Old PIII for MAME, AGP card doesn't help

[TheManuel]

I have an old PIII 800MHz that I want to use for a MAME cab and obviously it is pretty slow.  One of the things I did to try and help it along was to add an old AGP 2x video card to see if it helped.  The results is that in umk3, the fps went from 30 to 28.
I was hoping that adding a video card would actually help it, not make it worse since, as I have seen often quoted, integrated video "steals resources away from the computer".  I am starting to believe this is BS.

I had the exact same thing happen with an even slower computer (Celeron 700MHz) which did not actually have AGP.  I installed a PCI video card and the performance went down instead of up.  Somebody in a MAME forum told me it was because of the PCI interface but now, years later, I try the same trick with an AGP card (granted, 2x) and the result is similar.

Does anyone have any idea about what is going on?
Am I missing something?

Thanks.l

[TOK]

You didn't say what version of MAME you're using. I have a bartop running on a P3/933. The best way to improve performance is to run an older version of MAME. I'm running .84 and it runs great. If you're running anything from .100 on, you're never going to get good performance out of it.

[MaximRecoil]

I don't know why your performance decreased with the video card as opposed to onboard video, but I would suggest that you run a version of MAME that is from the same time frame that your processor was current in. Something from the .5x to early .6x range should be about right.

[ahofle]

Did you disable the onboard video in the bios after you installed the AGP card? 

Anyway, you aren't going to pick up much in terms of MAME performance (currently) with a better video card.  Try an older version of MAME as TOK suggested.

[TheManuel]

Thanks guys.
That is my current plan, to find an older version of MAME for those games that don't run at full speed and the latest version for all the rest to take advantage of the discrete sound improvements of recent releases. 
I guess for CPS-3 I am SOL  .

It still bugs me that a dedicated video card would lower my performance.  It just does not make sense.  I know MAME does not make much use of the video card but at the very least I would have expected the system to be less burdened.  The first thing I did was to see if the onboard video could be disabled through BIOS but it is an HP computer with the unfriendly BIOS.  Oh well.

[Fozzy The Bear]

It still bugs me that a dedicated video card would lower my performance. 

Sounds more like a drivers issue. Either poor video card drivers or the wrong video card drivers or you installed them over the top of the previous drivers and there's still some junk in the system screwing with them.

Best Regards,
Julian (Fozzy The Bear)

[Dervacumen]

Are those benchmark numbers from single runs, or are they averages of a few runs?  If you only benchmarked one time, it could be that there was a random background process running while you were running the benchmark after you installed the card.

[TheManuel]

I could not count how many times I have run it and the result is pretty consistent.
For consistency, I am checking the FPS right after I insert a coin (so to speak).

[xmenxmen]

Thanks guys.
That is my current plan, to find an older version of MAME for those games that don't run at full speed and the latest version for all the rest to take advantage of the discrete sound improvements of recent releases. 
I guess for CPS-3 I am SOL  .

It still bugs me that a dedicated video card would lower my performance.  It just does not make sense.  I know MAME does not make much use of the video card but at the very least I would have expected the system to be less burdened.  The first thing I did was to see if the onboard video could be disabled through BIOS but it is an HP computer with the unfriendly BIOS.  Oh well.

To add more desire for you, even my Athlon 1800 is not fast enough for that.  My P4 2.4 is.  So this would give u a decent idea of what you need. 

Also remember, mame is completely cpu dependent, so gpu will have no effect even if you had a 8800GTX, it's all cpu power when for the video processing.

[knave]

I would second the driver issue, there are too many variables that could afect this.  Also some onboard video is AGP, it's just hard wired on the motherboard.  I have a few cards that are like this. (they don't even have an AGP slot)

I also wouldn't worry about the 2 FPS difference.

I don't know what you want to run but I play most games full speed with an Athlon 800 with 256mb ram on a pretty recent release(.113). Granted with a good video card but it shouldn't make a difference.
By most games I mean classics through most neo-geo to just before street fighter 2.

So...to echo above...don't spend any more time on it than you have to.  Yse an earlier release. 

[TheManuel]

The AGP being hard wired in the motherboard makes a lot of sense.  After all, why create a new interface when you can take advantage of the existing AGP port.  That is probably what is going on here.
This setup should run most games full speed through CPS-2 with relatively recent MAME versions.  Like I said, I will use older versions for the tougher cookies.

