..:: PCSX2 Forums ::..

Full Version: Gsdx future discussion
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Pages: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
(03-15-2016, 03:31 PM)gregory Wrote: [ -> ]Well I know that iGPU aren't the fast chip in the world. But my question was, "are all iGPU too slow" or only some them and so which one.

We have to remember it's not always a case of "too slow". I realise a lot of people come here moaning their "epic" pc can't run bust-a-move but there are some people out there who just want to use PCSX2 for the amazing feat of running PS2 games on a PC, so segmenting the emulator off to just users with quick hardware is really not a good move for us at all, we should have a spectrum of compatibility, even if it is hardware which isn't capable of emulating it well.
It was to quantify the iGPU number. iGPU is more than half of the market, so in this case it means half of the user still rely on DX9. But honestly I'm a bit suprised that DX9 is much faster than DX10.
I'll do more testing in a few minutes, but I noticed a 10fps drop when I went from dx9 to dx11 in Chessmaster.
(03-15-2016, 09:04 PM)gregory Wrote: [ -> ]Well I know that iGPU aren't the fast chip in the world. But my question was, "are all iGPU too slow" or only some them and so which one.

There are probably some higher end iGPUs like the Intel Iris stuff which are more capable of at least competing a bit with discrete ones while getting good increased resolution levels, so those may not be that slow. But they are probably much more rare than the typical Intel HD3000-5000 type iGPUs, which are generally too slow to upres on much but DX9 on more mid-end or upper-mid-end games. Low end games like most NIS games they're typically fine for some level of upres on even DX11 and I'd expect OGL as well.

Intel HD4600 can get 2x on a decent amount of games with DX9 and even 3x with some relatively low end games like Soul Nomad. Meanwhile 3x on DX11 with Soul Nomad will tend to yield slowdowns during large graphical effects (spells, skills, explosions and such), which is kind of jarring and would probably make someone choose DX9 since there aren't many graphical errors in that game. Although I didn't try OGL with that game I'd expect it to probably be the same or slower than DX11 at 3x.

And on some higher end games like the Xenosagas, HD4600 usually gets about a 10 fps boost in speed with DX9 over DX11 when uprezzing 2x, which is the difference between playable in DX9 (mostly full speed except with mechs) and not so playable in DX11 (in the 50s much of the time although some scenes will be full speed for some reason, probably less polys or something). Still, too many graphical bugs in the XS games to really play in DX9. Some other games around that level of hardware requirement but with less graphical bugs affecting those games might have DX9 benefit people though.

I don't know too much about the really old iGPUs though from the time before Intel introduced their HD line. Probably they're just stuck with 1x HW or SW if their processors can still handle PCSX2.

I hope that gives you at least somewhat of an idea where somewhat modern iGPUs are in relation to uprezzing games on PCSX2. And obviously SW is always the fallback if someone is desperate to play a game but is GPU limited even at 2x. But uprezzing is a nice feature that most people want to take advantage of.

By the way, if you do remove the ability to do a custom resolution as the opening post says might be needed, a 1.5x resolution option like Dolphin has might be to a decent amount of people's benefit, although I don't know if 1.5x would be messed up by the same phenomenon that would make custom resolution untenable, since it isn't just a straight doubling of the resolution like 2x, 3x, 4x, etc.

(03-15-2016, 09:04 PM)gregory Wrote: [ -> ]It was to quantify the iGPU number. iGPU is more than half of the market, so in this case it means half of the user still rely on DX9. But honestly I'm a bit suprised that DX9 is much faster than DX10.

Well it does depend on whether they want to upres (although I would assume most people probably want to upres as much as their hardware can handle). Some may be fine with native or prefer SW due to less graphical bugs even if they have to sacrifice uprezzing to use SW mode.

Also I'm not sure it's "much" faster, maybe 10 fps gain in many ordinary situations but even more fps gain it seems during heavy graphical effects like explosions, where DX11 may drop but DX9 may not. Discrete GPUs, something like explosions may not make too big a difference, but I notice it can really lead to heavy drops on iGPUs depending on the size of the effect. DX9 for some reason handles these effects better performance-wise without looking glitched.

Also I'm kind of surprised if 50% of people who use PCSX2 use an iGPU but I have almost no frame of reference at all with which to guess how many PCSX2 users use what type of GPU. I would think iGPU users would be less likely to post on the forums though.
Honestly it would be nice to have an openGL test with a driver that support fully gl4.5, just to see. And please test with this settings too (default is -1)
Quote:override_geometry_shader = 0
Geometry shader help to reduce memory bandwidth and a bit the CPU power, but it is quite taxing on the GPU. Lots of particle will likely requires more Geometry Shader power.

Note: I'm not against to keep dx9 on the main build if it is use by lots of users.

Edit: test also accurate date (sometimes it is slower, sometimes is could be faster)
Unfortunately I can't do the GL4.5 test since my driver is on very shaky ground. I can run the OGL plugin but a lot of functionality is not available and a lot of the niche GLextensions don't work (can't do stuff like accurate date and the like), so I assume Nobbs66 will be the one running that test.

This is just an example to show what I'm talking about with the explosion effects typically dropping fps a lot more on DX11 (and OGL) than on DX9. Uprez on this is 2x, CPU: i5 4440, GPU: Intel HD4600, Windows 8.1. Ingame (before the explosion, just wandering around) is at a few fps off from full speed on DX9 but a bit lower on 11, maybe around 10 fps lower on average. OGL ingame is at around 60% speed, so generally much lower than both. Pic is the same explosion at the same point in different plugins. Tried to get the screen to be at the same part of the explosion but it's a little off. Still, this was generally representative for the amount of the drop.

[attachment=59350]
(03-15-2016, 09:42 PM)gregory Wrote: [ -> ]Edit: test also accurate date (sometimes it is slower, sometimes is could be faster)
And sometimes it is a deathtrap/PCSX2 stall.
Since we're doing Xenosaga 2 benchmarks, here is the game with all 3 renderers on a GTX670 card. They perform more or less the same.
Just showing this so people don't get the impression DX9 is always (much) faster. It largely depends on the GPU Smile
[attachment=59351]
@BlackTelomeres,
Could you check the impact of the geometry shader?

Quote: And sometimes it is a deathtrap/PCSX2 stall.
Normally it mustn't be that slow. Not anymore.
Hum, I added extra computing on the Vertex Shader (for future) and in Fragment Shader (to improve rounding). Maybe it could explain the huge speed diff with openGL. Intel iGPU feels like a waste of silicon.
Pages: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15