By: forestlaughing (forestlaughing.delete@this.yahoo.com), July 25, 2012 8:51 am
Room: Moderated Discussions
David Kanter (dkanter.delete@this.realworldtech.com) on July 25, 2012 1:37 am wrote:
> New computational efficiency data shows GPUs with a clear edge over CPUs, but
> the gap is narrowing as CPUs adopt wide vectors (e.g. AVX). Surprisingly, a
> throughput CPU is the most energy efficient processor, offering hope for future
> architectures. Our data also shows some advantages of AMD's Bulldozer, and the
> overhead associated with highly scalable server CPUs.
>
> Comments and feedback
> welcome!
>
> David
So what's the point of looking at peak DP floating-point? That's not really a good measure, well, of anything. One can pack a fpga full of vector pipes, and get way more peak throughput than these processors, if you don't need reasonable memory bandwidth and control structures to go with the ALU. Granted, the fact that all of these are successful commercial products, indicates that they have all maintained some reasonable ratio of compute and their other functionality.
Are we assessing these processors based on their applicability to the HPC marketplace? Not that linpack is a good benchmark, in fact it's a pretty terrible benchmark, but even that is better than simply looking at peak throughput. If we assess these processors for other applications, then why look at Floating-point. Almost nothing else is FPU bound.
I guess what I'm saying is, these graphs don't really show anything of interest. Do they?
> New computational efficiency data shows GPUs with a clear edge over CPUs, but
> the gap is narrowing as CPUs adopt wide vectors (e.g. AVX). Surprisingly, a
> throughput CPU is the most energy efficient processor, offering hope for future
> architectures. Our data also shows some advantages of AMD's Bulldozer, and the
> overhead associated with highly scalable server CPUs.
>
> Comments and feedback
> welcome!
>
> David
So what's the point of looking at peak DP floating-point? That's not really a good measure, well, of anything. One can pack a fpga full of vector pipes, and get way more peak throughput than these processors, if you don't need reasonable memory bandwidth and control structures to go with the ALU. Granted, the fact that all of these are successful commercial products, indicates that they have all maintained some reasonable ratio of compute and their other functionality.
Are we assessing these processors based on their applicability to the HPC marketplace? Not that linpack is a good benchmark, in fact it's a pretty terrible benchmark, but even that is better than simply looking at peak throughput. If we assess these processors for other applications, then why look at Floating-point. Almost nothing else is FPU bound.
I guess what I'm saying is, these graphs don't really show anything of interest. Do they?
