Automated Windows x64 builds?

[TheDax]

Is there any particular reason why Buildbot doesn't offer 64 bit versions of PPSSPP? Just curious.

Note that it's not really an issue for me personally, but to the end user who might want new x64 builds, it could be an issue..

[Ritori]

So dont has any 64 bit build?
Atleast 64 bit user can play 32 bit ppsspp

[Tyestor]

(06-04-2013 01:30 AM)Ritori Wrote:  So dont has any 64 bit build?
Astlest 64 bit user can play 32 bit ppsspp

What?

[Ritori]

(06-04-2013 04:04 AM)Tyestor Wrote:  

(06-04-2013 01:30 AM)Ritori Wrote:  So dont has any 64 bit build?
Astlest 64 bit user can play 32 bit ppsspp

What?

nothing wrong right

[TheDax]

He's saying this:

"So Orphis/buildbot doesn't have 64 bit builds? At least x64 users can play games on 32bit PPSSPP", more or less.

Just use your brain a bit. His English isn't the best, but it's understandable.

[Ritori]

(06-04-2013 04:11 AM)daxtsu Wrote:  He's saying this:

"So Orphis/buildbot doesn't have 64 bit builds? At least x64 users can play games on 32bit PPSSPP", more or less.

Just use your brain a bit. His English isn't the best, but it's understandable.

Ahahahahaha.... sorry, i am learn a bit now...

[TheDax]

No worries.

[Ritori]

I no worry at all it embrassing me T_T
is any people can read well but cant write good enough

sorry if i offtopic

[Orphis]

I could easily build 64bit. It's not really worth it though.

[TheDax]

The same could be said for JPCSP's 64 bit builds, no?

[solarmystic]

It seems like it could be beneficial to have it for a certain subset of users like myself.

Background:-

I was having a discussion with daxtsu about potential performance differences between 32bit builds and 64bit builds. We came to a preliminary understanding that it shouldn't make that much of a difference in real world performance, i.e. performance should be within 0-5% for the same build whether compiled in 32 bit or 64 bit when ran in a 64bit environment.

I was curious anyway to see if that understanding would hold up when applied in testing.

The Test:-

Procedure:-

The main procedure adhered to is the one outlined in daxtsu's benchmarking PPSSPP with cube.iso thread:-

http://forums.ppsspp.org/showthread.php?...4#pid26564

All PPSSPP settings are adjusted according to that post.

The candidates:-

I had daxtsu provide me with two of his own compiled builds based on the latest master, (1140), with no additional modifications. One 32 bit build and one 64 bit build was used for this test.

The rig used for testing is as follows:-

Intel Core 2 Duo T9550 @ 2.8 GHz
ATI Mobility Radeon 4670 1 GB GDDR3 @ 843/882 MHz
8 GB DDR3 RAM
Windows 7 Professional SP1 x64

The result:-

Using the provided 32 bit build, the rig averaged 918 VPS in the cube.iso benchmark, as compared to the 1449 VPS obtained using the 64 bit build.

32bit build



64bit build



The findings were well beyond the expected deviation in performance between the 2 builds (0-5%), so the tests were repeated 5 more times to ensure that the figures were not in doubt.

Additional related findings/Possible explanation:-

It seems like using the 64bit build yielded an increased VPS due the fact that the GPU was being completely utilized (leading to 1440 VPS) as compared to the 65-66% GPU utilization (which results in 9xx VPS) when the 32 bit build was used, as observed using GPUz's GPU utilization counter.

I speculate that the use of the 32 bit build brings along with it increased CPU overhead as observed from the Task Manager for this particular set of hardware, which prevents the GPU from working at 100% capacity for the same cube.iso benchmark.

PPSSPP showed 100% utilization of a single core (50% out of two, effectively) for the 32 bit build, meaning it was CPU limited even before the GPU could be fully pushed. The opposite is true for the 64 bit build.

32bit CPU utilization with Unlimited VPS:-



64bit CPU utilization with Unlimited VPS:-




Conclusion:-

It seems that for this one particular user scenario, the use of a 64 bit build brings along with it a tangible increase in performance that could possibly benefit others with similar hardware configurations.

[[Unknown]]

This could even be a better 64-bit driver for your video card, I think.

Anyway, I have to say cube is not representative of every game. For example, in Kingdom Hearts at the first playable scene in a new game, 32-bit gives me 503 vps, and 64-bit gives me 499 vps. I've seen 64-bit better for other games, so it just depends. I've never seen such a large difference in a real game, though.

Synthetic benchmarks are, as always, lies.

-[Unknown]

[Tyestor]

(06-06-2013 03:52 AM)[Unknown] Wrote:  This could even be a better 64-bit driver for your video card, I think.

Anyway, I have to say cube is not representative of every game. For example, in Kingdom Hearts at the first playable scene in a new game, 32-bit gives me 503 vps, and 64-bit gives me 499 vps. I've seen 64-bit better for other games, so it just depends. I've never seen such a large difference in a real game, though.

Synthetic benchmarks are, as always, lies.

-[Unknown]

Well...that's not really much of an issue, is it? What games would you want to play over 60vps? Speaking of Kingdom Hearts, there were problems a little while back that only appeared on 32-bit builds but not on 64-bit builds.

This site does daily/nightly 64-bit builds: http://www.emucr.com/search/label/PSP?&max-results=12 - It's too slow though Orphis' buildbot checks for builds every 10 minutes and there's almost always more than one build per day.

[solarmystic]

(06-06-2013 03:52 AM)[Unknown] Wrote:  This could even be a better 64-bit driver for your video card, I think.

Anyway, I have to say cube is not representative of every game. For example, in Kingdom Hearts at the first playable scene in a new game, 32-bit gives me 503 vps, and 64-bit gives me 499 vps. I've seen 64-bit better for other games, so it just depends. I've never seen such a large difference in a real game, though.

Synthetic benchmarks are, as always, lies.

-[Unknown]

You are of course, correct.

I redid the test for actual games, and the results are mostly within the expected deviation.

Games which offer significant performance increases in 64bit version are often those that aren't CPU limited/constricted to begin with at 2x RR with unlimited VPS.

Also, performance differences between the 2 builds are often negated when the emulator is run at a higher Rendering Resolution (4x RR or 1080p fullscreen), so that the situation becomes effectively GPU limited instead of CPU limited.

My point was that the 64bit build innately offers lower CPU overhead for that particular test, and it seems to be true when other games are tested at that same 2x RR which means that the GPU could be utilized fully at those given settings instead of being bottlenecked by the CPU.

You are obviously not going to notice it with your monster i7 rigs which have tonnes of the CPU power to spare.

[Orphis]

Unless you want to run the game at 20x the regular speed, it's then pointless to use a 64bit build since 32bit is perfectly working fine and there's no compelling reason to use it. It's useful for some other emu (Dolphin), not so much here for our low memory handheld console.