4Ghz Core 2 Duo vs M.A.M.E. 0.120 (benchmark results)

[headkaze]

This post might help

http://www.mameworld.info/ubbthreads/showthreaded.php?Cat=&Number=129879&page=0&view=expanded&sb=5&o=&fpart=1&vc=1

[John IV [MameUI64]]

Thanks for the run.  Yes you're missing the fastest configuration of .120u1 at 64bit.  The jumps Aaron made in the software sli and applying that to the other 3d drivers really gave some nice boosts.  He does his 64bit builds [PTR64=1 ]w/out PM=1 so it will be interesting to see if there are any changes w/ PM=1.

[taz-nz]

Since this thread is starting reads like a long list numbers, and the point of these benchmarks is to help out those of you looking to find out what you need to run the most demanding M.A.M.E. ROMs, I decided to post something other than benchmarks today, mainly because I'm still trying to wrap my head around two different Microsoft SDKs, 64 Mame compiling and my Hard Drives looks like MAMEDEV exploded all over them.

So for all those of you wanting to build a preformance M.A.M.E. PC on a budget, I did a quick shopping list and then found some prices over at www.clubit.com (There may be better and cheaper places out there, but they had the right gear at what looked like the right price)

	GIGABYTE GA-P35-DS3R Rev. 2.0 LGA 775 Intel P35 ATX Intel Motherboard Retail	$128.50
	Cooler Master RR-CCH-LB12-GP Hyper 212 CPU Cooler	$33.99
	Intel Core 2 Duo E4500 M0 Stepping Conroe 2.2GHz 2MB L2 LGA 775 Processor Retail BX80557E4500 SLA95	$123.50
	CORSAIR XMS2 2GB (2 x 1GB) DDR2-800 CL 5-5-5-12 Dual Channel Kit Retail TWIN2X2048-6400	$74.50
	Sub Total:	$360.49

Add a dirt cheap PCI-E VGA card, case with atleast a 400-450watt PSU & a 120gb HDD for your OS & MAME,and with a little overclocking you should hit 3.2ghz (8x 400fsb) and with a some more extreme overclocking and some luck you could get as high as 3.6ghz  (9x 400fsb), without the need for a lot expensive hardware.

(I suggest dropping the Multiplier below 11x and increasing the FSB to 400mhz, as the P35 chipset will easily handle 400fsb and the extra bandwidth doing this will give you will help hide the any drop in preformance you may get from the E4500 having only 2mb cache.)

That should get you 80-90% of my benchmark scores which as long as your running MAME64 0.120u1 in Vista 64, that will put most of the working ROMs on my benchmark list in the playable basket for you.

Thanks for the run.  Yes you're missing the fastest configuration of .120u1 at 64bit.  The jumps Aaron made in the software sli and applying that to the other 3d drivers really gave some nice boosts.  He does his 64bit builds [PTR64=1 ]w/out PM=1 so it will be interesting to see if there are any changes w/ PM=1.

As soon as I can get myself sorted with the whole compiling a 64bit build of mame, I'll post both In-game & CPU benchmarks for MAME64 0.120u1 with and without PM optimisation. (It really shouldN'T be this hard  , but like I said I'm rusty, and the documantion on compiling MAME64 sucks.)

This post might help

http://www.mameworld.info/ubbthreads/showthreaded.php?Cat=&Number=129879&page=0&view=expanded&sb=5&o=&fpart=1&vc=1

Thanks for the link, I think I have all the required files now, I've just got to install everything get everything in the right place on my HDD, and then I can try it all out.

[John IV [MameUI64]]

Just FYI, I released the 64bit version of Mame32 .120.2 after going to Vista-64 the other night.  Got some nice increases as you're seeing too.

Game   Driver   Arguments   .120.2-32   .120.2-64   Δ
crusnusa   midvunit.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   182.84%   220.81%   20.77%
starblad   namcos21.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   119.37%   127.74%   7.01%
gauntleg   vegas.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   139.10%   163.50%   17.54%
blitz   seattle.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   113.12%   145.63%   28.74%
daytona   model2.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   89.09%   120.89%   35.69%
gradius4   hornet.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   97.97%   79.57%   -18.78%
radikalb   gaelco3d.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   109.88%   163.93%   49.19%
ridgerac   namcos22.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   92.37%   147.79%   60.00%
scud   model3.c   -str 90 -video none -nosound -nothrottle -norc -nosleep   33.88%   28.48%   -15.94%
Vista Ultimate 64 • Intel C2D E6400 3.5Ghz • 2GB Corsair DDR2 800 • ATI X1950 Pro               20.47%

[taz-nz]

Just FYI, I released the 64bit version of Mame32 .120.2 after going to Vista-64 the other night.  Got some nice increases as you're seeing too.

Sweet, that saves me from another day of going around in circles trying to compile a 64bit build,  Does this mean there will be an official 64bit of MAME32 from now on ?    (I'd love a step by step how-to of what I need and where to put it, to make my own 64bit MAME builds if you have time to write one)

Running MAME in 64bit really does make for some great improvements, but it's interesting that gradius4 CPU score takes such a hit in preformance when you jump to 64bit, where as the 32bit versions show an increase in preformance with each new build, yet the 64bit builds take a pounding. (Maybe it's a little too optimised for the 32bit Enviroment to work well in 64bit currently.)

