GITS SAC performance regression?
#1
There seems to have been a performance regression between versionĀ 1.3.1-799-g000e11e and 1.3.1-898-g68a89bb in DX11 software mode for Ghost in the Shell: Stand Alone Complex PAL.

Using 1.3.1-898-g68a89bb I get ~22fps (EE 40% GS 95%). One CPU core is at 100% while the others are at 50% utilization.


Using 1.3.1-799-g000e11e I get [b]50fps[/b] (EE 90% GS 60%). All CPU cores have an equal load.

I tested both versions with default settings, except 'extra rendering threads' at 2 extra threads.
I tested ZoE 2 but there seems to be no difference in performance between the two versions.
Reply

Sponsored links

#2
Can you test any builds on between to try and narrow down where the problem lies? Thanks
[Image: ref-sig-anim.gif]

Reply
#3
(08-06-2015, 06:35 PM)refraction Wrote: Can you test any builds on between to try and narrow down where the problem lies? Thanks

1.3.1-872-g42f5159 tanks performance for me.
Reply
#4
Well that's one accuracy upgrade by gabest. Often higher accuracy is more demanding. Can you open a new issue in github repeating what you said there referencing this commit and including "@gabest" ? Possibly it will still stay as it currently is since accuracy has normally the higher priority over speed.
Reply
#5
(08-06-2015, 08:11 PM)willkuer Wrote: Well that's one accuracy upgrade by gabest. Often higher accuracy is more demanding. Can you open a new issue in github repeating what you said there referencing this commit and including "@gabest" ? Possibly it will still stay as it currently is since accuracy has normally the higher priority over speed.

Done. I'm not expecting anything but that's a rather steep increase. If I'm extrapolating correctly (maybe I'm not) you'd need more than twice the power. That's harsh.
Reply
#6
Perfect!

Since the decrease is such dramatical I would expect either a user configured hack that dis/enables the pr or the new code will be optimized.
Reply




Users browsing this thread: 1 Guest(s)