Nvidia GPU downgraded to PCIe Gen1 speed on ALTRAD8UD-1L2T

General Discussion
1

Hello !

I recently started to plug an Nvidia GPU on my ALTRAD8UD-1L2T system and I noticed that the GPU is stuck at PCIe Gen1 speed.

Did I missed something ? Does it ring a bell to you ?

My troubleshooting command:

mapfile -t PCI_IDs < <(lspci | grep -Ei 'samsung|nvidia|micron' | cut -f 1 -d ' ')
declare -p PCI_IDs
for pci_id in "${PCI_IDs[@]}"; do
  sudo lspci -vv -s "$pci_id" | grep -E '(^[0-9a-f]+:)|LnkSta:'
done

The output:

declare -a PCI_IDs=([0]="0000:01:00.0" [1]="0001:01:00.0" [2]="0003:04:00.0" [3]="0004:01:00.0" [4]="0004:02:00.0" [5]="000c:04:00.0" [6]="000d:01:00.0" [7]="000d:03:00.0")
0000:01:00.0 3D controller: NVIDIA Corporation AD104GL [L4] (rev a1)
                LnkSta: Speed 2.5GT/s (downgraded), Width x16
0001:01:00.0 Non-Volatile memory controller: Samsung Electronics Co Ltd NVMe SSD Controller PM173X (prog-if 02 [NVM Express])
                LnkSta: Speed 16GT/s, Width x8
0003:04:00.0 Non-Volatile memory controller: Samsung Electronics Co Ltd NVMe SSD Controller PM9A1/PM9A3/980PRO (prog-if 02 [NVM Express])
                LnkSta: Speed 16GT/s, Width x4
0004:01:00.0 Non-Volatile memory controller: Samsung Electronics Co Ltd NVMe SSD Controller PM173X (prog-if 02 [NVM Express])
                LnkSta: Speed 16GT/s, Width x8
0004:02:00.0 Non-Volatile memory controller: Samsung Electronics Co Ltd NVMe SSD Controller PM173X (prog-if 02 [NVM Express])
                LnkSta: Speed 16GT/s, Width x8
000c:04:00.0 Non-Volatile memory controller: Micron Technology Inc 7500 PRO NVMe SSD (rev 01) (prog-if 02 [NVM Express])
                LnkSta: Speed 16GT/s, Width x4
000d:01:00.0 Non-Volatile memory controller: Micron Technology Inc 7500 PRO NVMe SSD (rev 01) (prog-if 02 [NVM Express])
                LnkSta: Speed 16GT/s, Width x4
000d:03:00.0 Non-Volatile memory controller: Micron Technology Inc 7500 PRO NVMe SSD (rev 01) (prog-if 02 [NVM Express])
                LnkSta: Speed 16GT/s, Width x4
2

Give this card some serious use and check while it is busy.

When I had NVIdia card in my x86-64 desktop it was changing amount of lanes and speed during idle.

1 Like
3

This is correct. After loading a model in the GPU, the PCIe link has been upgraded to PCIe 4.0.

0000:01:00.0 3D controller: NVIDIA Corporation AD104GL [L4] (rev a1)
                LnkSta: Speed 16GT/s, Width x16
1 Like
4

You have no problems. Looks like PCIe has bugs.

5

This is from my Mt Snow, many U2 PCIe 3.0 x4 slots only run with downgraded speed.

image
image1630×704 233 KB

1 Like
6

You can also lock the GPU and memory clocks to prevent the PCIe link being changed.

sudo nvidia-smi -lgc 10000
sudo nvidia-smi -lmc 10000
1 Like