TR Forums

Here are some screenshots comparisons of Win 7 DX 11 vs oGL 4 on Unigine Heaven in openSUSE that were promised today:

Album http://imageshack.us/g/99/win7dx11score.jpg/

http://img525.imageshack.us/img525/6964/save1dx11.jpg
http://img100.imageshack.us/img100/9387/save1ogl.jpg

http://img580.imageshack.us/img580/33/save2dx11.jpg
http://img843.imageshack.us/img843/2940/save2ogl.jpg

http://img515.imageshack.us/img515/6773/save3dx11.jpg
http://img39.imageshack.us/img39/2810/save3ogl.jpg

http://img716.imageshack.us/img716/2383/save4dx11.jpg
http://img705.imageshack.us/img705/3696/save4ogl.jpg

http://img521.imageshack.us/img521/9978/save5dx11.jpg
http://img94.imageshack.us/img94/4455/save5ogl.jpg

And the final benchmark scores:

http://img99.imageshack.us/img99/6086/win7dx11score.jpg
http://img812.imageshack.us/img812/3670 ... lscore.jpg


As you can see, oGL is every bit as capable as DX 11 in terms of quality and speed.

Another myth busted.

Original files with md5sums

http://www.filedropper.com/screenshots

Cute and all, but has anyone cared since DX9? Khronos screwed up with OpenGL 3.0 which had all of its issues, and was released far too long after 2.0 and lost its way. Quake III was really the halcyon days for OpenGL. Going to take quite a lot to regain mindshare in gaming. It will still own its place in Workstations, though.

TheEmrys wrote:

Cute and all, but has anyone cared since DX9? Khronos screwed up with OpenGL 3.0 which had all of its issues, and was released far too long after 2.0 and lost its way. Quake III was really the halcyon days for OpenGL. Going to take quite a lot to regain mindshare in gaming. It will still own its place in Workstations, though.

Perhaps you missed this:

http://techreport.com/discussions.x/23355

As you can see, oGL is every bit as capable as DX 11 in terms of quality and speed.

I agree, but that's nvidia. AMD's not so great at OpenGL performance, and linux GL is probably worse. Not that I don't think there could be improvements. I certainly hope there will be anyway.

I think there are only a few people who think this debate is 'great,' one of them being the thread starter.

Here's my take on it: Windows can run both DX and OpenGL, thus giving maximum flexibility as far as what software can be run.

l33t-g4m3r wrote:

As you can see, oGL is every bit as capable as DX 11 in terms of quality and speed.

I agree, but that's nvidia. AMD's not so great at OpenGL performance, and linux GL is probably worse. Not that I don't think there could be improvements. I certainly hope there will be anyway.

That was the case years ago but AMD's blobs for about the last 2 years have been able to keep up just fine.

Deanjo wrote:

That was the case years ago but AMD's blobs for about the last 2 years have been able to keep up just fine.

Really? Good news then. I've heard the 6 series works ok, but not sure on the 7's. How about a mini distro review of openSUSE for the curious while you're at it. I'm considering dual booting with my spare 939 system to try out one of the newer distros, just need some time for it. Gonna redo my main rig with a ssd and win7 first though.

l33t-g4m3r wrote:

Really? Good news then. I've heard the 6 series works ok, but not sure on the 7's.

Do you mean the Radeon 7000 series? We just got a Radeon 7970 at work with Kubuntu 12.04 and it works fine for general desktop use with the catalyst drivers. On the other hand, I have not tried things like dual-head setups or video playback acceleration nor have i tested OpenGL performance. Using it primarily for OpenCL development and for that it works fine.

Deanjo wrote:

TheEmrys wrote:

Cute and all, but has anyone cared since DX9? Khronos screwed up with OpenGL 3.0 which had all of its issues, and was released far too long after 2.0 and lost its way. Quake III was really the halcyon days for OpenGL. Going to take quite a lot to regain mindshare in gaming. It will still own its place in Workstations, though.

Perhaps you missed this:

http://techreport.com/discussions.x/23355

This thread for some reason reminded me of the famous xkcd comic:
http://xkcd.com/386/

Banding and over dithered in the bricks and shadows...