I ran a set of the pure CPU benchmarks on your MAME32 64bit build, and update the score chart. (I finally figured out why I wasn't getting an output to file from Mame32)

CPU	ROM	Driver	0.119	0.120	0120u1	0120u2	0.120(64)	0.120u2(64)
E6400 @ 3.4ghz	crusnusa	midvunit.c	212.24%	212.12%	178.27%	182.84%	n/a	220.81%
E6850 @ 4.0ghz	crusnusa	midvuint.c	257.69%	257.32%	218.20%	214.79%	271.81%	269.31%
E6400 @ 3.4ghz	starblad	namcos21.c	234.33%	120.83%	116.97%	119.37%	n/a	127.74%
E6850 @ 4.0ghz	starblad	namcos21.c	299.30%	144.10%	144.00%	142.10%	153.57%	150.71%
E6400 @ 3.4ghz	gauntleg	vegas.c	47.92%	114.97%	132.38%	139.10%	n/a	163.50%
E6850 @ 4.0ghz	gauntleg	vegas.c	68.91%	166.17%	173.07%	166.03%	183.42%	191.32%
E6400 @ 3.4ghz	blitz	seattle.c	122.25%	87.86%	111.59%	113.12%	n/a	145.63%
E6850 @ 4.0ghz	blitz	seattle.c	153.43%	138.59%	143.28%	136.99%	164.15%	185.35%
E6400 @ 3.4ghz	daytona	model2.c	n/a	86.18%	87.22%	89.09%	n/a	120.89%
E6850 @ 4.0ghz	daytona	model2.c	n/a	106.17%	106.43%	104.94%	142.72%	140.42%
E6400 @ 3.4ghz	gradius4	hornet.c	60.36%	79.34%	n/a	97.97%	n/a	79.57%
E6850 @ 4.0ghz	gradius4	hornet.c	116.72%	164.94%	n/a	180.32%	122.16%	125.69%
E6400 @ 3.4ghz	radikalb	gaelco3d.c	77.46%	76.11%	95.72%	109.88%	n/a	163.93%
E6850 @ 4.0ghz	radikalb	gaelco3d.c	94.25%	93.97%	116.77%	129.44%	150.07%	200.22%
E6400 @ 3.4ghz	ridgerac	namcos22.c	65.56%	64.73%	90.77%	92.37%	n/a	147.79%
E6850 @ 4.0ghz	ridgerac	namcos22.c	80.00%	79.54%	114.89%	112.44%	106.77%	194.55%
E6400 @ 3.4ghz	scud	model3.c	34.41%	32.91%	33.00%	33.88%	n/a	28.48%
E6850 @ 4.0ghz	scud	model3.c	41.21%	39.60%	39.56%	38.93%	33.81%	33.89%

Looks like scud is the only ROM still holding out, but since it's a non-working rom and the emulation needs a lot of work, it only makes sense that it's scores are low. I had Task Manager open on the Preformance tab while running this lastest set of benchmarks, it showed that the model2.c & model3.c drivers are noticably lacking any form of multi-threading with one core being loaded to near 100% while the other is lucky to hit 10%, It will be interesting see if any of Aaron's software SLI magic can be applied to these drivers in the future.

For those still interested I'll run a full set of In-Game benchmarks tommorow with the 64bit build of MAME 0.120u2, but in the mean time here's a preview:

Propcycl:
0.120u2 (64bit)       151.88%         91.28fps

Blitz:
0.120u2 (64bit)       171.18%       102.71fps

Gauntleg:
0.120u2 (64bit)       142.84%         81.42fps

[AaronGiles]

Thanks for the benchmark results. scud and gradius4 both use the PowerPC, which has a DRC on 32-bit but not on 64-bit, so it's not surprising that 64-bit is slower.

In terms of future work, model2 and model3 can definitely be accelerated the same way, just haven't gotten around to it yet. System21 probably can be as well.

[Red]

Thanks for the benchmark results. scud and gradius4 both use the PowerPC, which has a DRC on 32-bit but not on 64-bit, so it's not surprising that 64-bit is slower.

In terms of future work, model2 and model3 can definitely be accelerated the same way, just haven't gotten around to it yet. System21 probably can be as well.

Thanks for all your work Aaron.  It is much appreciated.

Red

[taz-nz]

Thanks for the benchmark results. scud and gradius4 both use the PowerPC, which has a DRC on 32-bit but not on 64-bit, so it's not surprising that 64-bit is slower.

Just glad I be help, even if It only in a little way. I figured there had to be something about the 32bit code that didn't translate to 64bit I just didn't know what it would be.

In terms of future work, model2 and model3 can definitely be accelerated the same way, just haven't gotten around to it yet. System21 probably can be as well.

I figured as much, if it works as well as it did vegas.c and the other we should see some really nice preformance gains for it. I was suprised by how effectively it maxed out both cores on my CPU, If I get a chance in the next couple of days I'll run a set of the Pure CPU benchmark on one of the Q6600 quad cores on display at work, atleast one of them is running Vista 64 so I'll probably benchmark 64bit MAME 0.120u2 and see how it goes.

Anyway thanks for answering my queries', and if there is anything else i can do to help just let me know.

[taz-nz]

Here are the In-game benchmark results for MAME32 64bit 0.120u2 (i'll tidy them up into a easy to read table in the next couple of day, but for now it's a cut and paste job.


