Mini benchmark program

[||dav1de||]

Well it's just as easy for my CPU benchmarking routine to be in a function, and spawn that function multiple times in a threaded way. Since the data doesn't have to communicate, there is no complicated issue of race conditions or cross thread communication. I just invoke the function N times and wait til they all complete.

You didn't considered HT and CMT: if you take an i7 or an FX-8xxx if you give hime 8 thread and wait for the completion of the function, you'll have an higher time than if you give him 4 thread, because you won't have any performance loss due to resources sharing. Moreover, if enabled, Turbo Core/Boost DOES affect PCSX2 performances, so a good benchmark would not work in such a way to reduce its effect.

And just to understand...how can you really use PassMark to predict PCSX2 performances?

Example Passmark:
i7-3700k 2090pt
i7-4770k 2281pt
Haswell +9,1%

Example PCSX2
i5-4670k 4.1GHz 98,46FPS
i5-3570k 4.2GHz 86,72FPS
Haswell +13,5% (and haswell in this example has also a light frequency disadvantage, so at same frequency is something like +17%).

So PCSX2 and Passmark perfromances are not so well correlated.

Another example:
In Passmark an FX-9590 (4,7 to 5GHz) is slower than an i5-2500k, while in PCSX2 a 4.7GHz FX-8350 is not slower than that i5.

Maybe the difference is not extreme, but there's a difference. So if you want to have results consisten with the actual PCSX2 performances, use the PCSX2 CPU bench (or you may propose an updated version to see if thing changed with the latest versions).

[willkuer]

Well it is a starting point to use passmark. You also used only one game to compare benchmark results (probably FFX-2) and probably those two desktops haven't had the same hardware.
I think 9.1% is also quite close to 13.5%. And how you get to 17% I can also not directly see. And then you obviously have not even compared the same models but only the same generation step. So what is your actual point?

Also one should consider what was already said in this thread. It is about running the benchmark without using pcsx2 (especially the bios).

[||dav1de||]

Well it is a starting point to use passmark. You also used only one game to compare benchmark results (probably FFX-2) and probably those two desktops haven't had the same hardware.
I think 9.1% is also quite close to 13.5%. And how you get to 17% I can also not directly see. And then you obviously have not even compared the same models but only the same generation step. So what is your actual point?

Also one should consider what was already said in this thread. It is about running the benchmark without using pcsx2 (especially the bios).

I obtained 17% by estimating the performance of haswell at 4.2GHz (that is the performance at 4.1 multiplied by 4.2 divided by 41)
i7 and i5 differs only for HT and cache, that are not relevant when you consider ST performances in PCSX2. So if an i7 haswell is +17% than an i7 Ivy the same can be said for i5 Haswell vs i5 Ivy, for i3, Pentium, Celeron and so on. The point is that Passmark cannot give consistent results with PCSX2.

[Blyss Sarania]

In my experience, the single threaded performance on Passmark DOES heavily correlate to how well PCSX2 will run. For instance at a STP around 1800 almost everything is playable without speedhacks. At 1400, most games will work with a lot of tweaking, at 1000, only lighter games like FFX for instance will run, and of course 2 d games. Below that you are gonna be running 2d games only. And below 700 you can just forget it.

I thought about just using the passmark scores directly in my program, BUT I'm not sure that's allowed, I would have to ask Passmark themselves.

[||dav1de||]

In my experience, the single threaded performance on Passmark DOES heavily correlate to how well PCSX2 will run. For instance at a STP around 1800 almost everything is playable without speedhacks. At 1400, most games will work with a lot of tweaking, at 1000, only lighter games like FFX for instance will run, and of course 2 d games. Below that you are gonna be running 2d games only. And below 700 you can just forget it.

I thought about just using the passmark scores directly in my program, BUT I'm not sure that's allowed, I would have to ask Passmark themselves.

I would not call it a "heavy" correlation, it may work the way you used it right now, to make a rough estimation, but if you want to make something that say CPU A is better than CPU B, Passmark won't work well enough (I already gave you some examples).

Anyway I'm just saying that before starting to write the code, you should start with "accuracy" as your main objective, otherwise you could make an extremely simple software that takes the CPU architecture, frequency, and number of cores and makes an estimation based on that, for too old and too new CPUs it could just say "unsupported CPU" until you update it's simple database.

If you decide to go the "accuracy" way, I may suggest you to take a deep look at PCSX2 code, and use its main (and more performance limiting) functions, and base your code on it.

[Blyss Sarania]

I'm not good enough to base any code on PCSX2.

And that's why I said I was brainstorming because I'm not gonna do it unless I can do it right.

As far as the multithread issue, here is an idea. Except for GSdx software mode, PCSX2 uses 2 or 3 threads. So why not a two pass test, the first with 2, and if that CPU is not fast enough with only 2 threads, run a test with 3, and then recommend the MTVU speedhack if 3 threads are needed, and then include my own list of games that don't work well with MTVU as well.

[refraction]

what if it's a dual core cpu? MTVU might help the test but be detrimental to a lot of games.

[Blyss Sarania]

Well, if it's a dual core CPU, then don't run the 3 thread test? I'm sure it wouldn't be too difficult to add in some CPU detection.

[Blyss Sarania]

I just had an idea. It might be easier and more accurate to compile a list of CPUs that will work well with PCSX2, using Passmark and other data. Then just write a program to detect the CPU, compare it to that list, and make a suggestion.

[Nobbs66]

I just had an idea. It might be easier and more accurate to compile a list of CPUs that will work well with PCSX2, using Passmark and other data. Then just write a program to detect the CPU, compare it to that list, and make a suggestion.

Yeah, that would be easier, but wouldn't it just be easier to check which cpu you have and compare it to passmark.