(02-20-2013, 03:36 AM)etking Wrote: [ -> ] (02-18-2013, 09:42 PM)RachelB Wrote: [ -> ]Then either learn how to use your cpu, or find an OS that can properly manage turbo boost.
I already use process afinity but for many CPU models LTTC is always faster, regardless of how many real/HT cores are used or avoided. Removing LTTC made dolphin slower for me and all those mobile Corei7 Sandybridge people. The speed difference (1FPS) is not very big but is definitely there and valid and can clearly be measured objectively and reliably. And I will not reduce my internal resolution, the minor speed increase is just not worth the huge loss in image quality as I prefer to max out CPU and GPU equally.
Provide reproducible
data or I'm installing a plugin to ban you from the Development Discussion forums.
This is OT but what kind of data are you looking for? I reported my LTTC Benchmark data (Dolphin on screen FPS with and w/o using Fraps) on specific scenes in Zelda SS title screen and SMG Hub with and without process affinity set all the time during the discussion here and in the revision comments from the very beginning and also would like to report any further data if you can tell me what kind of data you need and how to measure it the best possible way. For example the only way to get the highest FPS in SMG (41) was to set LTTC on and set process affinity to avoid HT cores at the same time.
Even on older builds like 3.0-424 I did a lot of benchmarking back in the the day with the result that Zelda SS runs better and with higher FPS if LTTC is enabled. Please provide data that LTTC=on is slower on a mobile i7-26670QM found in many typical Dell configurations.
Also the data submitted by other people in these discussions (there was a graphical comparison somewhere) clearly showed that LTTC=on was faster on some of these machines.
And on my system the FPS is higher if I enable LTTC, even though the difference is minimal. I can perfectly live with LTTC being removed, since it seems to reduce performance for the majority of users (supposedly desktop users without HT-enabled mobile Core i7 CPUs) but the statement that LTTC is slower on every machine is just wrong.
I lost most of my older benchmark data but I can provide new benchmark results if needed since the benefit of LTTC= on is easy to reproduce on my system. However some factors like anti virus software, firewall, drivers, the 9+ active threads of Toshiba bluetooth stack and other unknown factors may have an impact on the results.
Record a movie (select the game, and then click emulation->start recording), do some stuff in game, stop emulation, and save the movie. Then play it back with log fps to file turned on (if using wiimote, you'll need to close and reopen dolphin each time for it to play back properly). This will ensure that exactly the same thing happens every time, so we can have a fair comparison. Once it finishes, stop the game, and you'll find an fps.txt file in /User/Logs. Make sure it runs for at least a couple of minutes, but the longer the better.
Anything else is a waste of time.
OK, I will be using Dolphin 3.5 official and the scenes will be
1) Zelda SS file select screen and Zelda SS outside of Sandhip
2) SMG hub with the map zoomed out
as these high CPU load scenes also benefit a lot from ICC enhanced builds.
But how can I playback the recording. If I use play recording and open the dtm file, in latest master it loads the state where I started recording and then nothing happens.
After a while, dolphin crashes, even it is has been restarted before starting the playback. In 3.5, dolphin crashes instantly when trying to playback a 10 second recording of SMG.
But I could create a savestate, empty the FPS file and let the game run for 30 seconds with no user interaction.
Don't start recording while the game is running. Just single click it, to select it, then start recording. The game will start automatically. You also can't use a real wiimote. 30 seconds is not long enough either.
Quote:But I could create a savestate, empty the FPS file and let the game run for 30 seconds with no user interaction.
This is not okay for wiimote games. Wiimote save states are bad and can't be trusted. It's not so bad with a gamecube controller, but still leaves room for user error.
As I cannot get the record function to run (dolphin 3.5 always crashes instantly when loading a dtm file) , I crated a save state, loaded it and let it run for 10 seconds. The results are exactly the same as in my previous benchmarks, where I just kooked at the FPS counter in specific scenes with no user interaction at all (like Zelda SS file select screen). No process affinity had been set in the below data:
SMG hub scene zoomed out, no user interaction LTTC off, 8 cores:
31
30
31
36
36
36
36
35
35
35
34
36
35
36
36
36
36
36
36
same state, LTTC on, 4 real, 4HT cores used:
30
31
36
37
37
37
37
37
36
36
37
37
37
37
37
37
37
37
LTTC Off, 4 real and no HT cores used for dolphin
35
36
36
37
37
37
36
37
37
36
37
37
37
37
37
36
36
LTTC on, 4 real and no HT cores used for dolphin
36
37
36
37
37
37
37
37
36
37
37
37
36
37
37
37
37
37
The best and fastest setting is LTTC=on + process affinity set to disable HT, however the lead is mimimal.
But from a usability perspective just setting LTTC=on would be much easier and more user friendly.
I did not benchmark 3 or less real cores as these are much slower. The benefit of LTTC is visible most in scenes with extreme slowdowns. To get back to the original topic, shoudn't we do such benchmarks for OpenMP Texture decoder as well (NSMB Wii has massive slowdowns with that function enabled)?
(02-20-2013, 08:08 AM)etking Wrote: [ -> ]As I cannot get the record function to run, I crated a save state, loaded it and let it run for 10 seconds.
Lol, why even bother?
And we don't care about your lttc off vs on comparisons. We already know lttc can give a speedup on cpus with turbo boost or HT. Compare lttc vs process affinity set appropriately.
So wait. You guys want people to manually setup the process affinity every time they want to run dolphin instead of just leaving an option turned on. Do I understand your logic correctly?
Yes, yes that is what they want.
(02-21-2013, 03:54 AM)AnyOldName3 Wrote: [ -> ]Yes, yes that is what they want.
No. You can easily configure it to be automatic.