Asus Proart X870e - 3 slot GPU is blocking the other PCIE slots?

Hi,

I bought a Asus Proart X870e, the space between the gpu slot and the slot below looks to be 3 slots.

I am looking to get a Gigabyte RTX5090 Windforce OC which is mentioned to be 65mm in height. is that 3.25 solts? sounds odd…
I have read on Google that a PCIe bracket/slot has 20mm height.

Is that going to block the pcie below the GPU?

I have not built a PC since the RTX2000 series, but sounds stupid to block the below pcie slot for 5mm… and limit any future expansion…

Please help, maybe I am overthinking it and will not be an issue in reality…

are all the pcie slots usable only with the 5090 FE because of the 2 slot design?

Please help, maybe I am overthinking it and will not be an issue in reality…

PS: my issues with 5090FE are:
Firstly, not sure if I can get one.
Secondly, I went for a Noctua NH-D15 G2 (for reliability and not stress about liquid) and an amd 9950x in a FD Torrent case. With the FE design, all the hot air will blow into the cpu cooler.
Thirdly, looks like the FE runs a little too hot for my taste… VRAM in the 90s C.

Looks like a 3.5-slot card, so yes, your second slot will be blocked on the ProArt. There are some AM5 boards with more room between the top two slots but the second slot might not be Gen5 x8 from the CPU like on the ProArt. You may be able to sneak a riser under the GPU and mount whatever you want to put in the second slot vertically. Or you could use a riser for the GPU and mount it vertically but that opens up a few other cans of worms.

Your best options are to find either another board or a smaller 5090. (Or live with only one expansion slot from the CPU.)

ETA: Check out the Asrock X870E Taichi (Lite). That appears to be the only AM5 board with 4-slot spacing and Gen5 x8/x8 slots from the CPU.

ETA 2: OK, and the Gigabyte X870E AORUS XTREME AI TOP, but the price is eye-watering.

Hey, have the same board.

Not sure the 3.5 slots matter that much? The chipset does sadly have a very stupid way to distribute its PCIe lanes. I.e. as soon as you put a device into the 2nd slot, it reduces the number of lanes afforded to the 1st slot to 8.

As @bambinone pointed out, a 3.3 slot GPU in in slot 2 mechanically obstructs slots 2, 3, 4, and 5. The ~15 mm the Windforce leaves open in slot isn’t all that much for air intake, particularly at 5090 power, so leaving slots 6 and 7 open as well isn’t a bad idea if operating temperatures and noise are concerns. Similar considerations apply for two slot GPUs with another card in slot 5.

Depends on the cards. A second GPU’s both pretty obstructing and contributes its exhaust heat to the downwind GPU, more directly so with passthrough. If it’s a four port ASM1164 SATA card it’s not very obstructing and only runs a few watts.

Happens to some extent with all GPUs and airflow configs where the CPU’s not on intake rads only. Haven’t seen data yet but the FE’s dual passthrough instead of pushing exhaust down into the motherboard’s may be a net win even if partially blocked by the G2. And with single passthrough there’s still backplate heat anyways.

The Torrent-G2 combination doesn’t help, what with putting the cooler up against the GPU, lack of top intake, and lack of clearance between the G2 and PSU intake. Speaking from putting 140 mms in Torrent Compacts, Noctua’s 150 mm compatibility rating there’s irrationally exuberant. The full size at least gives a couple hex punches more PSU clearance.

1 Like

Those lanes are not related to the chipset, but rather those come straight from the CPU.

The first slot drops to 8 lanes when the 2nd one is in use because the CPU only has 24 lanes available for general purpose, with 2 x4 of those being for NVMes, so only 16 lanes left in total.

This is standard behavior for a CPU’s PCI Express Graphics (PEG) lanes on enthusiast motherboards and has been for many generations, going back to the first SLI/CrossFire chipsets. And of course HEDT/prosumer hardware has had all kinds of PCIe lane bifurcation options for ages. Unfortunately in today’s PCIe Gen5 world this feature is pretty much only found on $400+ models. It was more prevalent around the $200 price point a couple years ago. It’s much cheaper for mobo makers to hang additional Gen3 or Gen4 expansion slots off the chipset instead.

So even though SLI is mostly dead, this feature is still useful to those of us who want to run more than one expansion card directly attached to the CPU for dedicated PCIe bandwidth, lower latency, fewer compatibility issues (i.e. no chipset quirks), and finer-grained IOMMU groupings. This could be a second GPU (e.g for a VM), additional NVMe drive(s) that won’t bottleneck on the chipset, a DPU, or anything else really.

Don’t disagree that there are many valid use cases for 2 GPUs on a board. Have mostly used server hardware in the past couple of years, where you’d do multi-GPU AI (inference) on these cards.

My whole point here is, what valid use cases for multi-GPU are there these days? Mostly professional applications. But for those you’ll pretty much need the bandwidth into the GPUs…

Thanks for your reply. I went with the proart because it has 3 pcie connectors which i find them more usfeul than a ton of M.2s

Gigabyte X870E AORUS XTREME AI TOP is absurd… i though the pro art is expensive, but in my country the Gigabyte X870E AORUS XTREME AI TOP is 800USD… not paying that…

I will not go a 2 GPU setup, but I wanted to be able to put a dual 25G NIC in the future and from what I have seen those are G4 x8 and also add a capture card which is G4x4. The other “non GPU” slot is a G4x4.

Also, the riser option might not be a solution because the Torrent does not have a vertical GPU mounting option…

TBH, looks like my only option is a RTX5090 FE or a liquid cooled GPU… I did not go with an AIO because of reliablity and I do not see myself going with a liquid cooled GPU. Also, the FE is impossible to get in my country (Romainia, EU)… Fun times…

I do run 2x 3090s on a x8/x8 setup, mostly for ML workloads, and with only a pair of GPUs the bandwidth from x8/x8 is more than fine.

Past 4 GPUs you might start to notice bottlenecks, but at this point a consumer platform wouldn’t have enough slots anyway.

Other tasks may be more bandwidth-hungry still.