Hacker News Reader

Comment by ChuckMcM | Hacker News Reader

Simulation fail? I'm surprised they didn't know exactly how much power it would draw, that seems to be a standard thing in chip design suites. If a company like Synopsys blew it I'd expect some financial recovery actions.

That said, unprecedented? Pretty much every chip company I know of has made a chip that didn't meet thermals. The Pentium 5 was the poster child for a while of a reaching too far.

View on HN

I wonder if A16 was backported from N3 to N4 which increased power consumption.

Piling on the speculation, the dynamic power draw when raytracing was enabled, especially the memory access patterns - and the poor performance in spite of the increased power draw - might have been the final blow.

But if an entire h/w project is punted to "the next release" there's more than just one feature that contributed. So maybe the halo feature, raytracing, plus the other new Metal 3 features all contributed?

The article doesn't say it was an issue with the tools; it sounds like the design was too ambitious.

Leakage power is not hard to quantify, but dynamic power is harder to evaluate, because it's workload-dependent. It's not always completely straightforward to get an answer.

Perhaps it used a new unproven process, making power dissipation difficult to predict?

Sounds like they just discovered it late in the design process, not in silicon.

One would think that the makers of smartphone chips would be particularly good at and experienced with calculating power draw.

Reply on news.ycombinator.com