By: anon (anon.delete@this.anon.com), July 1, 2013 1:45 am
Room: Moderated Discussions
Exophase (exophase.delete@this.gmail.com) on July 1, 2013 12:44 am wrote:
> The prospect of in-order FEC units on Silvermont led to a big discussion about the relevance
> of FPU performance in mobile CPUs (or is it CPUs in general?), but those units almost certainly
> handle all SIMD operations. So is integer SIMD (and shuffles, etc) also pointless?
Well on one hand, you have the market (which is very competitive, mind you) of producers and consumers deciding that a very significant amount of floating point performance has value.
On the other hand, you have some people on this forum asserting that it is largely irrelevant.
The ironic part of this is that Linus believes in the market when it suits his worldview ("market decided x86 is better than xyz despite some bickering from people in ivory towers, therefore your technical arguments are simply masturbation"). On the other hand, he is quite sure that the market is wrong to be driving floating point performance in this way.
> The prospect of in-order FEC units on Silvermont led to a big discussion about the relevance
> of FPU performance in mobile CPUs (or is it CPUs in general?), but those units almost certainly
> handle all SIMD operations. So is integer SIMD (and shuffles, etc) also pointless?
Well on one hand, you have the market (which is very competitive, mind you) of producers and consumers deciding that a very significant amount of floating point performance has value.
On the other hand, you have some people on this forum asserting that it is largely irrelevant.
The ironic part of this is that Linus believes in the market when it suits his worldview ("market decided x86 is better than xyz despite some bickering from people in ivory towers, therefore your technical arguments are simply masturbation"). On the other hand, he is quite sure that the market is wrong to be driving floating point performance in this way.