- 1492      
0.120      5570.54%   3342.32fps
0.120 64bit   6503.71%   3902.23fps

gaelco3d.c
- radikalb   
0.120      88.21%      52.93fps          
0.120 64bit   134.23%      
0.120u1      117.74%      70.64fps
0.120u2   64bit   177.31%      

- speedup   
0.120      102.93%      61.76fps   
0.120 64bit   157.14%      
0.120u1      127.66%      76.60fps
0.120u2   64bit   177.36%

- surfplnt   
0.120      91.29%      54.77fps   
0.120u1 64bit   139.72%
0.120u1      112.79%      67.67fps
0.120u2   64bit   162.83%

Mediagx.c
- a51site4   
0.120      185.95%      111.57fps   
0.120 64bit   234.34%      
0.120u1      186.04%      111.62fps
0.120u2 64bit   239.42%      

medvunit.c
- crusnusa   
0.120      225.23%      128.38fps   
0.120 64bit   271.99%
0.120u1      220.16%      125.49fps
0.120u2 64bit   269.87%

- crusnwld   
0.120      229.70%      130.93fps   
0.120 64bit   249.66%
0.120u1      221.42%      126.21fps
0.120u2 64bit   238.78%

- offroadc   
0.120      395.36%      225.36fps   
0.120 64bit   425.80%
0.120u1      365.57%      208.38fps   
0.120u2 64bit   427.64%

- wargods   
0.120      376.85%      214.80fps   
0.120 64bit   411.71%
0.120u1      330.00%      188.10fps
0.120u2 64bit   404.79%

namcos22.c
- airco22b
0.120      97.73%      58.64fps    
0.120 64bit   124.71
0.120u1      110.64%     66.38fps
0.120u2 64bit   150.83%

- alpinerd   
0.120      42.36%      25.42fps   
0.120 64bit   53.37%
0.120u1      65.95%      39.57fps
0.120u2   64bit   97.53%

- cybrcomm   
0.120      67.79%      40.68fps   
0.120 64bit   96.65%
0.120u1      95.78%      57.49fps
0.120u2   64bit   145.39%

- cybrcycc   
0.120      125.69%      75.41fps   
0.120 64bit   161.28%
0.120u1      157.40%      94.44fps
0.120u2 64bit   209.53%

- propcycl   
0.120      70.44%      42.50fps   
0.120 64bit   96.31%
0.120u1      100.39%      60.23fps
0.120u2 64bit   152.83%

- raveracw   
0.120      53.70%      32.22fps   
0.120 64bit   68.33%
0.120u1      81.37%      48.82fps
0.120u2 64bit   130.37%

- ridgerac   
0.120      75.84%      45.50fps   
0.120 64bit   103.33%      
0.120u1      108.13%      64.88fps
0.120u2 64bit   164.51%

- timecris   
0.120      70.84%      42.50fps   
0.120 64bit   93.17%
0.120u1      101.81%      61.09fps
0.120u2 64bit   158.05%

Seattle.c   
- biofreak
0.120      91.78%      55.07fps
0.120 64bit   140.22%      
0.120u1      101.70%      57.97fps
0.120u2   64bit   137.62%


- blitz      
0.120      122.34%      73.40fps   
0.120 64bit   153.09%
0.120u1      127.38%      72.61fps
0.120u2 64bit   173.04%

- blitz2k   
0.120      121.94%      69.51fps   
0.120 64bit   146.08%
0.120u1      123.73%      70.52fps
0.120u2   64bit   160.07%

- blitz99   
0.120      120.27%      68.55fps   
0.120 64bit   142.65%
0.120u1      120.26%      68.55fps
0.120u2 64bit   152.24%

- calspeed   
0.120      146.21%      83.34fps   
0.120 64bit   181.91%
0.120u1      165.08%      94.10fps
0.120u2      211.28%


- carnevil
0.120      241.11%      137.43fps   
0.120 64bit   287.40%
0.120u1      221.69%      126.36fps
0.120u2 64bit   268.86%

- hyprdriv   
0.120      140.84%      80.28fps   
0.120 64bit   185.50%
0.120u1      144.27%      82.23fps
0.120u2 64bit   175.13%

- mace      
0.120      173.23%      98.74fps   
0.120 64bit   217.20%
0.120u1      186.67%      106.40fps
0.120u2 64bit   255.76%

- sfrush   
0.120      144.95%      82.62fps   
0.120 64bit   199.68%
0.120u1      152.01%      86.64fps
0.120u2 64bit   204.46%

- wg3dh      
0.120      253.19%      144.32fps   
0.120 64bit   342.70%
0.120u1      265.32%      151.23fps
0.120u2 64bit   383.10%


Vegas.c
- gauntdl
0.120      108.58%     61.89fps    
0.120 64bit   123.46%
0.120u1      110.06%      62.73fps
0.120u2 64bit   132.08%

-gauntleg
0.120      108.57%      61.88fps
0.120 64bit   153.76%            (tends to lockup)
0.120u1      111.60%      63.61fps
0.120u2 64bit   152.79%

- tenthdeg   
0.120u1      62.03%      35.36fps   
0.120 64bit   76.21%
0.120u1      67.42%      38.43fps
0.120u2 64bit   83.69%

model2.c
- Daytona         
0.120      120.15%      72.06fps   
0.120 64bit   157.64%
0.120u1      120.08%      72.05fps
0.120u2 64bit   155.59%

hornet.c
- gradius4   
0.120      112.04%      67.91fps
0.120 64bit   92.69%
0.120u1      fails instantly with memory error
0.120u2      117.38%
0.120u2 64bit   96.02%

Model3.c
- scud      
0.120      40.98%      24.59fps   
0.120 64bit   33.57%
0.120u1      44.21%      26.53fps
0.120u2 64bit   n/a


namcos21.c   
- starblad
0.120      140.06%      84.03fps
0.120 64bit   146.24%
0.120u1      139.95%      83.97fps
0.120u2 64bit   144.54%

-swtrilgy (Emulation need a lot of work still graphics issues)
0.120u1      61.66%      39.97fps

Basically the results show if it's a working ROM it's playable and by a good margin in most cases, there are a couple like tenthdeg and aplinerd that still coming in a little low, but gauntleg looked up playable a couple of builds back and now it runs great so only time will tell with these two.

