How many threads does pcsx2 use? or can use?

[nosisab Ken Keleh]

I see that your FX is at default frequency with turbo enabled, thus using another core may force the CPU to work at a lower frequency and thus a small or even zero performance gain that the MTVU should give you in the specific situation becomes an overall decrease in performances due to the frequency.
The highest turbo for example works when 2 modules are idled, if you use 3 cores of 3 different modules...no more max turbo! If the 3rd core is in a module where there is the EE or the GS you may have performance losses due to the MTVU stealing resources to one of the other two threads.

What brings another interesting situation (not totally outside the thread topic).

I have a FX-6100 yet, it is waiting in it's box at the moment and was what I used before making the mistake for buying the FX-8150.

That's due I could easily overclock the FX-6100 to around 4.2GHZ and have reasonable temperature running prima 95 on all workers under the H-80 cooler.

That's not true with the FX-8150. The temperature rises too quickly if trying that same overclock to the point the liquid cooler is not enough to keep it down.

On the other side I found that using the default and intelligent turbo from the Asus M5A99-x Evo+AMD is enough to get all my games at ultra or near it, including demanding games like Skyrim, borderlands... and (arghh) Metro 2033 (although in this last case the 560 SLI seems to be the greater responsible, this game likes physX).

So, instead having 4.2GHz in all cores all the time the CPU has it where it is needed, if not for long that can be in all cores.

PS: Still I don't believe that is the cause of bad MTVU experience in this specific case, most probably due to the games I tested it not being the optimal to it. Even because aren't many the games (that I have) I can't run full speed all the time, so the games I tried it to get some performance was Zone of Enders and one specific scenery in Star Ocean: Till the End of Time. In both cases I still can get nominal speed pushing the speedhacks a notch more... so I really didn't bother much with the whole issue.

[||dav1de||]

What brings another interesting situation (not totally outside the thread topic).

I have a FX-6100 yet, it is waiting in it's box at the moment and was what I used before making the mistake for buying the FX-8150.

That's due I could easily overclock the FX-6100 to around 4.2GHZ and have reasonable temperature running prima 95 on all workers under the H-80 cooler.

That's not true with the FX-8150. The temperature rises too quickly if trying that same overclock to the point the liquid cooler is not enough to keep it down.

On the other side I found that using the default and intelligent turbo from the Asus M5A99-x Evo+AMD is enough to get all my games at ultra or near it, including demanding games like Skyrim, borderlands... and (arghh) Metro 2033 (although in this last case the 560 SLI seems to be the greater responsible, this game likes physX).

So, instead having 4.2GHz in all cores all the time the CPU has it where it is needed, if not for long that can be in all cores.

With PC games I think that you almost nevere reach the max turbo of your CPU, it will probably stay to the intermediate one (3.9GHz), but I can't be sure because on the FX-6xxx the highest turbo works when just one module is active (and the other two idled) so it's almost unaccessible. Anyway if you want to experience higher frequencies you can easily shut down some modules (or even one core for module), reaching higher frequencies with lower voltages.
Maybe we're too much off-topic now...

[nosisab Ken Keleh]

Yeah, but then I never stated MTVU would be bad for someones else machine, actually stated to try and keep if helping and disabling if not (that is valid for every speedhack).

Maybe, just maybe you are over weighting the gain MTVU is bringing to your machine + game combination or really getting good result with it. In my case, with the games I experimented it brought none, In zone of enders even EE cyclerate is bad news, and I can't grant is so in another machine also.

But then, all that time I'm trying to tell I have no problem with the setup as it is Those games run full speed, although I need to push VU cycle stealing to 2 with them, but, I bought Zone of Enders more to have it for testing purposes as well (and it was cheap in the shop's bargain bin) and did not play it beyond the very start I fear, more for lacking the time.

Now that I think about it, the only games including big robots I ever played whole was Xenosaga (if not counting FF VI).

[||dav1de||]

Yeah, but then I never stated MTVU would be bad for someones else machine, actually stated to try and keep if helping and disabling if not (that is valid for every speedhack).

Maybe, just maybe you are over weighting the gain MTVU is bringing to your machine + game combination or really getting good result with it. In my case, with the games I experimented it brought none, In zone of enders even EE cyclerate is bad news, and I can't grant is so in another machine also.

But then, all that time I'm trying to tell I have no problem with the setup as it is Those games run full speed, although I need to push VU cycle stealing to 2 with them, but, I bought Zone of Enders more to have it for testing purposes as well (and it was cheap in the shop's bargain bin) and did not play it beyond the very start I fear, more for lacking the time.

