On: Google TPU v5p beats Nvidia H100

[bhewes]

I am more interested in TCO of TPU v5p. Right now it seems only Nvidia is making money everyone else loses.

[lordswork → bhewes]

The article mentions TPU v5p is 2.1 x perf/TCO of H100.

[jsnell → lordswork]

I don't think it does?

> Google's v5p TPUs are up to 2.8 times faster at training large language models than TPU v4, and offer 2.1-times value-for-money.

That's a comparison of TPUv5p to TPUv4, not to H100.

[bhewes → lordswork]

Actual dollars would be nice, since that is what the end user understands. Otherwise what are they talking about 1 million unit orders? But nice catch I didn't see it.

[nabla9 → bhewes]

I think Broadcomm designs and builds TPUs for Google. Their stock was up 100% in 2023 and their profit margin was 39.31%.

[bhewes → nabla9]

They bought VMware for $61B in Nov of last year. Hard to think TPUs are driving much of this at Broadcom.

[surajrmal → nabla9]

Google's TPUs are entirely in house designed. Not sure where you heard broadcomm is involved.

[rreichman → surajrmal]

https://siliconangle.com/2023/09/21/google-reaffirms-broadco...

[windexh8er → surajrmal]

Google doesn't build their TPUs. Broadcom does.

[Y_Y → bhewes]

table of contents?

[kirubakaran → Y_Y]

I think they mean TCO: https://en.wikipedia.org/wiki/Total_cost_of_ownership

[bhewes → kirubakaran]

Thanks.