2.3.3 vs 2.4.0 speed

[mingle]

Looking good...

The only issue I currently have with WinAUE 2.4.0 is the speed decrease, compared to 2.3.x - about 18-25% slower.

Is this due to the extra code for filtering and the VSync stuff?

Cheers,

Mike.

[Toni Wilen]

Originally Posted by mingle

Looking good...

The only issue I currently have with WinAUE 2.4.0 is the speed decrease, compared to 2.3.x - about 18-25% slower.

Is this due to the extra code for filtering and the VSync stuff?

Cheers,

Mike.

This 2.4.1 only thread so technically your question is in wrong thread.

Anyway, you won't get any useful replies without much more information, like used configuration and logs (from both versions) and how do you notice the slowdown?

I forgot, I can give one sort of useful reply: Buy a better PC

[AmigaSurfer]

Originally Posted by mingle

The only issue I currently have with WinAUE 2.4.0 is the speed decrease, compared to 2.3.x - about 18-25% slower.

Is this due to the extra code for filtering and the VSync stuff?

Originally Posted by Toni Wilen

This 2.4.1 only thread so technically your question is in wrong thread.

Anyway, you won't get any useful replies without much more information, like used configuration and logs (from both versions) and how do you notice the slowdown?

I forgot, I can give one sort of useful reply: Buy a better PC

My PC is fast enought and I can confirm the slowdown with 2.4.0. Booting 2.4.0 is faster than 2.3.3. But starting YAM 2.7 needs definite more time with 2.4.0. I've done four tests with 2.3.3 and three tests with 2.4.0.

      Booting   Starting   SysInfo 3.24
      WinUAE    YAM 2.7    Mips   MFlops
2.3.3   16.6s     4.7s       755    794
      12.6s     4.6s       766    800
      24.2s     4.7s       734    799
      24.9s     4.7s       753    794
2.4.0   10.9s     8.0s       653    744
      10.7s     7.3s       639    737
      10.8s     7.2s       631    731

The logs are attached.

[Toni Wilen]

(moved from beta thread and just not because of 2.4.0 but because this really needs some more comments)

Does latest 2.4.1 beta change anything? It has some fastest possible timing/syncronization updates.

[lifeschool]

On speed issues. You may remember I sent you a speed comparison on the game Vroom some time ago - so here's an update. All four setups shown in the following video use the exact same config; the only thing I changed was to turn cycle exact on and off. My config (as seen at the end of the video):

A3000 + 030 25mhz - 2meg slow + 8meg fast
No JIT
Compatible Settings
50HZ full screen mode, both H and V centering, triple buffering.

This is the video (pitty EAB doesn't allow embeded videos)
[ Show youtube player ]

As you can see, 2.0.1 was the fastest and smoothest - the fastest to operate WB and the fastest to load the game. Booted WB in 3 seconds.

2.3.1 was noticably slower in all departments.

2.4.1b7 - cycle on - This booted in 3 seconds and ran WB fine, but as you can see it ran the game at sub A500 speeds - like 4mhz equivalent or something. Really slow to do anything - even though the CPU was set to Custom=25Mhz. But it still booted WB in 3 seconds.

2.4.1b7 - cycle off - this brought the game back to full speed again, quite slow to boot WB - 6 seconds, and slower to load the game, but ran even more quickly than the old 2.0.1 version in this mode. (You can see the video overtakes the 2.0.1 footage right at the end, meaning it's even faster! - BUT this is most likely due to the CPU being set to 25.3MHZ rather than just 25.000 - so they are practically the same speed).

Note: unlike the previous versions shown in the top two panels, 2.4.1b7 booted WB with a thick black border to the right and below - and this was the same on the game screen too. The WB and game screens were definitely off center by at least an inch to the left - coincidientally the same offset as the black border - without the border it would have been centered I guess.

It would be great to know the standard speed of a game like this on a real A3000 25mhz Amiga so that we know what speed the emul should be aiming for. No version of the emulator is the 'best of all worlds' for speed except the much more streamlined 2.0.1. Sorry to say that's the version I'm still using even now.

[Toni Wilen]

Use of single chipset based game to test speed does not make any sense because it is not just CPU but game specific CPU and chipset interaction that makes the end result.

CPU may be 1000MHz but it still may look like 1MHz if program is mostly waiting for some chipset state (like blitter) to change.

