By: itsmydamnation (no.delete@this.way.com), October 26, 2015 1:49 am
Room: Moderated Discussions
Contrarian (Contrarian.delete@this.hotmail.com) on October 25, 2015 12:25 pm wrote:
> mpx (mpx.delete@this.nomail.pl) on October 5, 2015 9:37 am wrote:
> > The 128-bit choice could also stem from having the same architecture for both x86 and ARM.
> > As ARM has no 256 bit vector processing, so a common denominator is 128-bit units.
>
> Intel has 2 read and 1 write ports to cache, same as all other high end CPU's, but Intel has three AGU
> units and thus can use all of those ports. For problems that fit in cache and need lots of bandwidth Intel
> will win. For much software this would just mean you become RAM bandwidth limited with fewer cores.
>
> Some games and other software where a primary thread is the performance limit would be an
> easy win for Intel. But for games AMD has better graghics to offset that to a degree.
>
> There are lots of other factors, many of which Intel will have a slight advantage
> in, AMD Zen was always going to be a small cut below Intel in performance.
>
> It looks like Zen will compete in the budget server market against ARM, and in the mid range laptop/desktop
> market. A big improvement from the budget laptop/desktop market sliver AMD has now. AMD might be back to
> struggling break even instead of going out of business. Back to a joke $7 stock. Sound reasonable?
>
So much like juanrga thinks all ALU ops are executed in 1 cycle, ALU's cant do mov's and registers don't exist.
I would point out since sandy bridge AGU and ALU has increased far more than the improvement in performance per clock* and that's with all the other core and uncore improvements. You could even possibly argue as far back as nehalem (1 load 1 store AGU), but there was a pretty big jump from it to sandy bridge but how much of that was cache overhaul, uop cache and other general improvements.
In the consumer K space intel had to release devils canyon to finally create an SKU that actually offered improvements over sandy bridge thax to sandys awesome overclockability. You consider these points and then look at your Zen speculation and it doesn't add up. Zen might suck, but its far more likely to be in the movement of data not the execution of it(much like bulldozer).
* excluding >128bit SIMD that most server and consumer software doesn't care about......
> mpx (mpx.delete@this.nomail.pl) on October 5, 2015 9:37 am wrote:
> > The 128-bit choice could also stem from having the same architecture for both x86 and ARM.
> > As ARM has no 256 bit vector processing, so a common denominator is 128-bit units.
>
> Intel has 2 read and 1 write ports to cache, same as all other high end CPU's, but Intel has three AGU
> units and thus can use all of those ports. For problems that fit in cache and need lots of bandwidth Intel
> will win. For much software this would just mean you become RAM bandwidth limited with fewer cores.
>
> Some games and other software where a primary thread is the performance limit would be an
> easy win for Intel. But for games AMD has better graghics to offset that to a degree.
>
> There are lots of other factors, many of which Intel will have a slight advantage
> in, AMD Zen was always going to be a small cut below Intel in performance.
>
> It looks like Zen will compete in the budget server market against ARM, and in the mid range laptop/desktop
> market. A big improvement from the budget laptop/desktop market sliver AMD has now. AMD might be back to
> struggling break even instead of going out of business. Back to a joke $7 stock. Sound reasonable?
>
So much like juanrga thinks all ALU ops are executed in 1 cycle, ALU's cant do mov's and registers don't exist.
I would point out since sandy bridge AGU and ALU has increased far more than the improvement in performance per clock* and that's with all the other core and uncore improvements. You could even possibly argue as far back as nehalem (1 load 1 store AGU), but there was a pretty big jump from it to sandy bridge but how much of that was cache overhaul, uop cache and other general improvements.
In the consumer K space intel had to release devils canyon to finally create an SKU that actually offered improvements over sandy bridge thax to sandys awesome overclockability. You consider these points and then look at your Zen speculation and it doesn't add up. Zen might suck, but its far more likely to be in the movement of data not the execution of it(much like bulldozer).
* excluding >128bit SIMD that most server and consumer software doesn't care about......
