NVIDIA's RTX 50 series might turn out to be the first mainstream MCM-based GPU from the company - if recent reports are to be believed. Not only that but Blackwell architecture could very well bring the largest performance leap in the company's history. NVIDIA recently confirmed that the Blackwell GPU architecture will be releasing in 2024. Since these are very preliminary rumblings, this post is very much a rumor so keep that grain of salt handy.
NVIDIA RTX 50 GPUs feature new SM design and a raytracing denoising accelerator
Now before I begin, I would like to point out that Blackwell architecture was planned to be the successor to Hopper architecture and the leak using Blackwell and RTX 50 interchangeably could mean several things. Hopper's parallel release in the consumer segment (RTX 40) had the codename of Ada Lovelace, so it is possible that Blackwell will also get its consumer counterpart. Alternatively, it is possible that Blackwell is actually the consumer codename (although admittedly, this is less probable) and the datacenter side of things will get a new name. Previous reports have indicated that Blackwell GPUs will be fabricated on TSMC's 3nm process.
The rumor comes from RedGamingTech and they have received some new information regarding the architecture. Firstly, Blackwell will feature an entirely new SM structure. Considering the underlying micro architecture is shifting to an MCM design with Blackwell - this is not surprising. Also, Blackwell will leverage a hyperspeed bus that will interlink the various SM and chiplets. A denoising accelerator will also be part of the ray tracing pipeline (modern path tracing setups don't actually trace the full sequence, they do it partially and a denoiser handles the rest) which should result in significantly improved RT performance.
There still seems to be no word on specifications although the source notes that there are various Blackwell GPUs being considered and a lot of the binning will depend on how AMD's current RDNA 3 and future RDNA4 offerings perform. Finally, RGT leaves us with the following teaser: "biggest perf leap in NVIDIA's history".
NVIDIA Hopper was the world's fastest 4nm GPU at launch and the world's first with HBM3 memory. It featured higher specifications than even the NVIDIA RTX 4090 (which contains 16,384 CUDA cores) at a net total of 18,432 CUDA cores. Blackwell will provide significant generational improvement over Hopper (as has always been the case). Four NVIDIA Blackwell GPUs have already been confirmed in a prior leak.
The NVIDIA Blackwell architecture was based after David Blackwell who was an American statistician and mathematician that made significant contributions to game theory, probability theory, information theory and statistics. He was also the first African American to be inducted into the National Academy of Sciences. Blackwell continues NVIDIA's trend of naming major architectures after prominent computer scientists and mathematicians and not much else is known about the Blackwell architecture at this point. It is very likely that we will eventually be introduced to the parallel nomenclature (as Ada Lovelace is to Hopper) that will complete NVIDIA's first MCM duo.