Speed testing is difficult, sometimes slow speed (that looks too slow) is actually the correct behavior. Unless using fastest possible CPU mode.

[wanderer]

Not sure how to measure this, but I've noticed that in 2.3.3 'Turbo speed' mode moves extremely smoothly while in 2.4 it moves in 'blocks' (the way a DVD player will behave if the disc is reversed in very fast speeds).

[lifeschool]

Tony: I understand that, but with cycle exact ON in my A3000 config the game is simply unplayably slow - and slower than an A500. Thats not right. There must be something a-miss. Yet at the same time I know trying to get every single game to play fair would be a difficult task. As I say, I have 1000% confidence in 2.0.1 -everything is centred and every game plays at the speed I remember from my youth.

[Toni Wilen]

Originally Posted by wanderer

Not sure how to measure this, but I've noticed that in 2.3.3 'Turbo speed' mode moves extremely smoothly while in 2.4 it moves in 'blocks' (the way a DVD player will behave if the disc is reversed in very fast speeds).

It happens if triple buffering and D3D mode is enabled. Triple buffering uses display driver vsync wait (to guarantee no tearing) and it can't be turned on/off on the fly so winuae tries to compensate and skips lots of frames.. (afaik 2.3.3 should also do the same)

Annoying Direct3D limitation.

[Toni Wilen]

Originally Posted by lifeschool

I understand that, but with cycle exact on in that config the game is simply unplayably slow - and slower than an A500. Thats not right. There must be something a-miss.

Ah, 68030.. 030 didn't even have cycle-exact mode until recently, it surely wasn't in 2.0.1 (Or worked totally different) so that at least can explain the speed difference.

030 cycle-exact mode isn't really cycle-exact either. Does 020 have the same slowdown too?

[lifeschool]

Cycle Exact was included in 2.0.1 but I remember reading recently that you completely changed how that works now.

But I don't understand this.. Now when I turn cycle exact off it runs the game at almost twice the speed I was getting before - Like A4000 speeds, even though the setting is Match A500 / A1200 or Cycle Exact. Now the game plays at almost the right speed with Cycle Exact ON (must have been something to do with my screen grabber interfering with things before - sorry). Tried 020 - very smooth, very fast, matches full game speed as it should. Then the mouse pointer began to hang on WB - and WB would begin to freeze randomly. Now I'm totally confused! Bug testing is a real headache it seems. I think I'll stick to my own thing and shut up.

[AmigaSurfer]

Originally Posted by AmigaSurfer

My PC is fast enought and I can confirm the slowdown with 2.4.0. Booting 2.4.0 is faster than 2.3.3. But starting YAM 2.7 needs definite more time with 2.4.0. I've done four tests with 2.3.3 and three tests with 2.4.0.

Originally Posted by Toni Wilen

Does latest 2.4.1 beta change anything? It has some fastest possible timing/syncronization updates.

Yes. I used the same config file with WinUAE 2.4.1b7 and starting YAM is fast again. See table.

         Booting   Starting   SysInfo 3.24
         WinUAE    YAM 2.7    Mips   MFlops
2.3.3     16.6s     4.7s       755    794
          12.6s     4.6s       766    800
          24.2s     4.7s       734    799
          24.9s     4.7s       753    794
2.4.0     10.9s     8.0s       653    744
          10.7s     7.3s       639    737
          10.8s     7.2s       631    731
2.4.1b7   10.9s     4.4s       647    737
          10.8s     4.3s       665    739

[Toni Wilen]

Great. Does 2.4.1b8 work better or worse than b7?

[AmigaSurfer]

Originally Posted by Toni Wilen

Great. Does 2.4.1b8 work better or worse than b7?

Sorry for the late answer.
Booting WinUAE and starting YAM needs the same time with 2.4.1b8. But the SysInfo results, especial Mips, are slower since beta 8.