[Haze]

could you benchmark this? http://www.mameworld.net/maws/set/puzzlekg  (Puzzle King Dance & Puzzle.. by Eolith)


I'd be interested to know how well it performs on the 4ghz machine, yes, it's a simple 2d game, but it pushes the e132xs (hyperstone) core pretty hard, so the speedups have little effect.

[divemaster127]

I purchased vista 64 yesterday as soon as I get my new quad 4 built w/vista I will also post framerates, im going to use his guide for overclocking also, much better than I have been able to do
dm

[Red]

Thanks for all your hard work on the benchmarks taz. 

It's looking like my next computer purchase will finally be able to play Blitz!

Do you have to compile a 64bit build of MAME to get these results?  Or can you just run regular MAME 64?

Did I miss it in the thread that said why you were compiling MAME 64?  I'm new to MAME and just don't understand the whole compiling stuff and why one would want to do it if the MAME devs have already done it for us with MAME 64.

Also, what's better for 64 bit MAME- Windows XP or Vista?  Thanks guys.

Red

[John IV [MameUI64]]

could you benchmark this? http://www.mameworld.net/maws/set/puzzlekg  (Puzzle King Dance & Puzzle.. by Eolith)


I'd be interested to know how well it performs on the 4ghz machine, yes, it's a simple 2d game, but it pushes the e132xs (hyperstone) core pretty hard, so the speedups have little effect.

This is at 3.5Ghz
mame32 -str 90 -video none -nosound -nothrottle -norc -nosleep puzzlekg
Average speed: 95.19% (89 seconds)

About 85% during gameplay part of demo.

[taz-nz]

could you benchmark this? http://www.mameworld.net/maws/set/puzzlekg  (Puzzle King Dance & Puzzle.. by Eolith)


I'd be interested to know how well it performs on the 4ghz machine, yes, it's a simple 2d game, but it pushes the e132xs (hyperstone) core pretty hard, so the speedups have little effect.

Core 2 Duo E6850 @4.0ghz running

Puzzlekg:
MAME32 01.20u2 (32bit)            119.68%           71.81fps
MAME32 01.20u2 (64bit)            109.00%           65.40fps

Go figure, who would have thought such a simple game could eat up so much processor time to emulate. But since most of the in game screen is static, it should be playable even at those scores as your unlikely to see wild swings in frame rate while playing.

I purchased vista 64 yesterday as soon as I get my new quad 4 built w/vista I will also post framerates, im going to use his guide for overclocking also, much better than I have been able to do
dm

I'm very interested to see what you overclocked scores are, I ran a set of benchmarks on a stock Q6600 today (results below) at I got more questions than answers out of do it.

Pure CPU benchmarks for Q6600 @2.4ghz, 4gb, 8800gts, Vista 64

CPU	ROM	Driver	0.120u2(64)
E6400 @ 3.5ghz	crusnusa	midvunit.c	220.81%
E6850 @ 4.0ghz	crusnusa	midvuint.c	269.31%
Q6600 @ 2.4ghz	crusnusa	midvunit.c	139.86%
E6400 @ 3.5ghz	starblad	namcos21.c	127.74%
E6850 @ 4.0ghz	starblad	namcos21.c	150.71%
Q6600 @ 2.4ghz	starblad	namcos21.c	90.47%
E6400 @ 3.5ghz	gauntleg	vegas.c	163.50%
E6850 @ 4.0ghz	gauntleg	vegas.c	191.32%
Q6600 @ 2.4ghz	gauntleg	vegas.c	133.76%
E6400 @ 3.5ghz	blitz	seattle.c	145.63%
E6850 @ 4.0ghz	blitz	seattle.c	185.35%
Q6600 @ 2.4ghz	blitz	seattle.c	128.91%
E6400 @ 3.5ghz	daytona	model2.c	120.89%
E6850 @ 4.0ghz	daytona	model2.c	140.42%
Q6600 @ 2.4ghz	daytona	model2.c	83.73%
E6400 @ 3.5ghz	gradius4	hornet.c	79.57%
E6850 @ 4.0ghz	gradius4	hornet.c	125.69%
Q6600 @ 2.4ghz	gradius4	hornet.c	65.95%
E6400 @ 3.5ghz	radikalb	gaelco3d.c	163.93%
E6850 @ 4.0ghz	radikalb	gaelco3d.c	200.22%
Q6600 @ 2.4ghz	radikalb	gaelco3d.c	109.55%
E6400 @ 3.5ghz	ridgerac	namcos22.c	147.79%
E6850 @ 4.0ghz	ridgerac	namcos22.c	194.55%
Q6600 @ 2.4ghz	ridgerac	namcos22.c	33.74%
E6400 @ 3.5ghz	scud	model3.c	28.48%
E6850 @ 4.0ghz	scud	model3.c	33.89%
Q6600 @ 2.4ghz	scud	model3.c	20.16%

I'm not sure what to make of these results, for most of the result the Q6600 results are in keeping with MAME scaling with CPU clock, but the score are almost those I like those I would expect from a 2.4ghz Core 2 Duo, there is not sign of any preformance gain on the Multi-thread drivers even though all 4 core were being used to about 75%+ load on average. 

