|
Intel Alder Lake bencmarks in MAME 0.238
#392794 - 12/19/21 08:35 AM
|
|
|
Finally got a chance to Benchmark a Intel i9-12900KF in MAME today, had to use DDR4-3600 CL18 due to availability, DDR5 is basically vaporware currently.
Benchmarks were done in Windows 10 21H2 64bit, with latest drivers and updates. NZXT Z73 360mm AIO & gigabyte Z690 Auros Elite AX DDR4 motherboard.
I did one run completely stock, and a second with Efficiency Cores disable since I was testing in Windows 10.
Game R7 5800X 12900KF 12900KF (no E-cores)
blitz 379.07 449.45 477.69
crusnusa 379.18 503.52 544.13
cubeqst 458.47 603.67 605.48
cyvern 1448.13 1947.30 1965.81
dkong 5548.59 7713.05 7860.82
drivedge 429.27 595.97 595.99
gauntleg 672.26 787.06 818.68
gnbarich 2244.50 2675.91 2699.13
gradius4 569.73 714.10 746.32
gtfore06 285.75 366.88 378.36
harddriv 802.73 1078.82 1090.38
kidniki 573.11 780.20 784.41
kof98 1763.90 2272.10 2279.21
mario 1335.68 1908.05 1922.75
mk4 323.87 466.24 479.28
pacman 13861.67 23179.80 23180.68
pinkswts 2307.93 2664.19 2820.03
pongf 615.21 717.17 763.93
propcycl 363.23 423.23 456.07
radikalb 308.41 423.89 438.95
roadblst 1361.35 1756.50 1772.83
robotron 7236.99 9750.31 9818.95
rvschool 443.51 627.76 631.70
scud 132.20 132.83 138.84
sf2 2908.19 3930.43 3971.62
sfa2 2170.54 3210.78 3242.85
sfiii 1970.62 2370.55 2394.85
slrasslt 617.52 752.30 821.49
spacfury 240.74 294.19 296.78
starblad 243.98 314.16 314.94
starsldr 76.61 101.96 107.48
sushibar 110.46 143.02 144.59
tekken 641.13 927.18 932.36
tekken3 326.88 460.10 461.76
vf2 231.46 262.71 288.29
vfkids 184.78 242.87 242.45
All I can say is it's fast. 
I'll try to get some more results tomorrow with Win 11 and with faster DDR4-3800 CL14 memory, but don't hold your breath as works been insane and I may not have time.
|
If all else fails, Burn the manual.
|
|
| John IV |
|
IV/Play, MAME, MAMEUI
|
|
|
|
|
Reged: 09/22/03
|
|
Posts: 1967
|
|
Loc: Washington, USA
|
|
|
|
Send PM
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: taz-nz]
#392797 - 12/19/21 07:34 PM
|
|
|
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: John IV]
#392799 - 12/20/21 11:04 AM
|
|
|
I feel your pain having to wait for RAM, I order my DDR4-3800 Cl14 kit on the 28/11/2020 and got it on 13/03/2021.
Work was quiet today so got a chance to run benchmarks in Windows 11. Windows 10 still faster on average.
