AMD silently removes memory encryption from consumer Ryzen CPUs, leaving users unaware that they may be vulnerable
https://www.tomshardware.com/pc-components/cpus/amd-silently-removes-memory-encryption-from-consumer-ryzen-cpus-leaving-users-unaware-that-they-may-be-vulnerable-security-feature-vanishes-after-newer-agesa-firmware-amd-engineers-go-radio-silent-when-pressed-about-the-changeOpen linkView original on mander.xyz485
Comments45
The headline is a little misleading: the feature has disappeared from consumer chips but AMD is not responding when asked why. As the article itself says: it's not clear if this is a deliberate decision, or a bug that has caused this issue.
The headline implies it was a deliberate action. Maybe it was, but at the moment we don't really know. But it is good that Toms Hardware is writing about this and drawing attention to this issue. It's concerning regardless of the reason, and it's also concerning how cagey AMD is being about addressing this issue.
A little more context as to when the engineer declined to continue the discussion:
Sounds to me like he had originally wanted to have it enabled for consumer CPUs, but some decision was later made to make this a feature only for higher end chips, even if lower end chips could technically support it. I can't really blame the engineer for wanting to stop the discussion at this point. He's most likely not the one making these decisions and the questions would be best asked to someone higher up.
I bet it's because something good!
It's funny how every big tech decision these last few years all sound like a shitty James Bond villain step in a shitty world domination plan, with shitty corpo writing.
👨🚀🔫👨🚀
Hold up, since when did consumer Ryzen CPUs have memory encryption support? I was sure that was always a EPYC exclusive feature.
I think that’s the crux of the article. The feature was there on some chips but not supported. A new update now prevents access to the feature.
Ryzen "Pro" too (their business line).
If it's deliberate and not put back, there's also the possibility the government made them remove it and not disclose why. So they can continue to access certain info and back doors and this security was giving them issues.
The government forced email providers to have a backdoor for them. It's the NSA's PRISM program. Been around since at least 2008.
YSK: This feature was disabled with a pushed firmware update.
Its true it was "not supported", but the CPU was/is capable of it.
The big issue here is did AMD disable it accidentally, or did they do it intentionally. If it was intentional why did they not announce it anywhere in the update notes, or anywhere else?
The article isnt very clear on this, but did they actually remove a critical feature from already sold products? Surely they can be sued for that?
Tom's is trash and should be banned. The original Ars article it mentions is better: https://arstechnica.com/security/2026/06/users-cry-foul-after-amd-stripped-memory-crypto-from-its-consumer-cpus/
Sounds like it was never really supported, but available. With the new BIOS update it's no longer available.
If that's the case, AMD shouldn't have problems saying so. Although it's still a very bad move from their part.
I suspect lawyers are involved.
Probably. Also PR to limit damages.
PR is just socially accepted lies
Eh, it protects against a certain class of attack when the attacker has physical access e.g. reading memory with memory probes while the computer is (still) on to get passwords etc., i.e. sophisticated attackers like customs, FBI. If they have physical access you're probably hosed anyway, but if you have the presence of mind to shut the machine off (not sleep, hard off if needed) memory encryption becomes irrelevant.
That is not correct. Data can persist in RAM even when powered off, especially if the sticks are frozen. https://en.wikipedia.org/wiki/Cold_boot_attack
Isn't that attack only viable within minutes of a machine being powered down? That seems like a huge caveat...
Not even, try seconds at most.
All things considered, a cold boot attack is only remotely feasible if the system is powered on when the attack begins. If it's powered off for any length of time, your memory will have decayed past the point of it being usable for the attack.
That actually is correct, because if you power your system down ahead of time, this attack is meaningless since there is only a VERY short window where this attack works. From your link:
If your attacker only has your cold machine that's been off since well before you hit the checkpoint, they can't do shit with that attack. At best they can boot the system up to verify your system operates as intended, but you don't have to provide any of the credentials to finish booting or unlock the TPM to load the key material into memory.
To add to that, even the original paper written with 1999-2007 era SDRAM/DDR/DDR2 is not optimistic about the scenario of a machine that was already powered down at regular operating temperatures:
And that only got worse with more advanced RAM, not to mention that they lost almost all of the data far quicker than that with only a couple % of bits surviving that long. For all practical intents and purposes, cold boot against an already-powered-down machine is a myth, the cooling has to be applied while it's on.
Ah, thanks, I stand corrected. Still a good practice.
FYI, the cold boot attack is only viable for a handful of seconds before your memory decays enough for it to be worthless for that attack.
Powering your system down yourself prevents this. Just make sure your system doesn't have fastboot enabled or hibernates instead of a true power off.
TIL. Thanks.
Create the problem, sell the solution situation?
Or just enshitifcation?
what problem did they create?
The lack of memory encryption....
are you saying AMD is the cause why the implementers of the x86 architecture did not encrypt RAM memory from the very beginning?
No, that is not what I am saying.
They did WHAT???!!?
They did a Trump Administration “We’re good on OpSec.”
It feels weird this was even ever a standard consumer feature. I wouldn't even really expect it on enterprise hardware outside of servers. This feels like stuff you only really need to think about if you're being directly targeted by a group with resources.
It seems like it is not a lot of overhead if any at all. Also the hardware design easily accommodates it. So why not if the work is already done?
There is still some overhead. Enough that if you are doing HPC and running on your own hardware you might want want to disable it.
I've disabled it on epycs for this reason but never touched it on ryzens.
that could be said about any and all kinds of encryption
I mean yeah maybe, this one is focused on protecting from threat actors with physical access which is kind of another level.
How's that attempt to get back onto consumers' good side again going for you, AMD?
I wonder if this is to reduce their value of being used as server CPUs.
What the fuck AMD?
Did you know they even had this feature before today
Yes, not that it should matter: Dropping it without notice is shitty regardless.
Is this different than the ddr5's memory encryption?
I don't think DDR5 has any encryption built in? Maybe you're thinking of the error correction controller that's on the module now? Memory with inline encryption is not very common, and as far as I know, not actually very secure if the CPU/TPM isn't the one holding the encryption key.
Right, that was what it was ty