Killing relative to what? Have you seen how much an old compute card can cost second-hand? Consider AMD's mi25 (Vega10). If you want one of those, you'd better be prepared to pay over $8k for one of those. Buying one new from amd.com will set you back $10k.
Deep learning is hot right now, and for the foreseeable future, dGPUs will provide the best AI deep learning performance. So if I am spending my R&D dollars on a next-gen dGPU, my first target is AI. Not gaming.
This is true, but it's not an unlimited market, nor is it enough to make up for the volume involved with mass market gaming or mining that they previously enjoyed.
The 'everything AND the kitchen sink' approach with RTX so far is both really expensive to the consumer, and not even all that profitable for Nvidia.
It's a weird choice, because it seems like having massive dies with Tensor/RT would have been sellable to the pro market (AI, Deep Learning, etc) for actually much much higher pricing. Like perhaps $12k-$20k for a slightly enhanced 2080ti.
On the flip side, a conventional 12nm gaming focused die the size of 2060 could have sold for $399 but been beyond 1080ti performance, and a true 1060 replacement could have been $249 with between 1070ti-1080 performance, both with reduced power draw/heat, and smaller PCB and HSF, further increasing profit.
In a world where mining didn't crash and every (above minimally effective) GPU sold, doing the single big RTX plan made sense to a point. Cheaper to not have to develop a pro and a consumer lineup. But with 10xx living so long, and mining losing the maniacal hype levels, they're left with a much reduced market willing to pay these prices, and investors are going to start asking questions soon unless sales really pick up. 2060 is moving alright, 2070+ are pretty much sitting on shelves and in warehouses, because even at a value to certain extremely specific markets the demand just isn't there, and for almost everyone who bought a GPU since 2016, there is no reason for them to buy a new one that's anything less than a 2080ti UNLESS they are prepared to spend a lot more.
I think it will all settle out, though the current situation is a weird and less than ideal one for almost everyone including Nvidia themselves, though for specific professional fields, these RTX cards and even the Vega 7 match well.
Facebook
Twitter