|
Any Ryzen 3000 series of benchmarks for MAME?
#384326 - 11/21/19 07:59 PM
|
|
|
I'm still rocking an old Core i5-2500k (albeit at 4.4ghz, so it's still plenty usable), but I've been thinking that my system is starting to get a bit long in the tooth and I wasn't sure if Intel's offerings still trumped AMD's for emulation or not.
I probably wouldn't even be getting a Ryzen process until the 4000 series (which apparently is a new architecture so the comparison to 3000 series benchmarks may be moot), but I wasn't sure if anyone can vouch for the Ryzen 3000 series in the most demanding scenarios (and anyone have any a Cascade Lake Intel CPU to compare?).
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Pr3tty F1y]
#384328 - 11/22/19 12:13 PM
|
|
|
Given that MAME is (unless something has changed recently) still heavily reliant on single-threaded workloads, the Core i9-9900KS would probably be the ultimate CPU for the job (it has the best single core performance of any x86 CPU ever made) Although there are probably very few games in MAME capable of running at playable speed on a Core i9-9900KS (or other similarly highly-clocked CPUs) but that can't run on something more modest like a Ryzen 7 3700X or 3800X.
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Pr3tty F1y]
#384330 - 11/22/19 02:26 PM
|
|
|
If it's not playable at full speed on a Ryzen 3700/3800/3900/3950...it won't be playable at full speed on Intels fastest CPU's either.
As someone who runs a 9900K (@5Ghz)the games that can't run at full speed are a LONG way off from being able to do so, such as the likes of Star Soldier & Naomi titles, barely run much past 35fps whilst in game.
..even turning off Hyper Threading and pushing clockspeeds upto 5.2Ghz, only adds a couple of FPS to them.
The difference in IPC between Ryzen & Intel is now so small as to make no tangible difference.
Edited by Ziggy100 (11/22/19 02:27 PM)
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: jonwil]
#384341 - 11/23/19 10:33 PM
|
|
|
> Given that MAME is (unless something has changed recently) still heavily reliant on > single-threaded workloads, the Core i9-9900KS would probably be the ultimate CPU for > the job (it has the best single core performance of any x86 CPU ever made) > Although there are probably very few games in MAME capable of running at playable > speed on a Core i9-9900KS (or other similarly highly-clocked CPUs) but that can't run > on something more modest like a Ryzen 7 3700X or 3800X.
Thanks for the recommendation and the thoughts regarding the comparison. It looks like I should be able to save some $$$ going with AMD.
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Ziggy100]
#384342 - 11/23/19 10:35 PM
|
|
|
> If it's not playable at full speed on a Ryzen 3700/3800/3900/3950...it won't be > playable at full speed on Intels fastest CPU's either. > > As someone who runs a 9900K (@5Ghz)the games that can't run at full speed are a LONG > way off from being able to do so, such as the likes of Star Soldier & Naomi titles, > barely run much past 35fps whilst in game. > > ..even turning off Hyper Threading and pushing clockspeeds upto 5.2Ghz, only adds a > couple of FPS to them. > > The difference in IPC between Ryzen & Intel is now so small as to make no tangible > difference.
Thanks much for being able to vouch for the 9900K. It sounds like the consensus is any fast chip will be fast enough..
However, if you don't mind me asking, how does N64/Aleck64 based hardware work on your 9900K? Can it manage a stable 60fps?
|
|
|
John IV |
IV/Play, MAME, MAMEUI
|
|
|
Reged: 09/22/03
|
Posts: 1981
|
Loc: Washington, USA
|
|
Send PM
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Pr3tty F1y]
#384343 - 11/24/19 01:38 AM
|
|
|
Used Super Mario 64, 42% w/ daily build, on 6700K at 4.5Ghz. So figure the 5ghz would be about 47% (assuming the bench mark scaling from the C2D era I found continues to hold).
c:\O\Games\MAME>mame n64 sm64 -bench 90 Average speed: 42.42% (89 seconds)
Note I also bench Star Soldier every build. http://www.mameui.info/Bench.htm
Edited by John IV (11/24/19 08:32 PM)
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Pr3tty F1y]
#384345 - 11/24/19 12:40 PM
|
|
|
Quote:
if you don't mind me asking, how does N64/Aleck64 based hardware work on your 9900K? Can it manage a stable 60fps?
Star Soldier is N64/Aleck64 hardware and it doesn't even come close to 60fps. Framerate varies between a low of 28fps and high of 45fps whilst in game. It all depends on whats being rendered on screen at the time.
But you can be assured that the latest Ryzen CPU's will be pretty identical to this, as long as you run them at 4Ghz+.
The days of Intel CPU's having upto a 30-50% performance advantage in MAME over AMD CPU's are (thankfully) long gone.
If I get the chance, I'll post some screenshots of it running on my PC to show you.
Edited by Ziggy100 (11/24/19 12:45 PM)
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Ziggy100]
#384346 - 11/24/19 01:04 PM
|
|
|
> if you don't mind me asking, how does N64/Aleck64 based hardware work on your 9900K? > Can it manage a stable 60fps? > > > Star Soldier is N64/Aleck64 hardware and it doesn't even come close to 60fps. > Framerate varies between a low of 28fps and high of 45fps whilst in game. It all > depends on whats being rendered on screen at the time. > > But you can be assured that the latest Ryzen CPU's will be pretty identical to this, > as long as you run them at 4Ghz+. > > The days of Intel CPU's having upto a 30-50% performance advantage in MAME over AMD > CPU's are (thankfully) long gone. > > If I get the chance, I'll post some screenshots of it running on my PC to show you.
