[The Nvidia Ampere GTC keynote has ended. Jensen Huang returned wearing his trademark black leather jacket. We all got a tour of the Nvidia CEO's kitchen and marveled at the number of spatulas he had (honestly, why so many spatulas?)

Kitchen utensils aside, the keynote consisted of eight videos covering autonomous driving, medical research, conversational AI, robotics, and other things that are important to Nvidia but not to us gamers. After seeing all of this, one might think that there was no real information about Nvidia's upcoming gaming graphics cards. And while there were a few references to gaming graphics, there were no direct details about the hardware itself.

However, during the GTC keynote, many details about the A100 were announced: the A100 is designed for the data center, but the same architecture will be featured in future next-generation GeForce graphics cards. [The 826mm square is sturdy, but we're not just talking about the physical size, we're talking about the fact that they squeezed 54 billion transistors into it; Nvidia's previous datacenter product was the Volta V100, so there's no direct comparison here to the current Nvidia claims a 20x performance improvement over Volta in certain tasks. For comparison, the chip had 21.1 billion transistors and was 815 mm square. In other words, the A100 has 2.5 times more transistors, but is only slightly larger; TSMC's jump to a 7nm process shows its worth.

If you're looking for real-time ray tracing, the second video has lots of footage of "Minecraft RTX" and a fun marble game called "Marble RTX". There was no mention of the ray-tracing RT core, as it is not used in data centers, but it does seem to improve the performance of the Tensor Core considerably. (However, Tensor Core performance does seem to have improved considerably.)

The actual number of tensor cores per SM has been reduced to four (Turing had eight), but Nvidia still claims an overall performance increase, up to 2.5x in some scenarios. Rumors of increased tensor performance seem to have been right on target, but we did not expect an actual overall reduction in cores.

What we don't know from these is what kind of performance we will get from the next generation of graphics cards. At this stage, the number of cores and clock speeds are unknown, but for the latest information, check out everything we have on Nvidia Ampere.