Earlier tweet (via Hardware Info) indicates that the GB100 GPUs will use a chip-based multi-chip module (MCM) instead of current monolithic designs.
MCM is an advanced packaging technology that partitions GPU components across separate dies. The leaker doesn’t tell us what the configuration will look like, such as whether it will contain multiple GPUs count – count Dies, or discrete memory dies along with a single GPU core – and that doesn’t indicate whether the GB100 will be Blackwell’s only GPU with a multi-die design.
This shift from a monolithic focus would give Nvidia’s data center GPUs some wiggle room for potential chip customization in order to compete with AMD’s Instinct MI300 series chips. Red Team’s data center chips cram the entire Zen 4 CPU inside a very massive GPU.
AMD also already has some gaming GPUs that we like so far, like the AMD Radeon RX 7900 XTX, although its inaugural chipset still doesn’t have multiple GPUs. count – count Die, just separate compute, I/O and memory.
After the dramas of the GA100 and GH100, it looks like the GB100 will finally use MCM.September 18, 2023
Anyway, the chips rumored here are specifically data center GPUs, which may not be very interesting to us PC gamers. Although the Blackwell architecture will be coming to our GeForce cards, under the name GB200. However, if Nvidia he have If we find a way to make multiple compute templates work within the traditional game show pipeline, we might still have something here to get very excited about.
Because that’s the hard thing. When it comes to GPUs in data centers, performing the full range of critical number crunching, rather than traditional frame rendering, is much easier when they are spread across multiple compute templates. For this reason, the SLI and Crossfire multi-GPU solution has continued to work its magic in computing long after its brief implementation in gaming.
This is essentially what we’re looking at when it comes to multi-compute MCM graphics cards: an SLI/Crossfire solution buried in a single chip. There’s no mention here specifically of whether Nvidia is looking to roll this out to next-gen GeForce cards.
We can be optimistic, but we would probably suggest that this is unlikely to happen in the next generation of GPUs. But Nvidia’s next-generation gaming GPUs have been rumored to not happen until at least 2025, which is plenty of time to know, right, Jensen?