Not as close as I hoped to the textures... but, the 1st set does show what i was looking for, banding and over dithering in the shadows of the bricks in the ogl shots. possibly bandwidth saving sacrifices possibly not. but since i'm not a dev, I will say as I did before i'm holding judgment until TR reviews the l4d port since unigine and l4d are different engines.

As for the other comment of trolling, suck it up. it is not my intent to lay bait, it is a comment on the past efforts by the Kronos group falling a little short while tooting a horn. If a little criticism and commentary on past events makes you call me a troll and "nananananana" .... then who is the fanboi?

l33t-g4m3r wrote:

Really? Good news then. I've heard the 6 series works ok, but not sure on the 7's.

When it comes to the Catalyst drivers, aka "the blobs", AMD has been doing a very good job for the last few years with supporting them in a timely manner. Again because much of the code base is shared with windows version, the level of support is near identical. The one thing that still does effect the Catalyst drivers is that they are slower then Nvidia to support newer revisions of the kernel,xorg, etc and usually takes a couple of months before they support those. With Nvidia that support is usually there when the new revisions go final or sometimes earlier (for example the latest Nvidia 304.30 drivers already support the X.Org Server 1.13 ABI even though it is not set to be released until Sept.)

How about a mini distro review of openSUSE for the curious while you're at it.

That has been actually on my "todo" list for a while. I will probably get around to it this month and hopefully have it ready for the 12.2 release. It would concentrate more on how openSUSE is different then other distros and focus more on those specific items that differentiates it from other distros such as YaST, zypper, the build service, KDE, etc. It would be something different then usual Ubuntu/Unity reviews that saturate the web.

/Sheldon/ Sarcasm ?

JohnC wrote:

Deanjo wrote:

TheEmrys wrote:

Cute and all, but has anyone cared since DX9? Khronos screwed up with OpenGL 3.0 which had all of its issues, and was released far too long after 2.0 and lost its way. Quake III was really the halcyon days for OpenGL. Going to take quite a lot to regain mindshare in gaming. It will still own its place in Workstations, though.

Perhaps you missed this:

http://techreport.com/discussions.x/23355

This thread for some reason reminded me of the famous xkcd comic:
http://xkcd.com/386/

wth is with the terrible edge aliasing in the ogl shots? im thinking jpg compression did not do the shots any favors... in fact in the shots where the sky box meets any world objects there are tons of compression artifacts...

maxxcool wrote:

wth is with the terrible edge aliasing in the ogl shots? im thinking jpg compression did not do the shots any favors... in fact in the shots where the sky box meets any world objects there are tons of compression artifacts...

Umm, ya think you are seeing what you wish to see. If there was "banding" and "edge aliasing" you would clearly see it in the diff.

http://imageshack.us/f/594/differencef.png/

nm, look at the grass... this is a terrible demo to do this comparison on. Their render techniques seem to skim in places the user is not supposed to look.
the grass is all blocky and terribly aliased...

19.9073 % difference with the majority of that comes from 2 factors:

1) the system settings on the pictures
2) the ever so slight difference of the lighting position (really hard to capture the exact same light position with the benchmark running a virtual day)

There is next to no difference in the items you think you are seeing.

Again though... moot. l4d engine <> unigine. Neat, not so pretty but not the same fruit. TR and Anand will settle it.

maxxcool wrote:

Again though... moot. l4d engine <> unigine.

While I agreed initially that there *can* be differences in rendering between OGL and DX, you're starting to make me regret it. There are some artifacts that really don't show themselves in stills but do in motion. But, it does sound like developers have made strides with OGL drivers.

I'd suspect that if OGL experiences a resurgence in gaming, there will be a lot of issues at first (see Rage for instance), but that they'll be ironed out fairly quickly. My stance was basically that because developers rarely use OGL for games, vendors don't worry too much about the performance and visual quality of them, but it's possible (likely?) that Linux adoption has spurred more interest, and Steam's Linux support will surely help as well.