But the really weird result is from ridgerac, which tanks total on the quad core.

I guess we will have to wait for more Quad core benchmarks at a couple of different CPU clocks before we get any real idea of a preformance pattern for Quad Core CPUs. (I've got a feeling the low FSB bus speed and thus limited FSB bandwidth is getting in the way of the Quad core really preforming)

Thanks for all your hard work on the benchmarks taz. 

It's looking like my next computer purchase will finally be able to play Blitz!

Thanks, I'm just glad people are get some use out of all it, and hope this put the myth that you need a 10ghz CPU to run CHD rom to rest once and for all.

If your not buying until next year the Intel Yorkdale (quad) and Wolfdale (dual) cored processors look like to hold a lot of potential for overclocking and may make 4.5ghz & 5.0ghz overclocks possible. (without the ultra extreme cooling) I don't think anyone here is going to be running out to buy a Core 2 Extreme QX9650 any time soon, so we will have to wait and see what happens next year.

Do you have to compile a 64bit build of MAME to get these results?  Or can you just run regular MAME 64?

Did I miss it in the thread that said why you were compiling MAME 64?  I'm new to MAME and just don't understand the whole compiling stuff and why one would want to do it if the MAME devs have already done it for us with MAME 64.

I currently using standard MAME32 0.120 builds by John IV for both 64 & 32 bit benchmarks.

All the official builds are non-optimised so they will run on any CPU, but by compiling your our build you can optimised the builds to the processor your using and thus gain an extra few percent in the benchmark scores, using a PM optimised build on a Core 2 Duo appears to add 5-8% to your scores.

I'd also like to start playing with the source to see if I can get to know it a bit and may help out in the future.

Also, what's better for 64 bit MAME- Windows XP or Vista?  Thanks guys.

Red

To tell the truth I'm not a huge fan of Vista, but it has been growing on me, I much prefer Windows XP as my main OS. But in saying that I'd have to say get Vista 64 for a couple of pratical reasons, 64bit XP never really took off and driver support for it is fairly limited, where as Vista 64 is fairly well supported and I expect that will only improve with time, 64bit XP also still retains some rather nasty memory addressing limitations for 32bit windows that Vista 64bit does away with.

You can always run dual boot like myself if you can't live with Vista on a full time basis.


 

[Tiger-Heli]

Thanks, I'm just glad people are get some use out of all it, and hope this put the myth that you need a 10ghz CPU to run CHD rom to rest once and for all.

In fairness, I believe Aaron originally posted the 10ghz spec when people were struggling with Pentium 4's and Athlon's at less than 2 or 2.5 Ghz.  This was before Quad-core or even C2D or A64 had come to market.

I suspect your quad-core at 4 Ghz, is roughly equivalent to one of the processors of the time running at 10 Ghz.

I appreciate all the testing and all the gains made by MameDev, though!!!

[divemaster127]

The reason I jumped to the quad core, on the benchmarks I ran, the quad is twice as fast as my dual core e6600 was.  I think if the quad had your magic like the other 2 computers, it would be much faster in mame.  But after I finish the build if I find the quad is slower than the dual core due to the FSB no problem, I will switch the quad for the E6750 dual core in my racing setup & use that one.
dm

[metahacker]

taz, thanks again for this wonderful thread & research

i've found it rather inspiring.  so much so that i am getting rid of my tinyxp and going to vista64...and i've got my 2.66 cranked up to 3.6 (so far).. no sweat running orthos 24-7.....maybe i can get lucky and hit 4.0 =) but not so sure about that.

[ARTIFACT]

why use DDRAW on all these games which are ... 3D?

D3D would seem to fit much better with these 3D games.
Just curious if you see better performance... I'd imagine... Technically DDRAW isn't meant to be used to render 3D like that.

[taz-nz]

In fairness, I believe Aaron originally posted the 10ghz spec when people were struggling with Pentium 4's and Athlon's at less than 2 or 2.5 Ghz.  This was before Quad-core or even C2D or A64 had come to market.

I suspect your quad-core at 4 Ghz, is roughly equivalent to one of the processors of the time running at 10 Ghz.

I appreciate all the testing and all the gains made by MameDev, though!!!

Your right my 4ghz Core 2 Duo (dual core) is equal to about a P4 8ghz, with the MAME Dev team's great work taking up the slack to the 10ghz mark.

I'm not saying that the statement "you need a 10ghz CPU to run CHD roms" wasn't accurate at the time it was first used, 3 years or so ago, (well it was more like 5ghz at first then people upped it 10ghz with time) and I not attacking the people who said it back then, but the Core 2 Duo has been out for well over a year now and I was still seeing people using it as little as a couple of weeks ago, and it's simply no longer true. 

 

I saw it as MAME tall poppy syndrome, where anyone that dared to ask what they needed to run all the MAME roms at full speed was shot down and told to go buy a used P4 and don't waste their money. I don't expect everyone to run out a build 4ghz MAME monster machines, but atleast now people know it can be done and can choose for themselves that preformance level they want to take their MAME box to.

I just hope it puts it to rest for a while, atleast until we have some new fully emulated ROMs that can't be run on current PC hardware.

