Am 25.03.2017 um 19:14 schrieb Mike Horvath:
> On 3/25/2017 11:30 AM, clipka wrote:
> while on modern Windows systems we expect a noticeable
>> gain.
> 
> 
> Compared to when? And by how much?

To quote my full sentence:

"Compared to 3.7.0, on Unix systems we expect to be roughly back at the
same performance, while on modern Windows systems we expect a noticeable
gain."

So that's compared to 3.7.0.

The performance gain varies somewhat with CPU model and, to an even
greater extent, with the scene to be rendered, but I'm seeing just shy
of +30% more pixels per second on my Intel i7-5820K machine for the
benchmark scene.

On AMD CPUs(*), expect a significantly lower gain (my guess would be
about +10% instead). This is because 3.7.0 already contained
optimizations specifically for AVX/FMA4-capable CPUs (benefiting AMD
CPUs only), which 3.7.1-beta.3 complemented with optimizations
specifically for AVX2/FMA3-capable CPUs (benefiting Intel CPUs).

(*AMDs new Ryzen is an exception to this rule, as AMD seems to have
dropped FMA4 support on that CPU in favour of FMA3.)


The newest speed increase of 3.7.1-beta.5 over 3.7.1-beta.4 is from
amending code I had introduced earlier in 3.7.1 development as part of
the refactoring work, which turned out to be a bit of a performance hog;
so these changes don't give a performance improvement over 3.7.0, but
rather counteract performance losses, so that the performance
improvements already introduced in 3.7.1-beta.3 can now live up to their
full potential.

Post a reply to this message