3 ms·
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
by ChuckMcM 4y ago
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.
- dboreham 4y agoPerhaps it used a new unproven process, making power dissipation difficult to predict?
- ip26 4y agoSounds like they just discovered it late in the design process, not in silicon.
- jasonhansel 4y agoOne would think that the makers of smartphone chips would be particularly good at and experienced with calculating power draw.
- wmf 4y agoI wonder if A16 was backported from N3 to N4 which increased power consumption.
- therealcamino 4y agoThe 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.
- sounds 4y agoPiling 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?