The reason I jumped to the quad core, on the benchmarks I ran, the quad is twice as fast as my dual core e6600 was.  I think if the quad had your magic like the other 2 computers, it would be much faster in mame.  But after I finish the build if I find the quad is slower than the dual core due to the FSB no problem, I will switch the quad for the E6750 dual core in my racing setup & use that one.
dm

Quad Core processors can pull of some extreme preformance gains running the kind of apps, I don't think the Q6600 did that bad in my benchmarks other than in ridgerac which I'm at a lost to explain, but I was expect a boost in scores on with the multi-threaded drivers and I just didn't see it. Maybe with a clock speed the quad core will shine more, but I guess we will have to wait until you post your result to see.  So I'm really looking forward to seeing you benchmark results.

taz, thanks again for this wonderful thread & research

i've found it rather inspiring.  so much so that i am getting rid of my tinyxp and going to vista64...and i've got my 2.66 cranked up to 3.6 (so far).. no sweat running orthos 24-7.....maybe i can get lucky and hit 4.0 =) but not so sure about that.

2.66ghz to 3.6 that a nice overclock, everything after 3.6ghz gets harder no matter what CPU clock you start with, but if it's a 100% stable like you say, and you've not already pumping insanely high core voltages into it, then I can say keep going and see where it take you, just watch those core temps.

I love TinyXP it's great for testing, 10mins to install and your in windows, now if only someone would make a TinyVista64, 15GB just for a base install is insane, my C: partition for XP Pro is only 6.4GB and it half empty, someone need to cut the fat out of Vista big time.

why use DDRAW on all these games which are ... 3D?

D3D would seem to fit much better with these 3D games.
Just curious if you see better performance... I'd imagine... Technically DDRAW isn't meant to be used to render 3D like that.

MAME doesn't use your 3D hardware to render anything, even if you running a ROM that uses all 3D graphics everything is done on your CPU, and then passed as 2D frames to your graphics card to display, MAME could easily take advantage of your 3D graphics hardware like many other emulators do, but the results you would get on screen would not be true to the orginal arcade machine.

Just to see what the difference would be I ran a quick benchmark on propcycl.
MAME32 0120u2 64bit      DDRAW        152.83%
MAME32 0120u2 64bit      Direct3D       116.72%

As you can see it was noticably slower and I also noticed a drop in the quality of what was being displayed. You would have to ask one of the Dev's to explain this.

If anyone know how I can extract from mame a complete list of parent ROMs based on the driver they used let me know.

[genesim]

In fairness, I believe Aaron originally posted the 10ghz spec when people were struggling with Pentium 4's and Athlon's at less than 2 or 2.5 Ghz.

The problem is that people keep repeating these things as if the facts never change.   Even back when I first pointed out the obvious multithread advantages there were many that argued in denial....without even testing the fact.

The funny thing is that while many have kept repeating the obsolete information, the devs starting working behind their backs.   

taz-nz,

Thanks for your work, and I also have been seeing speed jumps with my AMD 4400.   This is after it was configured correctly.

[Tiger-Heli]

If anyone know how I can extract from mame a complete list of parent ROMs based on the driver they used let me know.

Emuloader might be able to do this, depending on what you are looking for - also I think BuddaBing wrote a -listgen utility that would likely do a better job of that.

[ozzi22]

I wonder what the bench marks would be on linux 64

[ahofle]

This is slightly off topic, but also slightly on topic. 

Last night I overclocked my E6750 from 2.66 Ghz to 3.2 Ghz.  I simply adjusted the FSB to 400 Mhz and set the memory to 800Mhz for 1:1.  I was expecting to have to go back in and adjust the voltage, but I can't seem to make the voltage any lower than it already is (1.30 - 1.31V).  It seems to be running fine, but most of the guides I've read have people trying to find the minimum stable voltage to minimize heat and I have no way of decreasing the value (only increasing).     This is on an Abit IP35 Pro mobo.

[xmenxmen]

Refurb HP a6152n, Core 2 Quad Q6600, 3GB, 400GB, Vista Premium, $550 + $12 shipped (Frys Online)

http://www.fatwallet.com/t/18/778709/

Enjoy!!!!

[Tiger-Heli]

I was expecting to have to go back in and adjust the voltage, but I can't seem to make the voltage any lower than it already is (1.30 - 1.31V).  It seems to be running fine, but most of the guides I've read have people trying to find the minimum stable voltage to minimize heat and I have no way of decreasing the value (only increasing).

Generally, you raise the voltage to increase stability and lower it to reduce heat.

Typically, you bump it up a bit, find the max overclock it will boot at - drop down a bit from this for what you want to run at, and then drop the voltage from what you ramped up to to reduce heat if possible.

I am not current enough to know on your specific processor, but Taz is running 1.5 something volts and DM is running 1.7, so if you are stable at 1.3, I wouldn't worry, but that might be way off if say they have a different core config than you do, so take it with a grain of salt.

[taz-nz]

If anyone know how I can extract from mame a complete list of parent ROMs based on the driver they used let me know.

Emuloader might be able to do this, depending on what you are looking for - also I think BuddaBing wrote a -listgen utility that would likely do a better job of that.

thanks I'll check it out, I look for a way I can create a benchmark call file for a large number of roms without a lot of typing, if I can please a plain txt list, i can soon right a bit of code to format it into a batch file.

I wonder what the bench marks would be on linux 64

I would to like to see 64 bit linux benchmarks results too, but I unlikely to be running them any time soon, Linux + latest chipset motherboard + raid array = come back in 6 months when we there are drivers. (atleast that been the story the last 3 times I tried to install Linux on a new machine)

This is slightly off topic, but also slightly on topic. 