Windows 10 Windows 11
Game R7 5800X Stock No E-Cores Stock No E-Cores
blitz 379.07 449.45 477.69 439.52 444.29
crusnusa 379.18 503.52 544.13 509.81 481.67
cubeqst 458.47 603.67 605.48 587.07 586.67
cyvern 1448.13 1947.30 1965.81 1974.06 1926.66
dkong 5548.59 7713.05 7860.82 7617.03 7223.71
drivedge 429.27 595.97 595.99 595.86 587.75
gauntleg 672.26 787.06 818.68 794.13 741.42
gnbarich 2244.50 2675.91 2699.13 2636.11 2659.28
gradius4 569.73 714.10 746.32 684.79 627.36
gtfore06 285.75 366.88 378.36 362.82 343.25
harddriv 802.73 1078.82 1090.38 1079.84 1067.13
kidniki 573.11 780.20 784.41 773.43 782.37
kof98 1763.90 2272.10 2279.21 2281.92 2273.12
mario 1335.68 1908.05 1922.75 1894.59 1908.82
mk4 323.87 466.24 479.28 459.82 420.44
pacman 13861.67 23179.80 23180.68 23031.41 22973.07
pinkswts 2307.93 2664.19 2820.03 2749.78 2721.71
pongf 615.21 717.17 763.93 766.48 714.27
propcycl 363.23 423.23 456.07 421.08 382.35
radikalb 308.41 423.89 438.95 412.49 364.20
roadblst 1361.35 1756.50 1772.83 1771.71 1726.63
robotron 7236.99 9750.31 9818.95 9818.08 9722.61
rvschool 443.51 627.76 631.70 623.88 626.42
scud 132.20 132.83 138.84 129.48 135.64
sf2 2908.19 3930.43 3971.62 3986.52 3992.41
sfa2 2170.54 3210.78 3242.85 3237.81 3233.86
sfiii 1970.62 2370.55 2394.85 2400.21 2413.82
slrasslt 617.52 752.30 821.49 751.57 731.04
spacfury 240.74 294.19 296.78 293.80 293.36
starblad 243.98 314.16 314.94 310.97 311.44
starsldr 76.61 101.96 107.48 102.59 93.47
sushibar 110.46 143.02 144.59 138.69 143.90
tekken 641.13 927.18 932.36 868.51 921.25
tekken3 326.88 460.10 461.76 428.37 454.81
vf2 231.46 262.71 288.29 258.11 287.25
vfkids 184.78 242.87 242.45 218.16 241.07
I did do a benchmark run on a Dell Alienware Aurora R13 with 12900kf and DDR5 memory, but no one should buy that thing, the thermal design is a joke, and while it has DDR5-4800 CL36, it was running at DDR5-4400 with no way to change it, the bios was a joke. it preform almost universal worst that the DDR4 test bench I built.
I also tried DDR4-3800 CL14 memory, but as is often the case with early bios revisions, it wasn't stable with the aggressive memory timings, and while I did manage to get it to complete one run without crashing, it was slower than DDR4-3600 CL18, so it was clearly unhappy.
|
If all else fails, Burn the manual.
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: John IV]
#392809 - 12/21/21 02:36 AM
|
|
|
Hi!
I want to contribute benchmarks from 2 of my fastest machines
* AMD Threadripper PRO, 64 cores, 128 GB RAM, Ubuntu
* Mac Pro, Intel Xeon 28 cores, 64 GB RAM, Mac OS
However I never benchmarked mame. Can you tell me what command line string I need to enter to run each of the games on the page?
Edited by superdrew (12/21/21 02:37 AM)
|
|
|
|
|
Check out John IV page with long time charts.
http://www.mameui.info/Bench.htm
Be sure to adhere to the notes at the bottom of the tables.
I've included a batch file with the commands you'll need.
I think mine is still current(If not, someone will be quick to point it out.). Change the file extension from .txt to .bat and just run it in your MAME directorty. You'll also have to change the game name and/or add whichever other ones you want to check out.
I suggest you go back and look at some the the previous conversations about benchmarks, that is a good place to start too...
|
Pessimist: Oh, this can't get any worse!
Optimist: Yes, it can!
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: superdrew]
#392811 - 12/21/21 05:06 AM
|
|
|
MAME [game] -bench 90
I use a batch life automates the processes and outputs result to a txt file, but that will only work on Windows.
It's best to run all the games as normal first, as several of them have long first time setups, or need setting changed in service menu to launch directly into attract mode. Not doing this will give false scores as game will just holt at config screen during benchmark. (There is no video output during benchmark.)
While both your system are fast for multithreaded workloads, be warned MAME only uses the one or two cores for most ROMS, and basically doesn't scale past 4 cores. MAME only really cares about single core IPC and clock speed, but cache memory may have an effect.
