Experimental Windows Builds

[Nach]

I made a couple of crazy changes in the hope to learn a few things which will lead to improvements later.

These changes effect how much CPU power ZSNESW requires, and the timing of NTSC and PAL games.

Please let me know if these builds are faster or slower on your PC than ZSNESW v1.40. Please also let me know if you notice anything broken which isn't in ZSNESW v1.40 (or fixed for that matter). Things to look for is games going faster or slower than they should, sound glitches. Game to sound sync.
Noticable changes could be in the Tales of Phantasia and Star Ocean intro.

Without further ado, the T and G experimental versions:
http://www.ipherswipsite.com/files/private/zsnesw_t.zip
http://nsrt.edgeemu.com/zsnesw-g.zip

For your status reports, specify for which version you used, how it differed, was it a PAL or NTSC game etc...

And please don't go posting these all over the net, it's only for people who want to test; things could be and probably are broken. v1.40 from the official page is better than these builds at the moment.

[kevman]

You couldn't move the first to another mirror, could you? I'm downoading at a couple hundred bytes a second.

I'll test the second a little for speed, as this computer is slow enough that Zsnes loads it (ancient sound card pulling this K6-2+ 500 down) to 100%

Alright, using the NTSC English translation of FF5, I have noticed one change so far:

At 500 Mhz: Approx 45 FPS for both builds
same machine at 200 Mhz: Approx 24 FPS for the offical release, 21 FPS for the G release.

Everything else acts the same so far. I'll keep checking.
Also, I can test any speed between those above at 50 Mhz increments. Should I? (I can change my CPU multiplier while running)

[Oblivion]

First test: 500Mhz Celeron, fastforward on. All 3 builds took exactly or very close to 35 seconds to run the intro to Star Ocean. More results to follow.

[Nach]

First test: 500Mhz Celeron, fastforward on. All 3 builds took exactly or very close to 35 seconds to run the intro to Star Ocean. More results to follow.

Good, although you should listen to the intro too to see if any of them break the voices.

[Oblivion]

I was planning on doing that next. I raised my CPU multiplier on the Celeron machine from 2x (wtf?) to 8x, and the results are as follows:
T Build: 35 Seconds still.
G Build: 34 Seconds
1.40: Barely under 34 Seconds.

-------
Having listened to the entire intro, I have found that the G build had two small sound anomalies. Both were something of a static sound, the first was during the shockwave of when whatever planet is blown up. The second is in a fade-out from the ship to when the ship goes through said shockwave.
Settings: Stereo, 32000Hz, Gaussian Interpolation.
The bug occurred upon a restart of the emulator and such. The sounds are not the same.

-------
In Megaman X (Only Intro and Chill Penguin stages), I would say that the T build's Chill Penguin music sounded very slightly different, for perhaps 10% of the track. It just sounds somewhat off-tone or something, however, it is not a particularly significant difference. It probably wouldn't sound odd to someone playing the game for the first time. The G build sounds the same as 1.40.

-------
The Tales of Phantasia intro sounds the same in all three builds.

[Nach]

Thanks. Although some of that sounds weird to me, guess I'll have to look over my code more.

Can other people also test please?
It's also important people test PAL games.

[Oblivion]

I can test a couple of PAL games, but I don't have many. I should be able to test the speed of a 100Mhz machine shortly.
-------
I can't find the old 100Mhz laptop. I let Terranigma run through the intro, then started a game and didn't see or hear anything messed up in either the T or G builds.
-------
While running the Task Manager to see CPU time, I noticed the following ranges of CPU use, running F-Zero 2, in each build:
1.40 - 22-38%, Most frequently around 30.
T Build - 19-36%, Most frequently 30-34%
G Build - 13-30%, Most frequently in low 20s.
These tests were conducted with a 2.6GHz P4. Overall, it clearly appears that the G build uses less processor cycles, and the T build generally uses the same amount.
-------
Star Ocean Intro, CPU Use:
1.40 - 19-38%, Still mostly lingering around 28-30.
T Build - 19-38%, Greater fluctuation than 1.40, Average 30-33%
G Build - 8-31%, Mostly within range of 16-22%

Sidenote on sound in Star Ocean intro: The second sound anomaly that I mentioned previously occurs in all builds, but I had only noticed it in the G build prior to now. The first, has stopped occurring, although I heard it multiple times last night.

[Nach]

Looks like I'm going to need to make some more experiments soon.

When playing a game, say a fighting game, do the games seem to go faster or slower? I need both NTSC and PAL games tested.

[Nach]

At 500 Mhz: Approx 45 FPS for both builds
same machine at 200 Mhz: Approx 24 FPS for the offical release, 21 FPS for the G release.

Everything else acts the same so far. I'll keep checking.
Also, I can test any speed between those above at 50 Mhz increments. Should I? (I can change my CPU multiplier while running)

At 500 MHz, how much speed does the official give you? And what does the T release give you at 200 MHz?

Also please test a PAL game.

200, 300, 500 MHz speeds seem to be good #s to get an idea of what my changes do.

