As the AI market continues to balloon, experts are warning that its VC-driven rise is eerily similar to that of the dot com bubble.
If it crashes hard I look forward to all the cheap server hardware that will be in the secondhand market in a few years. One I’m particularly excited about is the 4000 sff, single slot, 75w, 20GB, and ~3070 performance.
Especially all the graphics cards being bought up to run this stuff. Nvidia has been keeping prices way too high, egged on first by the blockchain hype stupidity and now this AI hype stupidity. I paid $230 for a high end graphics card in 2008 (8800 GT), $340 for a high end graphics card in 2017 (GTX 1070), and now it looks like if I want to get about the same level now, it’d be $1000 (RTX 4080).
You can’t really compare an 8800gt to a 1070 to a 4080.
8800gt was just another era, the 1070 is the 70 series from a time where they had the ti and the titan, and the 4080 is the top gpu other than the 4090.
If you wanted to compare to the 10 series, a better match for the 4080 would be the 1080ti, which I own, and paid like 750 for back in 2017.
Sure, they’re on the money grabbing train now, and the 4080 should realistically be around 20% cheaper - around 800 bucks, to be fair.
Thing is though, if you just want gaming, a 4070 or 4060 is enough. They did gimp the VRAM though, which is not too great. If those cards came standard with 16gb of VRAM, they’d be all good.
An RTX 4070 is still $600, which is still way higher than what I paid for a GTX 1070, and the gimping of the VRAM is part of the problem. Either way, I’m fine with staying out of the market. If prices don’t regain some level of sanity, I’ll probably buy an old card years from now.
I figured the gear they were using was orders of magnitude heftier than those cards. Stuff like the h100 cards that go for the price of a loaded SUV.
They are, but training models is hard and inference (actually using them) is (relatively) cheap. If you make a a GPT-3 size model you don’t always need the full H100 with 80+ gb to run it when things like quantization show that you can get 99% of its performance at >1/4 the size.
Thus NVIDIA selling this at 3k as an ‘AI’ card, even though it wont be as fast. If they need top speed for inference though, yea, H100 is still the way they would go.
I’m not an expert just parroting info from Jayz2cents (YouTuber), but the big AI groups are using $10,000 cards for their stuff. Individuals or smaller companies are taking/going to take what’s left with GPUs to do their own development. This could mean another GPU shortage like the mining shortage andi would assume another bust would result in a flooded used market when it happens. Could be wrong, but he’s been correct pretty consistently with his predictions of other computer related stuff. Although, 10K is a little bit less than your fully loaded SUV example.