Last night I overclocked my E6750 from 2.66 Ghz to 3.2 Ghz.  I simply adjusted the FSB to 400 Mhz and set the memory to 800Mhz for 1:1.  I was expecting to have to go back in and adjust the voltage, but I can't seem to make the voltage any lower than it already is (1.30 - 1.31V).  It seems to be running fine, but most of the guides I've read have people trying to find the minimum stable voltage to minimize heat and I have no way of decreasing the value (only increasing).     This is on an Abit IP35 Pro mobo.

Your talking about undervolting and it's normally do to reduce heat and power useage, I have an undervolted and underclocked Athlon64 3000+ in my media box with passive cooling. People also undervolt with extreme cooling systems like phase change systems, but it your only useing air cooling you should be run stock voltage up to able 3.4-3.6ghz then you will need to increase it to overcome Vdrope, the catch is the more voltage you put in the more heat your cpu produces, the higher the CPU core temp the higher the voltage generally has to be to make the system stable, there is a point of zero return where all your doing is cooking your CPU.

Refurb HP a6152n, Core 2 Quad Q6600, 3GB, 400GB, Vista Premium, $550 + $12 shipped (Frys Online)

http://www.fatwallet.com/t/18/778709/

Enjoy!!!!

Good price on the surface, but it an HP box, so gutless 350watt PSU, no talk of the chipset used so probably something nasty, the bios will have zero overclocking settings. 3gb of points to another weild setup, probably only DDR2-667. If your looking for a cheap box with vista it's ok, but useless for overclocking or preformance computing. Stay away from brand name PC where ever possible they are built to look good and be cheap, and make the manufacture great margin at the same time, they hide a list of sins then you crack the case.

[u_rebelscum]

If anyone know how I can extract from mame a complete list of parent ROMs based on the driver they used let me know.

Hmm, mame -ls gives a full list, parents and clones. For not a complete list, there are maws (driver list), and the source files (of course , look at list at end of the file).

However, you'll have do some sorting to get the parents.  One way, in the source, the GAME MACRO goes:
GAME( YEAR, NAME, PARENT, MACHINE, INPUT, INIT, ROT, COMPANY, FULLNAME, FLAGS )
If the parents part is 0 then it's a parent, otherwise it's the name of the parent.

[Tiger-Heli]

thanks I'll check it out, I look for a way I can create a benchmark call file for a large number of roms without a lot of typing, if I can please a plain txt list, i can soon right a bit of code to format it into a batch file.

I know EL can select only parent ROMS (and roms based on game CPU or driver, for that matter) and can export the selected games to a text file.  I don't recall, but I think the text file might have:
mame=aburner
mame=asteroid
mame=bzone
etc.
but a little bit of search and replace "mame=" with "" in Word or a text editor might get you what you want, I hope!

[u_rebelscum]

If anyone know how I can extract from mame a complete list of parent ROMs based on the driver they used let me know.

Oh, and -listxml, too, but as is it's, umm, bulky.  RomLister can convert the xml to a list, as well as sort out clones, and output a batch file to run games.

[taz-nz]

If anyone know how I can extract from mame a complete list of parent ROMs based on the driver they used let me know.

Oh, and -listxml, too, but as is it's, umm, bulky.  RomLister can convert the xml to a list, as well as sort out clones, and output a batch file to run games.

Thanks for the tip worked great, created the XML file imported it into Romlister, filtered it for all working parent ROMs, and saved the list to a txt file. (3125 roms in total) Whipped up a bit of code in QBasic to turn the list of roms in to a call file for the benchmarking batch I use. (Yes I used Quick Basic, 8 lines of code and no need to compile.)I set the call file going and let it run for about 2.5 hours it got though about 350 roms, the worst score was for finalapr @ 5.99% you figure that one out.

I need to whip another bit of code to turn all the resulting txt files in to one file with all the ROM names and results. I may set the call file going and just let it run one day while I'm at work and see what I get in the way of result, some ROMs will get insane scores because they need user input to get pass the test screens, I seen a couple already that have scores around the 12000% mark. Probably not much use as a preformance test, but good for picking out the those roms that still have preformance issues.

[Red]

What's Tiny XP and where do I get it?  Thanks.

Red

[tommyinajar]

It 's a stripped down version of XP with alot of junk you don't need taken out, originally made for slower PC's.

It's NOT a Mic@$oft  release, it is a pirated ware. Make sure you own an original copy 1st.

Do a torrent Search for it.

[metahacker]

What's Tiny XP and where do I get it?  Thanks.

Red

google "TinyXP rev06"... "Tiny2003" is worth investigating as well.... it's quite an interesting build of windows. You can build your own copy of XP like TinyXP using nLite..... also, like he said, you need to have your own XP or W2K3 SVR license to use these legally.

[Red]

Thanks guys, much appreciated.

Red

[TheManuel]

What is PM and where could I find a PM build?
I would like to compare my benchmarks with yours (scaled down to 2.4GHz).

Thanks.

[Shoegazer]

Just read through this thread with great interest... very nice.

It's odd seeing all these great framerates.  I have a 2ghz dual-core laptop (core duo, NOT core2duo mind you), running xp32 and MAME32.  I realize I won't be able to even approach some of the stratospheric results I've been seeing from the likes of taz-nz, but still I did notice something weird: I checked "Multi-threaded rendering" and tried out every one of the games that have been tested in this thread (blitz2k, propcycl, etc).  Result?  0% increase in all cases.