[washuu108]

These are the average ranges of CPU usage using a 866MHz P3.

Super Bomberman 2 (NTSC)
1.40 - 79%-82%
T - 79%-82%
G - 75%-77%
Battle Mode, 1 human, CPU level 9, stage 1. (Other settings default) Tested through one round.

Yoshi's Island (PAL)
1.40 - 66%-69%
T - 66%-69%
G - 62%-66%
Sitting at the stage select screen immediately after the intro stage.

On both tests the T build sometimes showed an extra percentage point compared to 1.40, but overall the difference was negligible.

[Nach]

washuu108: Did the games seem like they played a bit faster or slower than usual in any of the special builds?

[washuu108]

Nope, timing has been the same in all three. Haven't noticed any problems with sound or graphics, either.

[Nach]

Please do all the same tests with this new build:

The I Build: http://nsrt.edgeemu.com/zsnesw-i.zip

[Oblivion]

I Build
-------
Star Ocean intro, 500Mhz Celeron, 2x multiplier: 34.5 Seconds
Since the multiplier increase barely was significant, I'm skipping that one.
Star Ocean intro sounded normal, and the first sound anomaly appears to be random, as it appeared again. I can't find any patterns with it.
Chill Penguin stage sounds normal.
ToP intro stuff sounds normal.
Terranigma intro stuff sounds normal.
F-Zero 2 - 11-30% CPU, mostly between 17 and 23 inclusive.
Star Ocean intro - 8-32% CPU, still mostly around 20.

[kevman]

At 500 Mhz: Approx 45 FPS for both builds
same machine at 200 Mhz: Approx 24 FPS for the offical release, 21 FPS for the G release.

Everything else acts the same so far. I'll keep checking.
Also, I can test any speed between those above at 50 Mhz increments. Should I? (I can change my CPU multiplier while running)

At 500 MHz, how much speed does the official give you? And what does the T release give you at 200 MHz?

Also please test a PAL game.

200, 300, 500 MHz speeds seem to be good #s to get an idea of what my changes do.

At 500 MHz, the T, G, and official builds all give the same FPS
At 200 Mhz, The T and G both give me 34 this time, with the official at 38.

Shit, I just realized that my sound card isn't working... I gotta fix that.

[Nach]

kevman: And how does the new I build stack up?

[kevman]

kevman: And how does the new I build stack up?

@500Mhz: The I build jitters back and forth between 47 and 51 FPS. The others all give a fairly consistent 50.

@200 Mhz: The I build switches back and forth between 33 and 35.



I also tested at 300 Mhz:
The G build gives me 40 FPS as well as the official. T gives 43-45.
I gives exactly 42. All really too close together to use.

I'm gonna re-enable my sound card now, I think. I'm gonna get a PAL game then.

[washuu108]

Since I had to redo the O, T, and G tests anyway, (couldn't match old values) I took better care this time in making sure the conditions were the same for each build, such as using a movie for testing instead of playing myself, and repeating each test. Also took some steps to reduce overall CPU usage compared to the last tests.

Test #1 - Super Bomberman 2 (NTSC) - Same conditions as before, just using a ZMV of the same battle for all builds.

Code: Select all

 ___ _________ _________ _________ _________
|Ver|   1st   |   2nd   |   3rd   |   Avg   |
| O | 58%-61% | 58%-61% | 57%-61% | 58%-61% |
| T | 58%-62% | 58%-62% | 58%-62% | 58%-62% |
| G | 55%-58% | 55%-58% | 56%-58% | 55%-58% |
| I | 55%-58% | 55%-57% | 55%-57% | 55%-57% |
 ¯¯¯ ¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯

Test #2 - Yoshi's Island (PAL) - ZMV of gameplay from level 1-1. First number is CPU usage at a certain point in the movie, second is peak usage up to that point. (This test was less accurate and really needed 3 trials, since the range was a lot bigger than the Bomberman one)

Code: Select all

 ___ _________ _________ _________ _________
|Ver|   1st   |   2nd   |   3rd   |   Avg   |
| O | 60%,92% | 71%,90% | 57%,92% | 63%,91% |
| T | 62%,92% | 59%,90% | 56%,90% | 59%,91% |
| G | 59%,81% | 60%,77% | 60%,74% | 60%,77% |
| I | 59%,76% | 60%,76% | 61%,79% | 60%,77% |
 ¯¯¯ ¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯

Edit: Forgot, still haven't noticed any differences in timing between the builds, or any other problems.

[Nach]

Edit: Forgot, still haven't noticed any differences in timing between the builds, or any other problems.

Can you try the intros to Star Ocean and Tales of Phantasia and listen for sound problems?

Excellent tests, thanks

If the next results are positive I may only have two more experimental builds with should offer some speed increases for slower machines.

[washuu108]

Didn't notice anything at all in Star Ocean, and whatever crackling/static I could manage in the experimental builds with Tales of Phantasia were in 1.40 too.

[daft]

Nach, has there been any new progress on the optimized builds? Are these meant to only show a noticeable difference for say, under 500 mhz machines or does it improve performance across the board?

[Nach]

Changes are in CVS.