4/3/2026 at 10:22:48 AM
Notable parts:- "GPU users should understand that the only cards known to be vulnerable to Rowhammer are the RTX 3060 and RTX 6000 from the Ampere generation"
- mitigations are enabling ECC on the GPU or enabling IOMMU in BIOS
So doesn't sound like a big deal for users, this is more of a datacenter sort of vulnerability. The fact that this attack is possible at all (you can turn small GPU memory writes into access to CPU memory) is pretty shocking to me, though.
by stratos123
4/3/2026 at 10:31:52 AM
Those are the cards that have been tested.It is very likely that the attacks work on most or all consumer Ampere cards, depending on what kinds of GDDR memories they are using. They might also work on more recent GPUs.
However, it is true that such attacks are normally useful only on multi-user machines.
The most important thing is that the attacks are prevented by enabling the IOMMU in the BIOS. This is a setting that should always be enabled, because it prevents not only malicious attacks, but also memory corruption due to bugs.
Unfortunately, many BIOSes have the IOMMU disabled by default, for fear of creating problems for some legacy operating systems or applications.
by adrian_b
4/3/2026 at 12:19:08 PM
most are GDDR5 and 6by v3ss0n
4/3/2026 at 8:16:05 PM
[dead]by okspAQ
4/3/2026 at 11:18:32 AM
Datacenters tend to have IOMMU turned on. Consumer devices are the ones that don't turn this on by default.by pclmulqdq
4/3/2026 at 8:36:46 PM
I think a new attack, GPUBreach, was also disclosed today that works even with IOMMU turned on. So thats not sufficient to defeat these attacks. Check out the updates to the article.by towardsproject
4/3/2026 at 6:31:22 PM
[dead]by okspAQ
4/3/2026 at 11:38:13 AM
> So doesn't sound like a big deal for users, this is more of a datacenter sort of vulnerability.If I understand it correctly though this can be used for priviledge escalation though, since it allows access to arbitrary memory.
by SkiFire13
4/3/2026 at 6:49:58 PM
I believe RTX 3060 is the most common card for people who want to have local LLM in their homelab.by integralid
4/3/2026 at 8:10:38 PM
Wait.. so just about every passed through GPU ( from that short lsit ) is fine?by iugtmkbdfil834
4/3/2026 at 7:13:41 PM
Ah so ars is still a pile of conde naste shit gotchaby halJordan