Ok but the way you tell about this problem seems that it's a well knwon and documented problem, instead, if I correctly understood, you're the only one who experienced it, and so I feel the urge to tell that it's not always true and I never experienced it. Anyway I'm not overweighting anything. I can tell you that I made some small test: +41% in FFX, from +3% to +45% in GT4 (respectively in a nightly track, that are probably bugged, and in a daily one), +20% in Naruto Shippuuden Ultimate Ninja 5, +14% in Xenosaga, +2% in RE: Code Veronica X. So sometimes the MTVU is almost useless, some other times is very effective.

[nosisab Ken Keleh]

Moved by curiosity made a kick test with Xenosaga III (at the very start, when you first gain control of Shion). Was needed to remove the frame limiter for that game runs almost all the time at steady 60 FPS.

Without MTVU ~ 86 FPS
With MTVU ~ 97 FPS

So, indeed there is a gain with this game, one I would never know if not removing that frame limiter.

Now, testing something that actually causes problems to me: Zone of Enders, no speedhacks (other than the basic), frame limiter off

the game's FPS varies too much from around 86 in the video like scenes and falling to bellow 30 in some moments when the machines are on the screen and staying bellow 50 all the time.

Without MTVU around 25 FPS at the wort moments

with MTVU fell below 20 FPS at some moments.

So, at least with the games I have been playing lately, MTVU was not needed, although I recon it could help benchmark with the framelimiter off in some games

So at lest for me MTVU is a mixed bag, I'll try it on those games where is needed some speed hack and observe if it does the job alone, if still needing the speedhacks and they are enough, better leaving MTVU out, or so I think. On the other hand, if needing, let's say VU cycle stealing at 2 without MTVU and being possible to have VU cyclestealing at 1 with it, MTVU becomes better option.

Edit: Maybe I must add that I don't see MTVU as a speedhack actually, where it works it is just a valid performance gain without the cons of actual speed hacks, so it is vastly preferable over the actual speed hacks where it brings actual performance gain.

By principle I use only the absolutely minimal necessary to have enjoyable experience and where possible, not mixing potential issues sources in the same bag. For example, reducing the virtual speed EE would run in actual PS2 mixed with VU cycle stealing increases a lot the chances of desync. If I can have only one the two, that is far better. Sadly there is not set rules, experimentation is still the best approach when dealing with speed hacks and that's dependent on the games.

[filenotfound]

Well this thread has alot of posts since i was last here XD

and to reply to earlier, yes i know extra rendering threads is only for software rendering (one reason why in the post i said software rendering threads and not extra rendering threads ),
and the fact it says software rendering mode options and hardware rendering mode options, should make it pretty obvious for everyone on which settings are for what rendering mode.
ok i take that back, most people dont know how to read, but ya XD

The fact that many games require software rendering for one reason or another, its kind of something worth knowing.

And with what refraction said, do you really need to have a core for each thread to get the benefit of it?

because i have an i5-2450M ( http://ark.intel.com/products/53452/ ), and its 2 cores with 4 threads.
and my laptop has an amd 1GB 128bit gddr3 7670m graphics card.

And i dont really gain or lose any fps (did it w/o frame limiting) with it on or off. On maybe be 1 fps less, but its hard to tell since its such a low value its almost negligible.
And the reason i think that its give no speed up or a very small speed loss is not due to it only being 2 cores, but because its a laptop cpu. It is clocked at too low of a clock speed. And the only reason i can even play at full or close to full speed in most games is due to turbo boost kicking in.

And since splitting the micro VU across the extra threads, it is able to increase the performance, BUT since im relying so heavily on turbo boost, which lowers due to the extra load on the threads, is why i get no actual speed benefit from it.


Now not really talking about my situation but what about say a core i3-2xxx/3xxx desktop cpu. Which are clocked at 3.0-3.2ghz at stock with 2cores/4threads.

Now would that gain performance by enabling MTVU? i think it would, but ofcourse am unable to test that lol

Also what about software rendering threads? in the situation above with an i3, would enabling 1-2 rendering threads on a 2core/4thread i3 help? or hinder it?


EDIT: hmmm ok well it seems i should have done better testing
after posting i decided to go more into it with testing on my cpu.

I turned off the pxsc2fx shader and i set the internal resolution to native, and closed most everything else that was running other then pcsx2.
Also turned off frame limiting.

it now seems in both hardware and software rendering modes, enabling MTVU does provide a speed up. Its anywhere from 15-25% increase in fps.

