On: Nvidia CEO Jensen Huang announces new AI chips: ‘We need bigger GPUs’

[YetAnotherNick]

How is 2.5x disappointing in one generation?

[dagmx → YetAnotherNick]

Did you skip the sentence immediately after that one?

It’s two fused chips. So 1.25x per chip. 25% uplift. Not 2.5x uplift. The 2.5x is for the whole package.

[downvotetruth → dagmx]

> two fused chips

Jensen's comment about being first was such a dig to Emerald Rapids.

[dagmx → downvotetruth]

Is it the first? The Apple Ultra series chips are two Max’s fused with an interconnect. In which case it’s both CPU and GPU.

I believe this is just the first for a GPU only product.

[throwaway11460 → dagmx]

Is that how it works? Why don't we just put many chips in one computer?

[dagmx → throwaway11460]

the massive blackwell SoC he showed is two Blackwell dies with an interconnect. It’s very similar to what Apple does with their Ultra series.

Then the B200 package is 2 of these plus a CPU. So a total of 4 GPU dies in each unit.

[abhinavk → dagmx]

> Then the B200 package is 2 of these plus a CPU.

That's GB200.

[chimney → YetAnotherNick]

Compare to the 10x that was Hopper uplift.

[YetAnotherNick → chimney]

Because it involved scaling in chip area needed for FP8. AI community realized that FP8 training is possible few years back so the transistors given for FP8 was scaled. Overall I think transistors grow just by ~50% per generation so most of the gains comes from removing FP32/FP64 share which were dominant 10 years back, but there is only some point it could go to.