Nvidia may not make the jump to MCM with flagship ‘Hopper’ GPU

We're still a while away from Nvidia releasing its next-gen GPUs, but rumours about the company's upcoming ‘Hopper' architecture have been floating around for a while. Previously, rumours claimed that the flagship GH100 GPU would use an MCM design – something that AMD is also working on for future GPUs. However, according to leakers, this is not the case, so GH100 is expected to use a monolithic die, one that is around 20% larger than the GA100 GPU package. 

The rumour shared by @kopite7kimi claims that Nvidia's GH100 Hopper GPU will use a monolithic design, with a package area of almost 1000mm². That's about 20% bigger than the Ampere-based GA100 GPU's package. The difference between the actual dies might differ.

GH100 does not have MCM, it's still monolithic.

— Greymon55 (@greymon55) January 29, 2022

This rumour contradicts older ones stating the GH100 GPU would feature an MCM (Multi-chip module) design. Still, @greymon55 claimed some Hopper products would come with that configuration, most likely referring to the GH102 GPU.

Summing up all the rumours we've heard about the GH100 GPU, the new architecture will be based on TSMC's 5nm node (N5 or N5P), featuring up to 18,432 CUDA cores (288 streaming processors on 16 GPU clusters) and an HBM2e interface. TDP is rumoured to be around the 1000W mark.

Many believe Nvidia will tell us more about the Hopper architecture at GTC 2022, but only if it can solve the trademark dispute with Dish Network (via Tom's Hardware). If the latter wins the case, Nvidia might have to change the name of its upcoming GPU architecture.

Discuss on our Facebook page, HERE.

KitGuru says: A 1000W TDP for an MCM configuration makes sense, but now seeing the GH100 GPU might come in a monolithic design, we're starting to wonder about the credibility of this rumour.

Become a Patron!