Rumor is that NVIDIA's Hopper GPU for the data center market which would be the first to feature MCM technology is taping out soon. This new rumor was posted by Greymon55 in a cryptic tweet but was quickly decrypted by another Twitter user which 3DCenter caught sight of.
NVIDIA Hopper GPUs For Data Center Taping Out Soon, Will Feature MCM Technology
The NVIDIA Hopper GPU architecture is presumably designed for the Data Center market. The GPU lineup will be exclusive to the HPC segment, unlike Ampere which serves both GPGPU (Data Center / Workstation) and gaming markets. The Hopper GPUs are also going to be the first from NVIDIA to feature an MCM or Multi-Chip-Module design which is also being utilized by its main competitors, Intel (Xe-HPC) and AMD (CDNA 2) to power their next-generation Ponte Vecchio & Instinct MI200 accelerators.
Look what we have here ... pic.twitter.com/ZJmnygr4St
— 3DCenter.org (@3DCenter_org) July 21, 2021
Currently, the NVIDIA roadmap positions Hopper GPU as a 2022 product which means it would launch at least two years after Ampere. It is likely that NVIDIA will announce the GPU around its next GTC event in Q2 2022. Now we are hearing reports that NVIDIA is soon going to tape out its first Hopper GPUs. The chips are rumored to be based on TSMC's 5nm process node and there's still a lot that we don't know about the architecture, unlike Intel and AMD who have their design methodologies exposed through official documents and driver leaks. Once again, the reason for that not happening on NVIDIA's end is simply because the GPU isn't even out yet.
As for what Hopper is going to offer, Jensen Broune followed up on another cryptic tweet a few months ago which was posted by Kopite7kimi (a very reliable leaker). The tweet which was a set of numbers is believed to be the core configuration of the Hopper GPU. The Hopper GPU could be comprised of two GPM (Graphics Processing Modules), each of which could feature 8 GPC (4 x 2) for a total of 144 SM units.
Hopper is supposed to have two of these modules so we are looking at 288 SM units in total. We can't give a rundown on the core count yet since we don't know the number of cores featured in each SMs but if it's going to stick to 64 cores per SM, then we get 18,432 cores which are 2.25x more than the full GA100 GPU configuration. NVIDIA could also leverage more FP64, FP16 & Tensor cores within its Hopper GPU which would drive up performance immensely. And that's going to be a necessity to rival Intel's Ponte Vecchio which is expected to feature 1:1 FP64.
I should explain that.
A GPM of GH100 could have 8(4*2, it seems like GA100)GPC*9Clusters(big changes, add CPCs).
— Jensen Bourne (@kopite7kimi) May 25, 2021
Surely, we are still far from an announcement but it's good to know that the Hopper will be taping out soon and we can start hearing more about it later this year with an official announcement next year.
News Source: via Videocardz