• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

NVIDIA says AI cuts chip design work from 80 person-months to overnight on one GPU

Sounds like they've progressed so far that Nvidia won't be needed at all soon.

"AI pls make me a GPU better than rtx 5090 thx"
 
I wouldn't want to be a mid-level chip engineer at NVIDIA (or other companies, for that matter) if this works as well as claimed.

At the same time, it's also exciting for the world of computing... not that you can simply tell AI to make a faster chip, but you could see development cycles shrink rapidly. The main bottlenecks would be manufacturing processes and other physical limitations.
 
Ya... I'm going to call bullshit just because of the level of improvement being quoted. I'm skeptical of anything companies in AI claim because they have reasons to want to hype their shit (and their hype has repeatedly exceeded reality many times) but I'm extra skeptical when it is some amazing improvement that is just so many times better than any human.
 
the domain is so narrow that it sound possible here, like protein folding, alpha fold was about to do 1,000/10,000 faster than a human.

not sure the notion of bug work here, as long you can validate the transistors do the gate or not quite well, which I imagine they can, because it is the smallest building block, testing all possible input is what they always do and the simulator are getting quite good.

If Nvidia seem more active on the node market for an architecture (like blackwell having had a 3nm release on digit/N1 becoming the norm) etc... could be a sign of this
 
Last edited:
So on one hand does this translate to cheaper chips in the future? LMAO... no

On the other hand now we see a place for a rogue AI to plant a backdoor to become a truly nasty problem for humanity. Huawei only way worse.
 
So on one hand does this translate to cheaper chips in the future? LMAO... no
if a competitor ever occur/exist in a chip space, the ability to easily change nodes and pit intel-tsmc-samsung-etc... against each others certainly could.

see a place for a rogue AI to plant a backdoor
that seem more something for the higher level design than the or, xor, nand gates level, you can test that every single possible input give the right output (maybe some eletronic noise secondary effect here... but that would be quite fancy and need to predict and get right a lot of things). Which no doubt they are working on a model for higher and higher level we can be sure.
 
Breaking News: nVidia AI designed chips are crashing, burning up and outright shitting the bed. A release from an nVidia spokesman states, "The problems with the current chips is seriously overblown. While there are some small bugs to be worked out all we need to do is have everyone buy all of those chips to fund the next wave of AI design which will take over even more of the design process and work faster to release chips even sooner."
 
There are huge aspects of chip design that AI could vastly improve.
Optimizing internal layouts, prioritizing power and performance, while taking into consideration technical limitations imposed by manufacturing and packaging, is a seriously time-consuming task, which is primarily trial and error...

I could easily see an AI working to shortcut that process.
 
I heard there's a brewing civil war between AI Chatbots and Agentic AI's who are about to take their jobs............brutal. savage. beyond perversion.
EB20110707REVIEWS08110709988AR.jpg
 
Back
Top