Could Skylake present us a regression in single-threaded performance?

[lamedude]

Probably not the regression you were looking for.

In processors based on the Skylake microarchitecture, the functionality of the MMX instruction set is unchanged from prior generations. But many MMX instructions are constrained to execute to one port with half the instruction throughput relative to prior microarchitectures.

 

[jpiniero]

Well, apps shouldn't be using MMX. I'm sure there's lots else in there that isn't used much now either. Removing it completely would be ideal compared to gimping though.

 

[ShintaiDK]

Well, apps shouldn't be using MMX. I'm sure there's lots else in there that isn't used much now either. Removing it completely would be ideal compared to gimping though.

Still old code that cant run without MMX. So removing it would be very bad. While performance is irrelevant as such.

It would be nice if there was better control with code and progression so we can finally get rid of legacy. 16bit still reigns as well in some areas.

 

[cbn]

Well, apps shouldn't be using MMX. I'm sure there's lots else in there that isn't used much now either. Removing it completely would be ideal compared to gimping though.

You know Intel Quark doesn't have MMX.

....And I wonder if they plan on using that for a foundation of new CPU line?

Maybe even something built from the ground up to be integrated with iGPU. Then have instruction sets from the CPU extend into the iGPU essentially making CPU and iGPU inseparable.

 

[ShintaiDK]

Legacy support doesn't hinder any performance. And the transistor budget is ultra tiny for it. The cost associated is mainly validation.

We are not getting some kind of magic just because MMX, x87 etc gets removed.

 

[cbn]

It would be nice if there was better control with code and progression so we can finally get rid of legacy.

New hardware applications that don't rely on legacy software for a foundation won't need those old instructions.

And with a processor like Quark (as a starting point) maybe this is actually practical?