• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Blackwell gpus

GEG33Pk.jpg

You could get a job at the royal family with those photo editing skills.
 
Last "leak" before announcement

Blackwell B100: uses an MCM architecture design with dual chiplets, using TSMc's CoWoS-L 2.5D packaging. The chips are connected to 8 stacks of HBM3e memory with a 192GB capacity

Total GPU silicon is going to be big, it's why the TDP is rumoured to be so high for Blackwell server products
 
Last edited:
Last "leak" before announcement

Blackwell B100: uses an MCM architecture design with dual chiplets, using TSMc's CoWoS-L 2.5D packaging. The chips are connected to 8 stacks of HBM3e memory with a 192GB capacity

Total GPU silicon is going to be big, it's why the TDP is rumoured to be so high for Blackwell server products
It would be interesting to see if nVIDIA would pull an actual MCM design before AMD to the GPU - not just the memory subsystem in chiplets, but either a multi-gpu single package as one or the "total" split of the GPU die into chiplets.
 
  • Like
Reactions: TNA
Last "leak" before announcement

Blackwell B100: uses an MCM architecture design with dual chiplets, using TSMc's CoWoS-L 2.5D packaging. The chips are connected to 8 stacks of HBM3e memory with a 192GB capacity

Total GPU silicon is going to be big, it's why the TDP is rumoured to be so high for Blackwell server products
Well, that sounds cheap. :D
 
Blackwell B100: uses an MCM architecture design with dual chiplets, using TSMc's CoWoS-L 2.5D packaging. The chips are connected to 8 stacks of HBM3e memory with a 192GB capacity
It has been rumoured for a year or so that Blackwell for AI/Professional would be using chiplet design and the gaming side monolithic.
 
It has been rumoured for a year or so that Blackwell for AI/Professional would be using chiplet design and the gaming side monolithic.

That is what I expect as well, I don't see the gaming cards being MCM as well unless there has being a massive improvement in the bandwidth of infinity fabric in the last 2 years. From what I can tell, B100 would be two GPU chips sitting next to eachother, surrounded by the 8 HBM3e memory chips and all of these chips sit on a giant interposer - this interposer from some older articles comes in at 5TB/s, which is the same bandwidth on the infinity fabric the 7900XTX has and one of the AMD engineers says 5TB/s was not fast enough to use MCM to create a dual chip gaming GPU and that only memory could be turned into chiplets for gaming.

Something like this:

The yellow section is the interposer that connects everything over its 5TB/s connection

 
Last edited:
Back
Top Bottom