PR (Peformance Regression) #2172

[JMC47]

yeah, Mario Galaxy is a little bit slower for me as well, but it's something like 4 - 5%

[dephined]

Are you talking about OpenGL, D3d or both? Because for me OpenGL seems to have the same speed as before (even in Super Mario Galaxy), while D3d has become slower simply due to the fact that now vps are perfectly bound with fps (before vps where a bit higher, so that has reduced global speed)

[mimimi]

Yes, i was talking about D3D. It's tested with 3xIR, no AA, no AF, efb2tex:
4.0-5964: 102-111 fps=vps
4.0-5971:  86-102 fps=vps

In the situation described before: Mario Galaxy, at start of the game on the viewing platform overlooking the quare, lake and castle.

With 4.0-5964 i did get a momentarily lag once, where fps and vps desynced. I think i have seen this kind of lag in many games before and i'm yet to see it after the pr2172 merge. But still, the performance got worse in at least one situation. Not that i really want complain about it, but stating that there isn't a performance regression is just not true.

[dephined]

I just tested SMG with D3d (4.0-5954 against 4.0-5973) and I don't think the fps-vps bahaviour is completely the same: I mean in 4.0-5954 they seem to "refresh" with a slightly different timing (fps are alway just a little bit behind) while on 4.0-5973 they're perfectly tied togheter...other than that I really cannot tell wich version is faster, on the viewing platform they both go for me from 45 to 55 fps...anyway even if 4.0-5973 would be a bit slower only for this game, I don't think it would be a big problem, since OpenGL really seems to greatly oupferform D3d in SMG (on my system it's around 20% faster - in that "testing place" fps nevert get under 60)

P.S. I'm using an Intel e8400 and Geforce gtx650ti on Windows 8.1

[kirbypuff]

(04-06-2015, 09:30 PM)degasus Wrote: I hope the performance is fine for you if you'll use all 4 cores?

I tried in normal (quad-core) mode and with an underclocked GPU.

* Quad-core makes the experience much worse, with constant FPS fluctuations (from slow to extremely slow, then back to somewhat fast (still slower than master), and finally slowing down to a crawl)

* With an underclocked GPU, the slowdown is even worse:

NSMB @6xIR, EFB2RAM
--------------------------------
Master before PR #2172: 58 / 68 / 40 / 47
Master after PR #2172: 43 / 52 / 22 / 36

Now up to 80% slower.

[kirbypuff]

Still experiencing the same slowdown problem with the latest (official) dev. build.
Since PR #2172 was merged to master, the performance drop is extreme (25% ~ 80%)

The unofficial build by Tino (Ishiiruka), on the other hand, doesn't have this issue.

A test build based on the latest dev. build with an option to turn off / bypass the GPU thread power saving (PR #2172) would be nice.