Anyone with a brain can figure out that ChatGPT is just more accurate predictive text. “AI” is a massive misnomer, it’s just fuzzy pattern recognition. Even LLMs are just predicting what word comes next over and over.
It can be a very useful tool, but it’s wholly incapable of doing anything but regurgitating mashups of its training data.
That is, unfortunately, sorely outdated. Particularly with the advent of tensorRT. Best case vs best case the 4080 is about twice as fast today
The actual rule has hard numbers, no need to speculate. And it’s no more than 300 TFLOPS of fp16 (or 150 fp32, 600 fp8, etc) so it ain’t TFLOPS that are the culprit. As for performance density, it’s equivalent to those figures at an 830mm^2 die, so again not that.
I doubt police are going to raid a company for a civil lawsuit over IP infringement. Whatever the raid is for I bet it’s more significant than this.
It still boggles my mind that on breaking into completely new territory they sat down, said “we’re going to make a MCM using 47 chiplets fabbed on 5 different nodes*”, and nobody suggested that maybe they should learn to crawl before attempting the steeplechase
Well the 512 rumour was kopite, and this is also kopite saying he misinterpreted a 128MB L2$ to mean 512
It’s highly likely to be a major architecture update, so core count alone won’t be a good indicator of performance.
It’s not specifically fp8, but TOPS*data size. Absolute limit is 4800, or 5.8/mm^(2). Above either is an outright ban. Above half of either needs a license.
They literally isolate GPU-only power draw in the review, so you know the 4090L is pulling 161 W to the 7900M’s 179 W. For 28% more FPS, not 10%
The higher power draw is mostly from the CPU. Efficiency complaints - at least for the CPU - are largely around idle power.