Quick Reference
One-page cheat sheet for the CMP 170HX lab desk. For narrative context see Introduction and Specifications.
PCI IDs and unlock targets
| Stock label |
Device ID |
Subsystem |
Unlocked HBM |
| 8 GB |
0x20C2 (10de:20c2) |
0x1585 |
64 GB (65536 MiB) |
| 10 GB |
0x2082 (10de:2082) |
0x1557 |
40 GB (40960 MiB) |
Compute (Full Specs, both SKUs): 70 SM / 4480 CUDA / 280 tensor. See Specifications.
Power and cooling limits
| Item |
Value |
| Default TDP |
250 W |
| Software ceiling |
~300 W |
| Idle GPU |
~30–40 W |
| Connector |
1× 8-pin CPU/EPS-style via adapter |
| Thermal hard stop |
~80 °C until the loop/air path is proven |
| Dry-run without coolant |
Power off within ~5 minutes |
PCIe layers
| Layer |
Stock |
After fix |
| Width |
Gen1 x4 (~1 GB/s) |
Caps: Gen1 x16 (~4 GB/s) |
| Generation |
Firmware Gen1 |
Software Gen2 (5.0 GT/s) via cmpunlocker |
| Caps parts |
(n/a) |
24× 0402 0.22 µF (ex. Samsung CL05B224KO5NNNC) |
Driver and unlock
| Item |
Value |
| Driver pin |
nvidia-open 610.43.02 or 610.43.03 |
| Tool |
amoghmunikote/cmpunlocker |
| Profiles |
--profile=8gb → 64 GB; --profile=10gb → 40 GB |
| Host firmware |
Above 4G on, Secure Boot off |
| BAR1 stock |
64 MB. A larger BAR1 is a separate kernel fixup, not the framebuffer unlock |
Key commands
# Identity
lspci -nn | grep -i nvidia
lspci -s <bus:dev.fn> -vvv | grep -E 'LnkCap|LnkSta|Subsystem'
# Memory / thermals / power
nvidia-smi
nvidia-smi -q -d TEMPERATURE,POWER,MEMORY
# Which nvidia.ko won?
modinfo -n nvidia
# Link after unlock / capacitor mod
sudo lspci -s <bus:dev.fn> -vvv | grep LnkSta
| Stage |
Memory |
LnkSta (typical) |
| Stock |
8192 or 10240 MiB |
2.5 GT/s, x4 |
| Unlocked, no caps |
65536 or 40960 MiB |
5.0 GT/s (if Gen2 in build), x4 |
| Caps only |
stock MiB until unlock |
2.5 GT/s, x16 |
| Unlocked + caps |
full unlocked MiB |
5.0 GT/s, x16 when both stick |
Related pages
References