Always good to see how different systems run MAME, good luck.
|
If all else fails, Burn the manual.
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: taz-nz]
#392817 - 12/21/21 01:59 PM
|
|
|
Quote:
> Finally got a chance to Benchmark a Intel i9-12900KF in MAME today,
>
> All I can say is it's fast.
Damn, that's a nice bump in performance.
I just wish my 9900K wasn't still so damn good at everything I throw at it, because I'd love a 12700K build, but just can't justify spending the money, to go from 'stupidly fast' to 'insanely fast' 
|
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: Ziggy100]
#392851 - 12/28/21 04:08 AM
|
|
|
> I'd love a 12700K build, but just can't justify spending the money, to go from
> 'stupidly fast' to 'insanely fast'
What about "ludicrous speed" then?
|
"Note to Noobs:
We are glad to help you but simply posting that something does not work is not going to lead to you getting help. The more information you can supply defining your problem, the less likely it will be that you will get smart-alec replies.
C.D.~"
|
|
| R. Belmont |
|
Cuckoo for IGAvania
|
|
|
|
|
|
Reged: 09/21/03
|
|
Posts: 9713
|
|
Loc: ECV-197 The Orville
|
|
|
|
Send PM
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: taz-nz]
#392947 - 01/06/22 04:13 AM
|
|
|
> While both your system are fast for multithreaded workloads, be warned MAME only uses
> the one or two cores for most ROMS, and basically doesn't scale past 4 cores.
0.239 has fixes for some issues that were discovered in MAME's job system. In English, that means I see consistent improvements for 3D games up to at least 8 cores on my 3900X now. I didn't test exactly where the perf stops going up, only that it's somewhere between 8 and 12.
|
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: taz-nz]
#392957 - 01/06/22 11:42 PM
|
|
|
I wonder what the just-revealed 12900KS with its reported 5.5GHz speeds might do in MAME...
|
|
|
| John IV |
|
IV/Play, MAME, MAMEUI
|
|
|
|
|
|
Reged: 09/22/03
|
|
Posts: 1967
|
|
Loc: Washington, USA
|
|
|
|
Send PM
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: R. Belmont]
#392959 - 01/07/22 01:10 AM
|
|
|
Which games / driver families benefit from this? I didn't see an uptick on the bench games w/ my 4 cores.
|
john iv
http://www.mameui.info/
|
|
| R. Belmont |
|
Cuckoo for IGAvania
|
|
|
|
|
|
Reged: 09/21/03
|
|
Posts: 9713
|
|
Loc: ECV-197 The Orville
|
|
|
|
Send PM
|
|
|
Re: Intel Alder Lake bencmarks in MAME 0.238
[Re: John IV]
#392997 - 01/07/22 06:06 AM
|
|
|
> Which games / driver families benefit from this? I didn't see an uptick on the bench
> games w/ my 4 cores.
Anything using poly_manager (Voodoo, Model 1/2/3, Naomi, the recently improved Konami 3D games). It used to peg out at 3 cores (it'd actually get slower at 4 and up). Now you can -numprocessors 8 (at least) if you have them and it'll give some improvements.
|
|
|
Stiletto |
|
They're always after me Lucky ROMS!
|
|
|
|
|
|
Reged: 03/07/04
|
|
Posts: 6472
|
|
|
|
|
|
Send PM
|
|
|
Reminder to John IV!
[Re: R. Belmont]
#393396 - 02/24/22 02:51 PM
|
|
|
> > Which games / driver families benefit from this? I didn't see an uptick on the
> bench
> > games w/ my 4 cores.
>
> Anything using poly_manager (Voodoo, Model 1/2/3, Naomi, the recently improved Konami
> 3D games). It used to peg out at 3 cores (it'd actually get slower at 4 and up). Now
> you can -numprocessors 8 (at least) if you have them and it'll give some
> improvements.
John IV - is there a way to squeeze this into your benchmarks charts? Have you started trying -numprocessors blah yet on the systems that use poly_manager?
