Asus WS X570-ACE MB - Is pairing this board with 3 2080 ti a good idea?

I want to do GPU rendering better and am hoping to upgrade soon. My goal is to put most of the money into the GPUs instead of the cpu and MB.

So my questions are:

1) How will I know if the 3rd PCI-E slot will deliverers enough power for the 3rd GPU?

2) Since the 3rd PCI-E slot works at PCI-E 4.0 8x then goes thru the chip set at PCI-E 4.0 4x will the data speed be caped/coverted to the 4.0 4x speed? (Ergo PCI-E 4.0 8x thru PCI-E 4.0 4x = PCI-E 3.0 8x)

Thanks for the help!

I do not think that any current gen GPU/card supports PCI 4.0. So you will need at least adjust your calculations for that.

coming soon TM next gen graphics cards will ( rumor mill ) and after initial launch 6 months/give or take a “3080 TI” will launch what the number scheme is does not matter the principal behind the TI cards is what it is.

1 Like

AMD’s Navi GPUs all support PCIe 4.0, but they aren’t as strong as the 2080ti. Maybe the Radeon 7 does also, it’s about RTX 2080-strength. That having been said, pcie 3.0 x8 is not a bottleneck in most workloads.

The third slot is a chipset slot, shares bandwidth with the m.2 SSD, the USB ports, etc. So that depends on just how busy the rest of the system is for the third slot. The other two slots are CPU slots. If you need more than that, you should get Threadripper or HEDT.

1 Like

Why do you need more than one GPU, it doesn’t make sense to me unless you plan to set up a virtual machine server, If that is the case then you don’t want a gaming GPU?

Blender or something else?

1 Like

The PCIe 4.0 slot is backwards compatible. Since the slot is working in 8x, I was wondering if than means the slot behaves as an 8x width PCIe slot, which would be great for me.

If that card is 70% fast than a 2080ti like the rumors say, that card might be worth the wait.

So you’re saying the PCIe 4.0 8x slot will be limited to a PCIe 8x width? If so that would be perfect for GPU rendering.

Also, HEDT is something I might have too consider. Maybe Threadripper second gen, because 3rd is too pricey. I just worry about those early motherboards giving me bugs or problems down the line. Also, I’m hoping to work in linux because I’ve gotten fed up with Win 10.

Yes, mostly it will be Blender with an OpenCL renderer and maybe a little Maya too.

It’s not for gaming, it’s for gpu rendering and some systems can have up to 16 gpus to render on. (not me of course)

No, PCie 4.0 x8 is PCIe 3.0 x16, but usually, in a multi-gpu configuration, that lane count goes down to PCIe 3.0 x8 on a mainstream non/HEDT/Threadripper motherboard. Keep in mind, that’s for a multi-GPU motherboard, usually one compatible with SLI, others may have the second GPU slot at pcie 3.0 x4, and the chipset slot is even worse, and a cheaper motherboard might only have that. (My B450 motherboard has one PCIe 3 x16 slot, one pcie 2 x4 slot, and several pcie 3.0 x1 slots…)

The ASUS WS though is about the best choice for a 3 GPU configuration on AM4.

To make things even more complicated, some motherboards have a bifucation option where some slots have less lanes in order to have other slots have more, with a different BIOS configuration.

1 Like

The WS is about the best 'board for a 3 GPU configuration on AM4 though, to make this a bit less confusing. All GPU slots will operate at PCIe 3.0 x8 in a three GPU configuration, in a one-GPU configuration, I think one slot will be pcie 3.0 x16. (PCIe 4 x8.)

1 Like

Keep in mind the chipset slot will be shared bandwidth with the m.2 drive, the USB ports, and so on, so its real-world performance may be less than the other two slots. If you can’t do that, you’d need HEDT or Threadripper.

1 Like

" Technology no place for wimps" - Scott Adams, Dilbert quote.

I get your point on the confusion. Lol, I thought it was just portions of the PCIe slot’s teeth would be disable to make it 4x, 8x, etc…

I feel it makes more sense to go with the HEDT route Now. Since this will be my daily driver too, it might not be wise to have the PCIe lanes maxed out and possible limiting performance. I guess it’s cutting it a little to close to be cheap lol!

Thanks PloniAlmoni!

1 Like

Well if you want to use three 2080ti´s,
then HEDT is pretty much the only viable option out there.
Main stream simply lacks the pci-e lanes pretty much.
And when you spend that much amount on render system,
the last thing you actually want to is to create a bottleneck.
Because you also wanne use fast as possible nvme storrage right?

It kinda depending on the budget.
Intel x299 with a core i9 cpu, or a AMD Threadripper setup.

1 Like

It isn’t so much that the CPU-powered slots are at pcie 3 8x, that’s not much of a bottleneck. It’s the chipset slot, if you’re using a fast nvme drive it shares resources with, that may present a problem even though it’s x8 on the WS. Enjoy your HEDT/Threadripper system. :grinning:

1 Like

Some clarifications:

M.2_1 is using 4 PCIe lanes supplied by the CPU - not sharing with any other device

When using PCIe 3.0 GPUs you can operate all three of them without any bottleneck at 3.0 x8 speed.

The chipset’s interface to the CPU is PCIe 4.0 x4, speed-wise essentially PCIe 3.0 x8 so that GPU doesn’t get slowed down compared to the other two that are handled by dedicated CPU PCIe lanes.

Only M.2_2 is sharing bandwith with the chipset but in any case it only has a x2 connection meaning it’s always bottlenecked when using a fast NVMe drive.

I’d be a bit worried about the power delivery to the board’s PCIe slots (3 x 75 W would be in spec) since it doesn’t have additional power connectors like other motherboards.

But a very nice board for home server stuff where you don’t have to go Threadripper/Epyc due to being able to use 3 x8 slots :slight_smile:


Hmm, I stand corrected about the chipset GPU slot then.


I just know about this board since I’m currently experimenting with it for a home server build with an HBA, ethernet controller and a couple of NVMe drives that would otherwise have made it necessary to leave AM4 for HEDT if the WS Pro X570-Ace didn’t exist (a route I didn’t like since it would have been wasteful (only needed about 30 PCIe 3.0 lanes)).