Quick Note: NVIDIA’s “Einstein” Architecture Was A Real Projectby Ryan Smith on March 19, 2019 12:30 AM EST
While it was never an official NVIDIA codename as far as roadmaps go, the name “Einstein” came up in rumors a few times earlier this decade. At the time, Einstein was rumored to be the architecture that would follow Maxwell in the NVIDIA lineup. And while we sadly didn’t find out anything new about NVIDIA’s future roadmap at this year’s show – or any sign of Ampere or other 7nm chips – I did inadvertently find out that the rumors about Einstein were true. At least, from a certain point of view.
While talking with NVIDIA’s research group this morning about some of their latest projects (more on this a bit later this week when I have the time), the group was talking about past research projects. And, as it turns out, one of those former research projects was Einstein.
Rather than just being a baseless rumor, Einstein was in fact a real project at NVIDIA. However rather than being an architecture, per-se, it was a research GPU that the NVIDIA research group was working on. And although this research project didn’t bear fruit under the Einstein name, it did under another name that is far more well-known: Volta.
So while this means we can scratch Einstein off of the list of names for potential future NVIDIA architectures, the project itself was real, and it was actually a big success for NVIDIA. As Einstein morphed into what became the Volta architecture, it has become the cornerstone of what are now all of NVIDIA’s current-generation GPUs for servers and clients. This includes both regular Volta and it’s graphics-enhanced derivative, Turing.