU but since your GPUs are different you can easily get the information from the Boinc Manager event log.

When Boinc starts there should be a couple of lines near the top of the log saying something like

19/12/2017 00:27:51 | | CUDA: NVIDIA GPU 0: GeForce GTX 1070 (driver version 378.49, CUDA version 8.0, compute capability 6.1, 4096MB, 3046MB available, 6803 GFLOPS peak)
19/12/2017 00:27:51 | | CUDA: NVIDIA GPU 1: GeForce GTX 1070 (driver version 378.49, CUDA version 8.0, compute capability 6.1, 4096MB, 3046MB available, 6707 GFLOPS peak)

I don't have any mis-matched cards in any of my setups but you should be able to tie up the GPU 0 and GPU 1 with the card name (GTX 1080Ti or GTX 1070 in your case)

Mark

[Woodles]

To define which GPU is used for each project, you need to modify your cc_config.xml file:

Code: Select all

<cc_config>
  <options>
    <use_all_gpus>1</use_all_gpus>
    <exclude_gpu>
      <url>https://boinc.thesonntags.com/collatz/</url>
      <device_num>0</device_num>
    </exclude_gpu>
    <exclude_gpu>
      <url>http://moowrap.net/</url>
      <device_num>1</device_num>
    </exclude_gpu>
  </options>
</cc_config>

Don't forget that you're excluding a GPU from the project so that cc_config.xml will allow GPU 1 to run on Collatz and GPU 0 to run on Moo.

Mark

[chriscambridge]

I decided to join you Will..

[wkdwill]

Nice card

Don’t think that cc_config has worked, I applied it last night, checked that collatz was downloading ok, but on checking my stats today, no more credits on Moo, I’ll have to check over everything when I get home.

[Woodles]

Hi Will,

Did you have both GPUs in your system when you downloaded the Moo tasks? Some projects tailor their tasks to what's available when requested and you may find that any tasks already downloaded are marked as being for two GPUs and are waiting for the second GPU to become available .. ie, never.

Check in the Boinc manager view for the number of GPUs the tasks are expecting to be available.

Mark

[wkdwill]

Hi Mark,

Yeah both GPUs were in the system when I downloaded, I edited the cc_config then shut down Boinc, restarted Boinc and aborted the tasks that were marked as 2 x GPU, then I hit update on Moo and left it to it, didn't check before leaving the house this morning, just checked now and all Moo tasks are marked as 2xGPU.

Capture2.JPG

I've checked the logs and this is the entry that states project GPU exclusion.

Capture.JPG

So it seems Moo is ignoring the cc_config?

Thanks,

Will

[UBT - Mikee]

Hi

No - Moo will only run when the two cards are available. So if 1 collatz is running it won't run the Moo WUs as there is only one card available and Moo wants 2 free.

Mike

[wkdwill]

Hi Mike,

Don't think I'm understanding this properly,

My understanding is I can set the cc_config so I can have Moo should run on my 1080ti and Collatz on my 1070 as the 1070 has the speed up code,
Moo still wants 2xGPU but if I suspend Collatz Moo won't run as it wants both GPUs, even if I abort those WU's and download new ones it still wants both GPUs, so in that case is there any point in my editing the cc_config? am I asking for something that is not featured in Boinc software?

Cheers,

Will

[UBT - Mikee]

Hi

I did try the code myself earlier and had the same problem. The code would probably be ok for say Primegrid and Collatz, but as Moo sees and wants both your GPUs (uses 2 GPUs to process 1 Work unit) I can't see the config mod working.

Mike

[wkdwill]

Oh well back to the drawing board it is then.

Thanks for your help

Will