With MTVU on in software rendering mode, first had software rendering threads at 0. Put it to 1, it increased fps by about 5% or so. I increase it to 2, it increase fps by about another 5%.
BUT in one test the sound was distorted (crackle/pops) when enabling 2 software rendering threads, but it only happened 1 of the times i enabled 2 threads, so may it was some kind of fluke or something was happening at the time in windows in the background or something.

so i think just having extra threads (w/o cores) does definitely provide a speed boost when enabling MTVU aswell as software rendering threads.

But still tested again, it seems when enabling 2x native resolution, it gives almost no benefit at all to having MTVU on.
its really weird, and i cant quite understand it.
MTVU off. dx11 hardware mode, 1/2 texture filtering or w/e the box is, and native resolution
74-77 FPS
MTVU on. dx11 hardware mode, 1/2 texture filtering or w/e the box is, and native resolution
85-88 FPS
MTVU off. dx11 hardware mode, 1/2 texture filtering or w/e the box is, and 2x native resolution
74-77 FPS
MTVU on. dx11 hardware mode, 1/2 texture filtering or w/e the box is, and 2x native resolution
74-77 FPS

tho i was only testing with 1 game, dark cloud 2. i may go test on more games now, but wanted to add a update to my post.

[nosisab Ken Keleh]

yes, you got the spirit, with speedhacks the best is experimenting, although people can say to you what works better in their machine that may not be the best in yours (or too much or not enough, or maybe the best possible setup indeed, everything is possible)

But... (there is always a but), my advice is testing with the framelimiter on, the actual gaming situation, the principle is solid, if not needing, don't use. That is specially valid for the actual speed hacks, although your CPU is not ideal for MTVU, where it helps it is far preferable over speedhacks.

Edit: a brief and general vision about speedhacks and what they do, not necessarily directed to the OP alone.

Code:

MTVU - Not really a speed hack, what it does is dividing the load among cores, so it does not introduce errors by itself (except if plain crashing). Should be the first attempt always. It is specially effective when the VUs are the most overloaded and a minor gain (or none) if the VUs module is already less loaded than EE module itself, as to say it helps less when EE is the real bottleneck.

VU cycle stealing - it tries to get the most steady data flow for the VUs processing forcing the interruption of EE normal processing instead waiting it's turn to be served by it (so effectively "stealing" cycles EE could be using to process something else). at less aggressive set it works fine without introducing greater issues and actually providing a huge boost in performance in most cases. If abused that may cause EE to stall and then the game becomes laggy and sluggish what is the first symptom of it being in excess.

EE cyclerate - that one is aggressive by nature, it works on the principle that a slower machine is easier to emulate. So what it does is reducing the clock the actual EE (PS2 CPU) would be working, as to say like underclocking actual PS2. The main symptom it being in excess is the game becoming desynchronized, mostly perceivable in the sound but may be noticeable as wrong response time to controllers and such. At it's first point (appearing as 2 in the setup) it works well for most games, at the extreme point it will break almost all games out there. I only use EE cyclerate when the other approaches failed.

So, if the game is presenting some glitch or oddness, those symptoms may help deciding what to try and reduce first and even to decide if the collateral effect is yet better than the raw low FPS from not having speedhacks.

PS: Both MTVU ans VU cycle stealing are based in the principle that graphics (meaning vector processing) is most of times more demanding on the emulation than the EE normal processing, and they are less effective in those games which forces exceptional EE processing, physics comes into mind hence games including big robots or racing games are harder on the emulator.

Edited to put into code window what is not essential for the reply and can be freely skipped by most casual users.

[refraction]

Mtvu is only a speedhack due to its implementation amd how the vu works. By design it is considered a non-valid way of handling the vu and is prome to problems. This is why it's a speedhack and not a normal feature/option

[nosisab Ken Keleh]

Mtvu is only a speedhack due to its implementation amd how the vu works. By design it is considered a non-valid way of handling the vu and is prome to problems. This is why it's a speedhack and not a normal feature/option

Yes, I do agree it is not (at least yet) to be put as normal feature, what I meant with not considering it a speedhack in the normal sense is due to where it works it does not introduce collateral issues by itself (but since it may fail in some situations is understandable the option to disable it and the advice not using it like any other speedhack where not needed). PS: Or so it should be in an ideal world, the mention of non-valid way to handle the VU suggest more than simple attempt in making the module multithreaded.