Dev version (vNext) performance difference with v17?

Support and general discussion.
Post Reply
eddman
Posts: 74
Joined: Wed 09 Feb, 2022 9:30 am

Dev version (vNext) performance difference with v17?

Post by eddman »

I recently downloaded the vNext 19051875 32-bit build. It seems the emulation performance is a bit lower compared to v17.

ASUS P55T2P4
Pentium MMX 200
256 MB RAM
Voodoo 3 2000 / Speed: Default / Bilinear: ON / Screen filter: OFF / Threads: 4 / Recompiler: ON (vNext: Dither subtraction: OFF)
Sound Blaster 16
IDE HDD / VHD
Windows 98 SE

To eliminate as many variables as possible, I copied over the config and nvr files and also loaded the same HDD file. The windows' power plan is on "High performance".

These shots were taken about a minute after booting into windows and idling at the desktop.

Are these later builds perhaps more accurate, hence heavier, or maybe there is an active debugging layer?

P.S. The 64-bit build is even more slower.
Attachments
vnext-19051875.png
vnext-19051875.png (125.28 KiB) Viewed 2203 times
v17.png
v17.png (98.01 KiB) Viewed 2203 times
User avatar
SarahWalker
Site Admin
Posts: 2054
Joined: Thu 24 Apr, 2014 4:18 pm

Re: Dev version (vNext) performance difference with v17?

Post by SarahWalker »

Release builds are/have been built with profile optimisations. Dev builds are not. That would account for much if not all of the difference.
User avatar
leilei
Posts: 1039
Joined: Fri 25 Apr, 2014 4:47 pm

Re: Dev version (vNext) performance difference with v17?

Post by leilei »

The flag I usually use for self-compiled WIP builds is -flto. It's a little more obscure to set in a cmake makefile (imo) but I haven't noticed any sharp drop of performance from a vNext -flto build (built on GCC 11) compared to a pre-cmake -flto build (built on GCC 4.7).

Also logging in a non-release build can eat away at performance too.
Post Reply