Ever since our first inkling of Nvidia's next-gen Lovelace GPUs, folks have assumed that it was a refresh of the current Nvidia Ampere architecture. Was that underselling what Nvidia has in store, however?
Noted Twitter leaker kopite7kimi is now saying that the original GPU planned for Lovelace, which they estimated would a little better than double the performance of the Ampere GPU it was replacing, is no longer the GPU Nvidia is working on. Indeed, the new AD102 GPU is more than just a refresh of the existing Ampere architecture.
This would mean that the next-gen AD102 GPU, which will likely power the Nvidia RTX 4080 and RTX 4090 graphics cards, is more like a whole new architecture rather than a refinement of what previously existed. PC Gamer notes that this could very well be in response to the rumored performance of AMD RDNA 3, the next-gen GPU architecture expected in AMD's Radeon RX 7000-series GPUs.
The new tweet clarifies what kopite7kimi said about a year ago, which was that while the Lovelace AD102 GPU would see a performance increase of about 2.2X, the RDNA 3 Navi31 GPU would see a 2.5X increase over the Navi21 GPU found in the AMD Radeon RX 6900 XT.
If Nvidia got wind of what AMD was cooking up, that could have put a lot of pressure on the engineering team to seriously boost performance rather than provide a more modest improvement over the GA102 GPU found in the RTX 3080 and RTX 3090, and it puts other recent reports about a 900W RTX 4090 graphics card from Nvidia in a new context.
Analysis: competition is great (when it works)
There is something to be said for competition. Forcing workers around the world to compete for ever-lower wages in a race to the bottom that immiserates everyone is what happens when competition goes off the rails (at least from the perspective of everyone but company execs who pocket the savings for themselves and their shareholders. Happy May Day y'all!). We've seen a lot of this kind of competition lately, especially in the tech space, but every once in a while, competition can actually work to consumers benefit.
If its true that Nvidia was just planning on an Ampere refresh for Lovelace (always take rumos with a grain of salt, even if leakers have been reliable in the past), that wouldn't have been nothing, but it also would have been safe. Nvidia is in the pole position in the graphics card market, and when you're in the lead, it's very easy – and common – to get complacent and try to coast to a victory. Ask Intel how well that worked out for them.
You can't coast to victory though, and AMD has every reason to push harder and take risks to overtake its archrival, and nothing spooks the an leader than seeing a hungry rival gaining ground quickly in the rearview.
Nvidia Ampere (for those lucky enough to have gotten their hands on it) is incredible, especially at the higher and lower ends, where the RTX 3090 Ti delivers just jaw dropping performance and the RTX 3060 Ti with DLSS can actually make 4K gaming something that everyone can enjoy (at least to some extent). Doubling that would be impressive, sure, but nearly tripling what we've seen from this generation? That would be game-changing.
Sure, we could get closer to 3X improvement in the generation after Lovelace, but if AMD can force Nvidia to get there in the next generation, than we all win, and the improvements with the generation after Lovelace will be that much better for it.
