Incorrect. The leaks clearly state MCM is for GB100 only, 200 is all monolithic. There won't be any MCM gaming GPUs in the Blackwell generation
What I do find interesting is that Nvidia would launch server GPUs first on new architecture an entire year before it lets gaming GPUs get the new architecture- ensuring that all silicon used to build new architecture GPUs is sold to corporations for mega profit for an entire year before gamers can get any
It's basically Apple's strategy - your new architecture is launched and sold at crazy prices and mass market products have to wait a full year before they can get the architecture, creating a constant performance gap between consumer products and industrial products. Now they just need AMD to do the same and we'll have entire new way to screw gamers, and based on the rumours that RX8000 GPUs will be mid range cards only, I'd say it sounds like AMD is following Nvidia's strategy too
I think you are incorrect stated there wont be any Blackwell MCM gaming GPUs. Look at AMD Navi 31 and 32 was the first AMD high end gaming MCM GPUs and also Mi300 was the first AMD data center MCM GPU. Now we know Nvidia are now catch up AMD data center MCM GPU with Nvidia first MCM GPU Blackwell GB100 so Nvidia will need to catch up with Navi 31 MCM GPU design.
Blackwell will be mark the 3rd time Nvidia will use the same architecture in both data center and gaming since Pascal and Ampere.
I guess cut down GB102 was aimed sell to China and others will be cancelled now after US enforced A800 and RTX 4090 sales ban listed China and others regions.
Now we have some information on GB100 that it will have 10 Texture Processing Clusters, 8 Graphic Processing Clusters with 160 SM so mean it will have 20,480 CUDA cores but we now have some very interesting information on GB202 that it will have 12 Texture Processing Clusters, 8 Graphic Processing Clusters with 192 SM so it will have massive 24,576 CUDA cores compared to GB100's 20,480 CUDA cores. I think GB202 RT will have 192 4th generation RT cores because it has same numbers as 192 SM.
AD102 has 76.3 billion transisters on TSMC 4nm with 144 SM and 609mm2 die size that is slight smaller than GH100's 80 billion transisters with 132 SM and huge 814mm2 die on same TSMC 4nm process and also it has same 18,432 CUDA cores count as AD102.
So if you look at GB202 numbers, it look like a massive chip contain maybe 160 billion transistors and it could be bigger than GB100 or rough the same size but it did not make any sense that GB202 is a monolithic chip while GB100 is a MCM chip. I dont think you can shrink to fit on 3nm 609mm2 or 750mm2 die size. Look at AMD Mi300 for example it has 146 billion transistors with 1017mm2 die size total on TSMC 5nm.
Q3 2025 is a long time away for Blackwell gaming GPUs launch but interesting Nvidia will launch next generation data center X100 in Q2 2025 after Blackwell GB100 in Q2 2024 so I think I figured out the X architecture name will be Xanthopoulos as here is only 1 X long name on physicists names list.