PIII 933 and E6300 Benchmarks Added to Mame Benchmarks Site

[Dmod]

I've been trying to decide whether to upgrade my PIII and if so what to replace it with.  I ran some benchmark tests for comparison to other results at

http://benchmark.mameworld.net/header.html

I've been using v102.  In some previous testing, I found that anything above v102 results in a drastic slowdown on the PIII.  I followed the guidelines at the site for MAME parameters.  I just eliminated the ones that aren't supported anymore.

I tried to find ROMs that had been tested against most of the other processors and selected the ones that I thought were borderline and would give good comparisons against other processors.  As a result, most of the ROMS I tested were challenging and the results tend to be lower than desired frame rates.

Hopefully this data will be useful to others.  I have another machine that I plan to test (Core 2 Duo E6300) and will post those results when I get around to it.


ROM  MAME Build  CPU  MHz  FPS 
1943  0.102  Intel Pentium III  933  136 
batrider  0.102  Intel Pentium III  933  62 
cbaj  0.102  Intel Pentium III  933  25 
crusnusa  0.102  Intel Pentium III  933  15 
dariusg  0.102  Intel Pentium III  933  28 
gdarius  0.102  Intel Pentium III  933  24 
kof99  0.102  Intel Pentium III  933  71 
mk  0.102  Intel Pentium III  933  47 
mslug3  0.102  Intel Pentium III  933  66 
nbajam  0.102  Intel Pentium III  933  40 
radm  0.102  Intel Pentium III  933  18 
rtype  0.102  Intel Pentium III  933  92 
sfex  0.102  Intel Pentium III  933  23 
ssf2t  0.102  Intel Pentium III  933  71 
starblad  0.102  Intel Pentium III  933  14 
stunrun  0.102  Intel Pentium III  933  41 
tekken  0.102  Intel Pentium III  933  22 
term2  0.102  Intel Pentium III  933  53 

[spystyle]

Dmod,

You know all about MAME's speed, nice!

I can pick your brain

What versions of MAME do you think are fastest? I had the idea that .60 and the like were very quick compared to the newer builds.

What versions would you recommend for budget systems? Like P2 500mhz and P3  500mhz?

Do you have any documentation detailing MAME's speed and how to speed it up?

Thanks,
Craig

p.s. Thanks for sharing your batch file to benchmark mame! It's wild to see DigDug intro at 400 fps. I quickly tested ver 36 against 61 and 92. I found 61 was practically the same speed as 36, but 92 was 100 FPS slower than 61.


digdug mame92 379pfs
digdug mame61 475fps
digdug mame36 478fps

[Dmod]

Spystyle,

I'm no expert but I did some testing on my PIII between v078 and v110.  The Mame Benchmarks site isn't mine.  But I've found it very useful, and just recently added some of my own data to it.

For some more info, check out this thread:

http://forum.arcadecontrols.com/index.php?topic=63584.0

I tested a bunch of versions between 0.078 and 0.102 and found only a minor slowdown, so I've been sticking to 0.102 in order to include some of the newer ROMs.  Above 0.102 things slowed down drastically.

An alternative to the regular MAME build is to use fastmame78 or fastmame84.  This yields higher speeds.
 
----------------------
EDIT:  This is probably a better answer to your question...

Aaron Giles recently benchmarked PacMan all the way from 0.37 to 0.113.  See the results here:

http://aarongiles.com/?p=208

[spystyle]

Thanks for that, it's interesting.

Although Pac-Man is a very simple game. Like he wrote at the bottom, Rastan would be a good choice, I am anxious to see how that test comes out.

But now that I have your benchmarking batch file I suppose I could do it myself :)

Cheers,
Craig

[Dmod]

I'd like to see something like mslug3.  When I tested mslug3 I saw a big drop in fps for versions between 102 and 106 which pacman doesn't seem to show.

[u_rebelscum]

I tested a bunch of versions between 0.078 and 0.102 and found only a minor slowdown, so I've been sticking to 0.102 in order to include some of the newer ROMs.  Above 0.102 things slowed down drastically...

Aaron Giles recently benchmarked PacMan all the way from 0.37 to 0.113.  See the results here:

I've done some testing too, and for me it depended more on the settings I used and which computer I tested on and the specific game run, than the version of mame. 
One game (rastan) actually significantly increased in speed after the 0.106 video change with some settings (directdraw) on two of the three computers I tested, but dropped or stayed level with other settings and/or on the third computer (most other games decreased even with the same settings). 
Then there were a couple games than increased in speed as many times as decreased in speed between any two consecutive versions of mame (even though the size in decrease averaged higher than the average increase for these games).
Vector games that I tested had a few large jumps between a few specific versions of mame, but under different settings and different games the direction of the jump often was different, even though between the same versions.
The most consistent decreases across most games and all three of the computers I noticed, however, seemed to be due to mame's changing default settings.  (Even so, it didn't always happens for all game and all settings I tested.)

I didn't post the results before because anything could be "proved" with the inconsistent results, by using only the results that support whatever is being "proved" and ignoring the rest.  Here's a few charts, the first two different games with the same settings, and last one one game with three settings; as you can see, this was done a little while ago:

[u_rebelscum]

I'd like to see something like mslug3.  When I tested mslug3 I saw a big drop in fps for versions between 102 and 106 which pacman doesn't seem to show.

I tested MslugX, included in above post, but I have a png of that game by itself; heck I'll include MK2 and rastan by themselves since I have those already too.  Like I said before, something's weird with rastan, dd, and the test computer.

[Dmod]

Really interesting results.   I guess it's not that surprising that the "step changes" in MAME performance vary quite a bit depending on the ROM tested.

I tried browsing through the MAME revision docs and MAWS to find some commonality among the ROMS that showed sharp declines above v102 but I'm a monkey's uncle if I could find any correlation.

Since they all don't drop on the same version, it seems like each drop is due to a different code change in the MAME code.

[Dmod]

I've added benchmarks for my E6300 Core 2 Duo @ 1.86 GHz using v111.  They're up on the site.

For anyone who's interested, it seems like the E6300 runs at about 4x the speed of my 933MHz PIII.  Results for the 2.4-2.5 GHz Athlon 64 seems to run MAME at close to the same rate as the E6300.

These estimates are rough since not all results were obtained using the same MAME version.