My 2nd core is turned on in the BIOS, so what gives?  Weird.  I'm probably going to upgrade to a Penryn laptop next year when Intel releases it, though in the meantime it would be nice to see at least SOME performance increase out of this old dog...

Shoegazer

[TheManuel]

I did not see much from enabling MT either.
When you compare your results to those of the author of the post, make sure to run his same command line parameters.

[taz-nz]

What is PM and where could I find a PM build?
I would like to compare my benchmarks with yours (scaled down to 2.4GHz).

Thanks.

PM stands for Pentium M (mobile) the Pentium M, Core Solo, Core Duo, Core 2 Duo, & Core 2 Quad are all in same family of processors, and thus if you use the PM switch to compiling an optimised MAME build, these processors will gain a few % in preformance, but that build of MAME will no longer run on other processors, thus the offical MAME builds don't use this or any of the other CPU optimisation. Where do you find a PM optimised build, the easiest is to make one yourself by compiling the MAME source code.  Everything you need to compile your own 32bit build can be found here http://mamedev.org/tools/  and here http://mamedev.org/release.html. If you work out how to do it for a 64bit build send me a set of step by step instruction. (I still haven't gotten around to doing this myself)

MAME scales almost one to one with CPU clock speed, so if you get a score on a2ghz processor, that processor at 3ghz will get approx 1.5x it's score at 2ghz. So you can scale my scores the same way, 2.4ghz is 60% of 4ghz so your CPU should score around 0.6x my scores, allow +/- 10% for differences in FSB and cache memory etc. , this should get you in the ball park. (but your really need to hit 3ghz or better to make a lot of these games playable)

Just read through this thread with great interest... very nice.

It's odd seeing all these great framerates.  I have a 2ghz dual-core laptop (core duo, NOT core2duo mind you), running xp32 and MAME32.  I realize I won't be able to even approach some of the stratospheric results I've been seeing from the likes of taz-nz, but still I did notice something weird: I checked "Multi-threaded rendering" and tried out every one of the games that have been tested in this thread (blitz2k, propcycl, etc).  Result?  0% increase in all cases.

My 2nd core is turned on in the BIOS, so what gives?  Weird.  I'm probably going to upgrade to a Penryn laptop next year when Intel releases it, though in the meantime it would be nice to see at least SOME performance increase out of this old dog...

Shoegazer

the -MT does make a difference but it varies from ROM to ROM, some MAME drivers see a big boost others don't. The gain also scales with your CPU, so what might be a 1% or 2% gain for you might be a 5% or 6% gain for me, so at slower clock speeds it may be hard to pick any gains form the margin of error in the benchmarks.

I did a quick benchmark of some of the ROM to show the difference I get.

I use the command line:

mame32-64u2.exe -noautoframeskip -frameskip 0 -seconds_to_run 240 -nothrottle -nosleep -video ddraw -skip_gameinfo -effect none -nowaitvsync -noreadconfig -mt [ROM NAME] >.\bench\[ROM NAME].txt  (with and without the -mt switch)

ROM                with -MT       without -MT

radikalb          177.31%          159.18%
speedup         177.36%           158.01%
a51site4         239.42%          224.46%   
crusnusa          269.87%          253.39%
wargods         404.79%          385.23%
propcycl          152.83%          147.59%
timecris           158.05%          149.28%
blitz                 173.04%          167.22%
carnevil           268.86%          258.99%
wg3dh            383.10%          364.30%
gauntdl           132.08%          129.23%
gradius4           96.02%            93.07%
starblad          144.54%          139.86%

As you can see there is a 5-20% gain to be had on my system from using the -MT switch. (you should see some improvement too, but it may not be as great)

Your best bet to gain some preformance is to compile a PM optimised build and run that I saw about 8-10% gain with the use of a PM optimised build it could be more, I would love to be able to post a set of PM optimised MAME 64bit 0.120u2 benchmarks, but I jsut haven't had time this week to have another go at compiling a 64bit MAME build.

[TheManuel]

Thank you very much for taking the time to write that helpful post.
I will be a little more discering with the next benchmark as I did not run it using your process without the mt parameter disabled.  I did use your parameters exactly as you described them, mt enabled.   
I took your numbers for mace and scaled them down to 2.4GHz (my processor).  I am running MAME Plus 120u1 and my % frame output matched almost exactly (just a little over) your scaled number for version 120 but fell short of 120u1 by about 8 percentage points.  However, this is confounded by the fact that your 120u1 version was PM optimized so I will have to give that a try and see where I fall.
I will definitely get into compiling my own version.  Unfortunately, I don't konw if I will be able to help you with a 64bit optimized version as I don't have Vista so I will not be able to tell whether it's working or not.

A recommendation for the futue is to run a fixed number of frames rather than a fixed number of seconds.  The reason is that if someone wants to compare his benchmark to yours (if you care, anyway), and his/her computer is slower, that computer will run a shorter sequence of the game in 240sec so it will not be a 100% apples to apples comparison since some parts of the attract mode of a game are more taxing than others.  This could explain also why I see a slight improvement over your scheduled numbers.  Just an observation.

PS: this also goes to show how overwhelmingly dependent MAME is on processor speed and how little FSB, video card, memory speed and the rest matter.

[TheManuel]

Can anyone tell me what available switches are there while compiling with mingw and how do I specify them?  I also have a PIII that I would like to use on my future cabinet and would like to test different builds on it.
I could not figure this out from the compiling instructions.