Might also be fun to compare -numprocessors 1 ... -numprocessors 8 on your new Alder Lake!
- Stiletto
|
|
|
| John IV |
|
IV/Play, MAME, MAMEUI
|
|
|
|
|
|
Reged: 09/22/03
|
|
Posts: 1967
|
|
Loc: Washington, USA
|
|
|
|
Send PM
|
|
|
Re: Reminder to John IV!
[Re: Stiletto]
#393399 - 02/24/22 07:53 PM
|
|
|
Here's a quick look at blitz. Normally I run the benches close to mame defaults w/ -norc so -numprocessors is auto in that instance. Here we see manually setting to 3 gives the best results:
mame blitz -bench 90 -numprocessors 1
Average speed: 298.46% (89 seconds)
mame blitz -bench 90 -numprocessors 2
Average speed: 463.74% (89 seconds)
mame blitz -bench 90 -numprocessors 3
Average speed: 487.89% (89 seconds)
mame blitz -bench 90 -numprocessors 4
Average speed: 469.54% (89 seconds)
mame blitz -bench 90 -numprocessors 5
Average speed: 438.11% (89 seconds)
mame blitz -bench 90 -numprocessors 6
Average speed: 417.06% (89 seconds)
mame blitz -bench 90 -numprocessors 7
Average speed: 383.09% (89 seconds)
mame blitz -bench 90 -numprocessors 8
Average speed: 349.03% (89 seconds)
mame blitz -bench 90 -numprocessors auto
Average speed: 461.08% (89 seconds)
|
john iv
http://www.mameui.info/
|
|
| John IV |
|
IV/Play, MAME, MAMEUI
|
|
|
|
|
|
Reged: 09/22/03
|
|
Posts: 1967
|
|
Loc: Washington, USA
|
|
|
|
Send PM
|
|
|
|
Here's a bench run with numprocessors set to 1, 3, and 8. Percentages are against the 'numprocessors auto' column.
Pink Sweets is a cv1k game which they all die if numprocessor is set to 1. Mametesters 6828.
So for the most part auto handles things except for a bump in blitz and scud when using numprocessors 3 and similar oddities w/ numprocessors 1.
As usual this is via 'mame.exe gamename -bench 90' plus -numprocessors X
[ATTACHED IMAGE]
|
|
|
| Stiletto |
|
They're always after me Lucky ROMS!
|
|
|
|
|
|
Reged: 03/07/04
|
|
Posts: 6472
|
|
|
|
|
|
Send PM
|
|
|
Re: Reminder to John IV!
[Re: John IV]
#393402 - 02/24/22 10:11 PM
|
|
|
> Here's a bench run with numprocessors set to 1, 3, and 8. Percentages on the bottom
> are against the numprocessors auto column.
>
> Pink Sweets is a cv1k game which they all die if numprocessor is set to 1.
> Mametesters 6828.
> So for the most part auto handles things except for a bump in blitz, scud, and
> crusnusa when using numprocessors 3.
Thanks John IV!
If I am reading this right, numprocessors 1 got a couple of bumps too? The crusnusa, scud, vf2 ones for sure, the rest probably be statistically identical.
Weird, kinda missing the point of numprocessors with numprocessors 1!
- Stiletto
Edited by Stiletto (02/24/22 10:14 PM)
|
|
|
| John IV |
|
IV/Play, MAME, MAMEUI
|
|
|
|
|
|
Reged: 09/22/03
|
|
Posts: 1967
|
|
Loc: Washington, USA
|
|
|
|
Send PM
|
|
|
Re: Reminder to John IV!
[Re: Stiletto]
#393404 - 02/24/22 10:31 PM
|
|
|
Yes, you're right about the few bumps on np=1.
Wonder if the numprocessor 'auto' code can get some tweaks to account for the outliers.
|
|
|
Index 
Threaded 
Intel Alder Lake bencmarks in MAME 0.238
Reply
[Re: 
