New Posts  All Forums:

Posts by sumitlian

that and your profile pic perfectly complete the expression.
In terms of IPC, x86-64 seems to be dead for serial improvements or may be Intel is only putting all the efforts to improve vectorization only.
Yeah that would make sense then. I just was surprised by his example where he automatically assumed Vulkan as an inferior alternative to DX. Guess then he too can't complain about it if it doesn't exist.
Number of instruction decoded per cycle do not tell the whole story, there are tens to hundreds of clocks are used during the fetch/decode and execute and memory operations (both in Intel and AMD) before you get the final result. Even if Zen has to decode more instructions per clock, improvements in other areas might do well in AVX2 like scenarios. But for now, it is all theoretical. We don't even know if Intel compiler is allowing full flops for AMD CPUs after seeing the...
4 decodes per cycle, Zen should be able to do 256bit AVX (or AMD's equivalent for similar SIMD processing), they are probably talking about four doubles per clock.AVX2 was designed to be decoded as 8 single precision per clock, Zen doesn't seem to do 8 instructions per clock, that means AVX2 can't be fetched in a single cycle.
You have this confidence to fantasize anything you want and yet you can't/don't provide any actual proof to support it.I've yet to hear from a single developer claiming that Vulkan is worse than DX12.
I second this.
Oh man I missed I think nVidia's power efficiency is due to deferred rendering as well.
So has never been AMD caught using a "Cripple Intel" function in anywhere.Do you remember ? "it is not AMD's job to optimize code for non Genuine AMD CPUs".
By "Optimized to favor one side" you mean, as in Intel compiled Cinebench ?
New Posts  All Forums: