Voltage glitching turns an 8GB mining card into an 80GB A100
Using voltage glitching to unleash the A100 in my mining card
A voltage fault injection attack on the GA100 die lets an operator blow arbitrary fuses, turning a CMP 170HX mining card into a functional A100-80G. The unlocked card runs about 1.2x faster than a retail A100 because all ~120-126 SMs are enabled instead of the yield-safe 108. NVLink, PCIe Gen4, NVJPG, NVDEC, and MIG can also be unlocked. The vulnerability was disclosed to NVIDIA.
In a properly secured design, a blown fuse should always induce a more restrictive state, because a blown fuse cannot be unblown in an economically feasible way. NVIDIA’s design allows one to relax restrictions by blowing more fuses!
- htrp
amazing to think there was a time (literally not too long ago) where you would throw 80 GB of HBM onto a chip and turn 90% of it off
- myself248
Instead of putting the host to ACPI S3 state, could an external PCIe riser be used with a separate power supply?
Seems like that would considerably speed up the process, by letting the host continue to run while only the target cycles.
- notenchange
This is great. I’m actually trying to unlock the 80G in a CMP 170HX the software way for the last 2 weeks (still no luck).
I just don’t know if I’m brave enough (and knowledgeable enough) to go this blown fuses way. Seems a lot more risky.
P.S. I now see this is for the 8GB model, which is insane. The 10G one is the one that until now was proven to have 80G, but never fully accessible through CUDA (it was accessible using CPU memory copy though - much slower).
- amelius
But what if nvidia used those fuses because an A100 had too many fabrication faults so it could only be a cheaper mining card?
(which is the more likely scenario because they can make much more money on enterprise/datacenter products, see DRAM)
- roger_
Glad I jumped on these cards, excited to give this a try!
Hard to believe $1k was a deal a few months ago (and $200 before then).
- SillyUsername
I just wish they had something like this for older Tesla p100s, p4s...
- nullc
Considering how unobtainium GPUs have become nvidia could get a PR win by publishing unlocks for all their outdated hardware. ... people are figuring it out anyways, might as well embrace it.
- dylanger
Great usage of glitching!