Am 16.02.2017 um 03:21 schrieb dick balaska:

> So the i7 is a wee bit faster.  Also, 4 cores means the chip can do 4
> things at once.  The i7 has 2 threads per core; it can sort-of do 8
> things at once, but 2 threads share each FPU, I expect waiting for the
> FPU to be involved.

I think it is realistic to expect hyperthreading to give about 15% more
pixels per unit of time.


> But those scenes with BorderChars, the i7 is actually *slower* than the
> i5.  I would think even if a thread is always waiting on the FPU, it
> should still be minimally 6% (34/32) faster than the i5.
...
> I can't imagine anything you could do different. It's just SDL.  The
> issue has to be deep in the optimization code of the M$ compiler.
> Or ... I blame clipka.

:)

I tend to blame it on different cache sizes. Or different number of CPU
clock ticks required to execute particular instructions. Or differences
in the CPU clock. Or in the branch prediction. Or in the instruction
pipeline depth. Or... well, I guess you get the picture: CPU performance
is an utterly complicated topic.

Post a reply to this message