Star Soldier is 50% ingame on my now many years old i7-4790K at stock speeds, which some people suggest is now quite behind the performance curve.
I imagine we've come somewhere since then?
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: John IV]
#384349 - 11/24/19 03:54 PM
|
|
|
> Used Super Mario 64, 42% w/ daily build, on 6600K at 4.5Ghz. So figure the 5ghz would > be about 47% (assuming the bench mark scaling from the C2D era I found continues to > hold). > > c:\O\Games\MAME>mame n64 sm64 -bench 90 > Average speed: 42.42% (89 seconds) > > Note I also bench Star Soldier every build. > http://www.mameui.info/Bench.htm
Well, totally beats the pants off of my 2500k @ 4.4ghz which is looking much older than I thought:
C:\Emulators\Mame64>mame64.exe n64 -cart "d:\Roms\Nintendo - N64\Super Mario 64 (USA).7z" -bench 90 Average speed: 26.53% (89 seconds)
C:\Emulators\Mame64>mame64.exe n64 -cart "d:\Roms\Nintendo - N64\Super Mario 64 (USA).7z" -bench 90 Average speed: 27.28% (89 seconds)
Then again, I'm wondering how much the CPU bug/security mitigations have affected performance (although I thought they mostly impacted IO, not just straight number crunching).
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Pr3tty F1y]
#384396 - 11/26/19 04:36 PM
|
|
|
> Then again, I'm wondering how much the CPU bug/security mitigations have affected > performance (although I thought they mostly impacted IO, not just straight number > crunching).
That is indeed a very interesting question. Maybe you dont know it, but there are different Windows 10 types/builds. AFAIK only version 1607 of Windows 10 has no micro-updates and no CPU bug/security mitigations (Spectre etc.). So that is the only version, that keeps the CPU unaffected/unthrottled.
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: uman]
#384397 - 11/26/19 05:09 PM
|
|
|
> > Then again, I'm wondering how much the CPU bug/security mitigations have affected > > performance (although I thought they mostly impacted IO, not just straight number > > crunching). > > That is indeed a very interesting question. Maybe you dont know it, but there are > different Windows 10 types/builds. AFAIK only version 1607 of Windows 10 has no > micro-updates and no CPU bug/security mitigations (Spectre etc.). So that is the only > version, that keeps the CPU unaffected/unthrottled.
Windows doesn't load CPU microcode updates. If you're running Windows, any CPU microcode updates you want to use need to be applied by the BIOS or an EFI module. (Linux does load microcode updates itself.)
Spectre mitigations do reduce performance of system calls. The "variant 2" mitigations tend to have the biggest impact. It's possible to disable "variant 2" mitigation on 1607, 1703 and 1709 with this update: https://support.microsoft.com/en-us/help...ectre-variant-2
There's also information on how to disable some Spectre mitigations with registry settings here: https://support.microsoft.com/en-us/help...nerabilities-in
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Vas Crabb]
#384398 - 11/26/19 06:02 PM
|
|
|
First i thank you, for these infos. At least i was partially right (or wrong, depending on your view). I dont trust any Microsoft infos anymore, simply because i was a person who was affected from a (automatically) update in January 2018. My machine was in a infinite boot-trap after the update and i could not use it for nearly six weeks. No matter if you did a fresh install or trying to repair the old system. The only solution was keeping the machine offline, as you would avoid any Microsoft update. The support was awfully bad, treating many customers with the same problem like shit and even told them that it is the customers fault, which clearly was not the case. They did this, simply by the fact, that they didnt have a machine with the same problem.
I found my solution on a german IT website, which was using 1607 and using ultimate or enterprise editions of Windows 10 , because you can avoid updates (for a month or two). I use this solution till today without any problems. > Windows doesn't load CPU microcode updates. If you're running Windows, any CPU > microcode updates you want to use need to be applied by the BIOS or an EFI module. > (Linux does load microcode updates itself.)
I did not know this, but if this is true, i can only say that till today there is no ASUS Bios update for my X99 board, that would do a CPU microcode update. > Spectre mitigations do reduce performance of system calls. The "variant 2" > mitigations tend to have the biggest impact. It's possible to disable "variant 2" > mitigation on 1607, 1703 and 1709 with this update: > https://support.microsoft.com/en-us/help...ectre-variant-2 > > There's also information on how to disable some Spectre mitigations with registry > settings here: > https://support.microsoft.com/en-us/help...nerabilities-in
see above. you can trust this... i will not. 1703 and 1709 is just a question of time, when Microsoft will do that in a silent update. AFAIK 1709 is already patched. 1607 is the only version, where Microsoft said, it will not happen.
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Vas Crabb]
#384441 - 11/28/19 09:09 PM
|
|
|
Quote:
Windows doesn't load CPU microcode updates. If you're running Windows, any CPU microcode updates you want to use need to be applied by the BIOS or an EFI module. (Linux does load microcode updates itself.)
That appears, based on what Microsoft themselves are saying, to be incorrect.
|
---
Try checking the MAME manual at http://docs.mamedev.org
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Firehawke]
#384451 - 11/29/19 02:09 AM
|
|
|
> That appears, based on what Microsoft themselves are saying, to be incorrect.
Interesting. Previously they expected it to be done outside the OS.
|
|
|
|
Re: Any Ryzen 3000 series of benchmarks for MAME?
[Re: Vas Crabb]
#384452 - 11/29/19 02:13 AM
|
|
|
Spectre is certainly scary enough to make Microsoft feel the need to change policy since many motherboard manufacturers refused to do anything at all for older boards.
|
---
Try checking the MAME manual at http://docs.mamedev.org
|
|