Anyway we'll see soon how Cannonlake (low power 10nm) goes as they're releasing soon. The physics score from that 3dmark would be interesting to know but I can't find it.Īnyway with the latest Intel roadmap not mentioning Icelake (10nm octocore) at all, maybe we won't see it until next year. I've since seen an Intel chart reproduced on chipwiki showing a claimed 26% increased performance for same power or 52% less power for same performance on coffee lake 14nm VS broadwell 14nm at 0.7V, an eyeball estimate on that chart suggests ~35% lower power at same performance of cfl 14nm vs Kabylake 14nm, but that's not at 1.1-1.3V which is where the H chips will end up In other words those two extra cores will cost power, or frequency. That kind of leap is seen with a node shrink. Click to expand.I think you misread I was saying I thought there is no way in hell there's a 33% increase in power efficiency from a simple process tweak on a process that's been tweaked twice already.
0 kommentar(er)
