So Nvidia had their big GTC keynote last Tuesday (the 20th) and announced the RTX 40-series, and my first reaction was the same as everyone else's: okay, the 4090 looks like an absolute monster, $1,599, releasing October 12th. Fine. Expected. GPUs have been creeping toward absurd for years now and I've made my peace with it, mostly because I'm not buying one.
But then I actually read past the headline number and noticed the thing that's been bugging me for four days straight: there are two RTX 4080s, and they are not the same card wearing two different amounts of memory. There's a 16GB version at $1,199 and a 12GB version at $899, and I assumed, like I think most people assumed on Tuesday, that the difference was just VRAM and maybe a small clock bump. Nope. Different die entirely. The 16GB card uses the AD103 chip, the 12GB card uses AD104, and the 12GB one has something like 22% fewer CUDA cores and a narrower memory bus on top of the smaller VRAM pool. That's not a memory tier. That's two different GPUs that historically would've been called a 4080 and a 4070, both wearing the 4080 badge because, I guess, $899 sounds better attached to "4080" than it does attached to "4070."
I want to be clear I don't think this is some unprecedented act of corporate evil, companies fudge product tiers constantly. But this one's blatant enough that it's already generating pushback across the usual hardware forums, and it should. If you buy the 12GB "4080" expecting 4080 performance minus a little headroom, you're going to be disappointed, and Nvidia knows that, and priced it $300 under the real one specifically so people wouldn't stop and ask why.
Here's my small, dumb, personal complaint buried in the middle of this post: I still have a GTX 1070 in my desktop from 2016 that I bought for $379 at launch, and it plays basically everything I throw at it on lower settings, and every time one of these launches happens I feel a little more insane for not upgrading and a little more validated for not upgrading, both at once. Six years and it still boots Half-Life: Alyx in a pinch (badly, but it boots). I keep telling myself I'll wait for reviews. I've been telling myself that since the 3080 came out in 2020.
Speaking of which, worth remembering where GPU prices were even a few months ago. Crypto mining demand cratering after Ethereum's move off proof-of-work has already been pushing used 3080s and 3090s down on eBay, and now you've got a whole new generation stacked on top of a secondary market that's still digesting the last one. If you don't need a card right now, this is genuinely a weird time to buy anything at MSRP. Let the market sort itself out for a month or two.
Anyway, back to the naming thing, because it's the actual story here, not the price. Nvidia has done this before in smaller ways (remember the 1060 3GB vs 6GB situation, where two cards with the same name had different core counts too), but stacking two separate silicon dies under one product name, at a nearly $300 price gap, in the same generation, feels like a new low bar. My guess, and it's just a guess, is that enough site editors and YouTube reviewers make noise about it that Nvidia either renames the 12GB card before launch or quietly lets it die a slow death in the reviews cycle. I'd put decent odds on a rename actually happening, honestly, once the embargo pieces start dropping in a few weeks and every single review has to spend its first three paragraphs explaining why there are two "4080s."
If you're in the market for one of these, do yourself a favor and read the die specs before you read the marketing slide. The name on the box tells you nothing useful this generation.