         Booting   Starting   SysInfo 3.24
         WinUAE    YAM 2.7    Mips   MFlops
2.3.3     16.6s     4.7s       755    794
          12.6s     4.6s       766    800
          24.2s     4.7s       734    799
          24.9s     4.7s       753    794
2.4.0     10.9s     8.0s       653    744
          10.7s     7.3s       639    737
          10.8s     7.2s       631    731
2.4.1b7   10.9s     4.4s       647    737
          10.8s     4.3s       665    739
2012-04-23:
2.4.1b7   11.0s     4.5s       668    740
          10.8s     5.0s       629    735
2.4.1b8   11.2s     4.5s       301    637
          11.3s     4.2s       278    627
2.4.1b9   11.3s     4.2s       303    636
          11.1s     4.2s       268    629
2.4.1b10  11.0s     4.2s       308    640
          11.1s     4.3s       300    641

[Toni Wilen]

I am quite sure SysInfo results are total bogus (or it wraps around) because I usually get from 230 to 400 MIPS depending on how fastest possible CPU time is shared and my CPU is 4.6GHz i7 2600k. There is no way you can have 2x-3x bigger results

"A4000 68040 25MHz" ratio appears to work better, it has always increased after I have gotten newer CPU. (175-180 now, not much difference between WinUAE versions either)

Could you do your tests using above SysInfo value?

Anyway, stop watch and timing some long operation is the only way to get sane results, most Amiga benchmarks either lose accuracy completely or timing counters wrap around (multiple times!) if CPU speed is "impossible" from 1990s point of view..

[zipper]

Originally Posted by Toni Wilen

I am quite sure SysInfo results are total bogus (or it wraps around) because I usually get from 230 to 400 MIPS depending on how fastest possible CPU time is shared and my CPU is 4.6GHz i7 2600k. There is no way you can have 2x-3x bigger results

Well, running on Amiga Forever, 720QM @ turbo, about 2.9 GHz, Sysinfo gives 689 Mips for me; v2.4.0. on default settings.

[AmigaSurfer]

Originally Posted by Toni Wilen

I am quite sure SysInfo results are total bogus (or it wraps around) because I usually get from 230 to 400 MIPS depending on how fastest possible CPU time is shared and my CPU is 4.6GHz i7 2600k. There is no way you can have 2x-3x bigger results

The Phenom II X6 1090T 3.2 GHz has twice as much L1 cache and twice as much L2 cache than the i7 2600k. :-)

Originally Posted by Toni Wilen

"A4000 68040 25MHz" ratio appears to work better, it has always increased after I have gotten newer CPU. (175-180 now, not much difference between WinUAE versions either)

Could you do your tests using above SysInfo value?

        SysInfo 3.24
        A4000 68040 25 MHz
2.3.3      93
           94
2.4.0      87
           86
2.4.1b7    86
           86
2.4.1b8    68
           68
2.4.1b9    68
           68
2.4.1b10   66
           66

Same trend than Mips and MFlops. But WinUAE boots fast and YAM starts fast. This is the main thing. :-)

[Toni Wilen]

It is now 100% confirmed SysInfo MIPS calculation is totally broken, at least if CPU is too fast. Laptop with Q720 i7 shows 600-700MIPS and 60-90 A4000 speed result..

Anyway, I managed to get SysInfo A4000 result back to 2.3.3 range (~90, was ~60 previously) when using above laptop and limiting max extra CPU emulation time per scanline.

Not sure if it makes any real-world difference or slows down something else, please test once more, thanks. (http://www.winuae.net/files/b/winuae.zip)

[zipper]

Yes, confirmed; running @933 MHz I still got 555 Mips.

[AmigaSurfer]

Originally Posted by Toni Wilen

It is now 100% confirmed SysInfo MIPS calculation is totally broken, at least if CPU is too fast. Laptop with Q720 i7 shows 600-700MIPS and 60-90 A4000 speed result..

Anyway, I managed to get SysInfo A4000 result back to 2.3.3 range (~90, was ~60 previously) when using above laptop and limiting max extra CPU emulation time per scanline.

Not sure if it makes any real-world difference or slows down something else, please test once more, thanks. (http://www.winuae.net/files/b/winuae.zip)

I updated AMD Catalyst from 12.3 to 12.4 in the meantime. So I've made fresh tests.

                 Booting   Starting       SysInfo 3.24
                 WinUAE    YAM 2.7    Mips   MFlops   A4000
2.3.3             24.5s     5.8s       774    800      93
                  25.0s     9,9s       768    802      93
2.4.0             10.9s     6.5s       656    731      87
                  10.6s     8.0s       649    742      87
2.4.1b10 23.04.   11.2s     4.3s       284    639      67
                  11.1s     4.2s       293    634      68
2.4.1b10 25.04.   11.2s     4.3s       392    643      73
                  10.9s     4.5s       356    637      71