Folding Forum

Hi all,

Currently planning a new 4x 1080ti build and was hopping for some advice.

The cheapest 40 lane cpu I can find in the UK is an E5-2603V3 at £215 https://www.scan.co.uk/products/intel-x ... cpu-retail

Couple of questions about this

1 - Will this cpu bottleneck 4 1080tis? I know you don't need a crazy cpu to feed a gpu but 1.6ghz is a little low.

2 - Can you put a xeon into a normal x99 gaming board? This is the cheapest board with 4 double spaced pcie slots I can find: https://www.scan.co.uk/products/gigabyt ... e-m2-(pcie)-4-way-sli-4-way-crossfire-usb-31-type-a

The alternative could be to get a 6850k but this is double the price of the xeon, which whilst a small number in the context of the whole build, I don't want to spend £200 if its not necessary

Thanks in advance
Gary

Your motherboard link did not work for me. I think this is the same MB.
https://www.scan.co.uk/products/gigabyt ... -31-type-a

The vendors web site lists E5-2603v3 compatibility
https://www.gigabyte.com/Motherboard/GA ... -rev-10#sp
https://www.gigabyte.com/Motherboard/GA ... upport-cpu

This motherboard will work.

Nvidia uses an infinite loop to look for PCI-E activity, you can't use faster CPUs to improve an infinite loop, 1.6 Ghz will be fine.

I am running 2 of those xeons in a 2p motherboard. Asrock Rack EP2C612WS. It has 7 PCIe slots and can easily fit 4 1080tis with 80 lanes. You can start with one cpu and as needed add in 2nd cpu. Motherboard is not much more expensive than the gigabyte board.

des1957 wrote:I am running 2 of those xeons in a 2p motherboard. Asrock Rack EP2C612WS. It has 7 PCIe slots and can easily fit 4 1080tis with 80 lanes. You can start with one cpu and as needed add in 2nd cpu. Motherboard is not much more expensive than the gigabyte board.

Nice MB. Its manual has a block diagram on page 14 and a table on page 23 showing you get 2 @ x16 plus 5 @ x8. What do you get in actual practice What's the rig PPD
http://asrock.pc.cdn.bitgravity.com/Man ... 2%20WS.pdf

Right now, running both 1 cpu client with 10 cores and 2 gtx1070s overclocked. Averaging about 1.4 m per day. Waiting on Vega release, to see if I go with Vega or 1080ti. Whole system is watercooled. Cpu and gpus average right 50c.

Thanks for all the replies - really helpful.

I've not heard of infinite loops and cpu clock speed not effecting gpu performance but this is a bit of a game changer. I guess this means that as long as you leave 1 core free per GPU then your GPU folding performance should not be effected by whatever else the cpu is doing - correct?

I read that CPU folding can adversely effect GPU folding and that in order to maximise PPD you should leave at least 2 cores free per GPU or better yet avoid cpu folding altogether. If this is untrue then I will be delighted to resume CPU folding once more!

Have you confirmed that 4 1080 TIs will fit on double spaced pcie slots? These cards are a lot thicker than the 1080s are. I'm using an open air wooden case so it's not an issue for me. Also be aware that each requires two 8 pin power cables so make sure your PSU has enough.

SteveWillis wrote:Have you confirmed that 4 1080 TIs will fit on double spaced pcie slots? These cards are a lot thicker than the 1080s are. I'm using an open air wooden case so it's not an issue for me. Also be aware that each requires two 8 pin power cables so make sure your PSU has enough.

They will "fit" side-by-side; however, given the lack of air flow, I wouldn't recommend it for what we do.

I've already got a rig with 4 blower style 1080tis with that board. They fit fine and the temps aren't too bad as I run at 100% fan speed. As you'd imagine it's LOUD so I run it in my garage.

I was planning to watercool it but I'm a bit nervous of the cost/complexity/maintenance, so it may just stay in the garage.

The plan for the new rig is to get 4 EVGA hybrid cards and run it in my office. Should hopefully mean the system is quieter than blower style!

As for PSU I'm using an EVGA G2 1600W. That has plenty of 8 pin power cables and plan to get exactly the same in the new system

I'm using that same PSU. It has plenty of power but iirc I am using every cable that came with it. Otherwise I might put something on the two 1X pcie slots I have left. I think next build I'll go with two smaller PSUs for that reason.

ASUS was showing off a custom-build 8x1080ti 4u rack build recently - all blower-type cards but the article didn't specify if it was their standard "Turbo" blower card (was definitely NOT an FE but it did look like the Turbo).
They claimed the cards ran at some absurdly low temp (53C or some such) while doing some sort of work on them but even with massive airflow I take THAT claim with a couple pounds of salt.

The cards didn't appear to have any more space between them than standard "every other slot" spacing on something like the Biostar 4 x PCI-E 16x motherboards have, but the case DID appear to have some serious fans in the front.

QuintLeo wrote: ... while doing some sort of work on them ...

In the non-FAH world of GPUs, the current rage seems to be AI. Does anybody know how the heat generated by AI compares to the heat generated by FAH?

I guess it is similar as the GPUs also run several days with full load until some deep learning training task is finished.
The server racks fans run at high speed with much noise pushing a big amount of 20°C air through the case from an air conditioner.
That is much more cooling than you would get in a normal PC case or an open case.
(The wind noise is why we call our server room bug-proof)

foldy wrote:(The wind noise is why we call our server room bug-proof)

It's also great for us tinnitus sufferers

Aurum wrote:It's also great for us tinnitus sufferers

Because of my inner ear issues, I don't hear hardly any of our fan noise.