Hey all. Can you give me a sanity check?
I’m planning to use RTX Pro 6000 in SFF build and with a riser cable to run AI workloads locally. Since 6000 already using risers internally should I be concerned about signal integrity and stability?
Are there any test I can run to verify that?
Make sure to use a pcie5 riser.
On Linux you can check for PCIe errors in the logs (perhaps you need to enable AER in bios), and check that the link speed is not downgraded using lspci.
If there are issues you can downgrade to PCIe4 in the bios, which shouldn’t cost much performance (as long as the GPU doesn’t need much access to system RAM).
Hard to predict if it will work but at least there’s tools to diagnose and mitigate.
This is a mixed bag, and not all risers are built the same. I currently have several of these cards on risers varying from 100mm to 300mm with mixed results, all of them are “PCIe 5” … but they do not act the same. While I haven’t had any issues with functionality, I do have a mix of messages depending on the riser and the slot the card is in. That said, when I put all of them in Gen4 mode in the BIOS, they work flawlessly with no messages. My next steps are to figure out which brands / lengths work best without errors. Once I have that data I will share it.
I’m running it on one of these from Amazon (30cm straight) and it’s working just fine for me:
LINKUP-Future-Vertical-Compatible-Straight/dp/B0D5F72XTL
Working perfectly fine for me on my ASRock WRX90 Evo. From what I understand it could be very much dependent on the motherboard itself as well. I wanted the ASUS WRX90 SAGE honestly, but after hearing the horror stories between getting stable RAM running and full utilization of PCIe slots I opted for the ASRock. Couldn’t be happier/zero issues so far. 