Thank you everyone.

[SirPeale]

Also, try lowering your resolution.  Mame doesn't like higher res.

[TheManuel]

I have a bunch of ini files with different resolutions and refresh rates to get all games looking right on a PC monitor.  For the more taxing games, I use lower resolutions.
You are right.  With older computers, higher resolutions take quite a hit.

[u_rebelscum]

What video card did you add in?  What was the onboard chip set? (or motherboard if you don't know the chip set.) 
Some of the really old cards were slower than the at-the-time CPUs doing the work on at-the-time PC games; some even used the CPU more than some onboard chips.  Any of these cards would be even worse with mame.  (You probably aren't using one of these since the drop is so minor, but maybe the next generation?)

There are many many points that might be at work with the onboard vs AGP card:

a.  Mame currently favors directX 9.0 tuned chips & drivers with mame's default settings.  If your card is only dX 8.0, mame automatically drops down to use that (but it's not as good AFAIK).  Worse is if your card says it does dX 9.0, but does it poorly; on these cards it might be better to set mame to use 8.0 (or see next point).

b.  Mame currently favors direct3D tuned cards with mame's default settings.  If your card does directDraw better than direct3D 9.0 and 8.0 (see above point), set mame to use dd instead of d3d.

c.  OTOH/Also, Mame now defaults to use windows desktop's resolution.  If your card isn't powerful enough for that res with d3d or dd (see above two points), enable switchres and set mame to use a lower resolution (one that your card can use well).  (This point is an expansion of Peale's post.)

d.  Mame uses only a few parts of d3d, so if the card is better than the onboard chip at stuff mame doesn't use, it's of no use.  (Ditto with dd if you use that.)

e.  A few onboard chips are pretty good, and even fewer steal little to no resources.  Rare, but it's possible your onboard chip really is better at d3d @ your resolution.  (IOW, you shouldn't treat generalizations as absolute truth.)

f.  As mentioned by others, video drivers might be at work, and effecting all the above five points.


I know I missed other points, but.... 2 fps is less than 7%, and with all the variables going on, that's not outside the range of error IMO.  If you ran benchmarks a few times for, say, 120 seconds each (mame -str 120...) and the change was over 5%, I'd agree something is going on.

In the end, you'll get higher speed increases with a fast CPU.  Sometimes an earlier version of mame is faster, but not always (again, generalizations can be not true, sometimes).


BTW, discrete sound emu/sim can take a lot of CPU power.  You should be okay, but don't be too surprized if your CPU isn't fast enough for a few discrete sound games.


BTW2:

Also remember, mame is completely cpu dependent, so gpu will have no effect even if you had a 8800GTX, it's all cpu power when for the video processing.

Not exactly true. 
Mame is CPU depentdent on emulation, so GPU will not help on emulation.
Mame, however, can and does use the GPU for none emulation stuff, like convert the original res to the displayed res, artwork, scanlines, ect.  This is even more true now than it was years ago.
So a slow GPU can slow down mame, but a super-fast GPU will not be faster with mame than a good-enough GPU.

[TheManuel]

Lots of intersting stuff there, u.

The MB southbridge chipset is VIA VT82C596 and the onboard video is NVidia Riva TNT2.  This is an HP PIII 800MHz that was donated to me by co-workers for my arcade project along with an Athlon T-Bird 800MHz.  The video card came from the latter and is a GeForce2 MX/MX 400 (powered by).  Coincidentally, the Athlon PC runs the game at the same speed as the PIII when they both have the same video card.  The Athlon's MB does not have integrated video.

I'll address your comments by letters:

a. I don't know if even DX 8.0 was available when these products were made 

b. I will play with switching back between D3D and DDraw.  I always use DDraw thinking it would be faster by default.

c. I am sure I am switching resolutions to 600x480x60Hz when running the benchmark (the monitor tells me from the OSD)

d. I was hoping that simply having the card, the CPU would be freed of some load but maybe all I'm freeing is the memory of which the computer has plenty enough for that game (512MB)

e. Again, I have been trying this with DDraw but I could very well be a case of a very crappy card being outperformed by onboard video

f. Drivers are always a possibility.  I went from scratch and uninstalled all the old drivers for both, onbard and AGP card and then installed fresh the latest drivers for that card and the result was the same.

As far as a margin of error for the benchmark, it would be less than 1FPS in this case.  I always start the game, insert coins and press P1 to go to the character selection screen.  10 out of ten times I get the same number of FPS in each case so it is pretty clear that the total variability is dominated by the video cards and not the measurement system.

I have always been aware that MAME is mostly CPU dependant.  By adding the card I was simply hoping to give it a little boost.  I can't count how many people give the advice of buying a separate card rather than onboard video to "not steal away from the CPU".  I am pretty convinced by now that this is rubbish.  Video cards are for PC games, video rendering etc...

Thanks for all your comments.

[u_rebelscum]

The MB southbridge chipset is VIA VT82C596 and the onboard video is NVidia Riva TNT2.  This is an HP PIII 800MHz that was donated to me by co-workers for my arcade project along with an Athlon T-Bird 800MHz.  The video card came from the latter and is a GeForce2 MX/MX 400 (powered by).  Coincidentally, the Athlon PC runs the game at the same speed as the PIII when they both have the same video card.  The Athlon's MB does not have integrated video.

Considering the tnt2 was a great (at it's time) video card, yes card, you have one of the few video card quality onboard video chips, so....

d. I was hoping that simply having the card, the CPU would be freed of some load but maybe all I'm freeing is the memory of which the computer has plenty enough for that game (512MB)

... Ya, probably only helping memory.     However, it might not even be that, depending on how the memory was handled.  (I have no idea how the onboard TnT2 worked.)

b. I will play with switching back between D3D and DDraw.  I always use DDraw thinking it would be faster by default.

Probably right to use ddraw with those cards, but no harm trying d3d.

a. I don't know if even DX 8.0 was available when these products were made 

 

e. Again, I have been trying this with DDraw but I could very well be a case of a very crappy card being outperformed by onboard video

I'd guess they're about the same level in proformance.  Sure the GeForce 2 MX is two generations after the TnT2, but since it's the MX (read "value") version, it's performance is probably about the same as GeForce 256 (the previous gen to GeForce 2, and the gen after TnT2).  And IIRC, the TnT2 beat the GeForce256 in a few tests; sure it lost more, but mame wasn't one the the tests.

As far as a margin of error for the benchmark, it would be less than 1FPS in this case.  I always start the game, insert coins and press P1 to go to the character selection screen.  10 out of ten times I get the same number of FPS in each case so it is pretty clear that the total variability is dominated by the video cards and not the measurement system.

If you really want to benchmark, use mame's -seconds_to_run option (aka -str).  Not only will the runs being compared be the exactly same, but you will have higher precision numbers than mame's onscreen FPS.  That said, you'll probably get the same results as you've been getting, but even if you're getting the same numbers your way of testing there are variables that can be removed by using -str.

...By adding the card I was simply hoping to give it a little boost.  I can't count how many people give the advice of buying a separate card rather than onboard video to "not steal away from the CPU".  I am pretty convinced by now that this is rubbish.  Video cards are for PC games, video rendering etc...

It's general rule.  IOW, it's usually true.  IOW2, it's sometimes not true.
In your case, you're replacing a video card (that just happens to be onboard) with another a little newer but "value" video card.
So in your case, it sounds like it's rubbish. 
Please don't make the same mistake you are knocking by taking your one instance, (converting it to a general rule,) and believing it's always true.

Don't make me get on my soapbox ranting about general rules incorrectly being said and read as altruisms, again.

[TheManuel]

Even though in my cases, I've dealt with two instances of this issue, generalization is alwasy dangerous.  I agree. 

At any rate, I went back last night and pulled the card out and ran the game with the onboard video again and now I am getting the same number of fps as with the video card.  Clearly something changed since I did the first tests with the onboard video but I have changed so many things that I lost track and it now confounds the whole issue. 

At any rate, it does not seem like it is much of an issue either way.  I will simply find older versions that allow me to play the slow games better.
I'll concentrate now on deciding whether to use an arcade monitor or a TV but I am leaning toward the first.

By the way, what was the name of your old mame build that handled multiple mice and such?  I will not be needing it now but humor me for nostalgia's sake.

Thank you.