Having said that, there might be no difference at all now. There certainly was in the late 90s/early 2000s. The advantage of developers only using DX is that the driver teams really only have to focus on one set of drivers for gaming. The downside is that according to people who know (aka John Carmack) OGL has a bit less overhead, is quicker to adopt feature requests, and is cross-platform.

I don't really think there's a great debate about it with the exception of a few who, for their own reasons, believe that OGL's rendering is or will be inferior without much in the way of facts to support it. OGL has basically been on the sidelines, so there was nothing to debate (I realize there are nongaming applications for OGL and they were actually the primary use).

If OGL game(s) become the primary cross-platform benchmark though, I could see some questionable "optimizations" by hardware vendors just as they have done with DX drivers in the past.

Jason181 wrote:

I'd suspect that if OGL experiences a resurgence in gaming, there will be a lot of issues at first (see Rage for instance), but that they'll be ironed out fairly quickly. My stance was basically that because developers rarely use OGL for games, vendors don't worry too much about the performance and visual quality of them, but it's possible (likely?) that Linux adoption has spurred more interest, and Steam's Linux support will surely help as well.

+1. I don't think there is much technically to differentiate between GL and DX APIs nowadays. It comes down to driver implementation, and driver implementations will improve over time if more games use them.

That said, driver development for DX is definitely a bit simpler than OGL. For example, for OGL, every vendor essentially ends up writing their own frontend for things like parsing GLSL, whereas for DX a lot of the frontend machinery is actually provided by Microsoft. HLSL shaders get compiled to optimized DX assembler entirely by Microsoft's tools, and driver writers only need to compile the DX assembler to GPU ISA, which is a lot simpler and less error-prone than compiling something high-level like GLSL. In OpenGL, every vendor has to do GLSL->Proprietary IL->GPU ISA. Thus, in OpenGL there is more duplicated effort by the vendors and with slightly different frontends, there are more chances of bugs or incompatibilities.

(Disclaimer: I may or may not know what I am talking about.)

The screens seem to differ:
1) Displacement mapping is more pronounced in OGL
2) Gamma is washed out in OGL (could be SUSEs fault, Linux seems to be always washed out overall)

The performance in OGL seems to be better, min FPS are higher by 2. Unfortunatelly no idea what happens with max FPS, as OGL is VSYNC capped. Can you rerun the test with VSYNC off? It's interesting if max FPS are higher in Linux too, and if so, by how much.

Either way, 10 vs 12 min FPS shows that OGL is quite faster, which reinforces the results Valve got from L4D2.

I would love the fix for OGL gamma though (or more likely, Linux gamma). Maybe they use the new SRGB gamma extension? Turning this one on ruins everything unless it's accounted for beforehand.

AMD's OGL driver does not like tessellation this benchmark, nor does it like crossfire, but best case is 35% hit by using OGL, and over 60% with normal tessellation and crossfire enabled. The benchmarks linked are old, but I ran Heaven with 8xAA, Extreme Tessellation, and was getting less than half the fps of D3d11. I didn't even finish benchmarking because it was so bad.

I did included normal tessellation as well because tessellation is not a strength of 6970s, and we're not trying to benchmark the card's performance, but the driver.

1680x1050, 6970s in crossfire, Vysnc off:
DX Extreme Tess, 8xAA 37.4/14.2
OGL Extreme Tess, 8xAA Too embarrassing to post.
DX Normal Tess, 8xAA 48.9/20.2
OGL Normal Tess, 8xAA 19.0/9.2

Crossfire disabled:
DX No Tess, 8xAA 36.5/15.8
OGL No Tess, 8xAA 23.1/11.4

Well, this certainly explains the performance issues I was having in Rage (8k textures = slideshow; had to reduce to 2xAA).

The future of OpenGL in the gaming segment will depend entirely on how much control Microsoft is able to exert over gaming in general going forward. At the end of the day, it doesn't really matter which one is "better" from a purely technical perspective.

Deanjo, your minimum framerate numbers might not be correct (depending on your perspective); I found I had to start the benchmark, cancel it, then restart it because the initial loading dropped minimums way down. That would explain why my minimums are so much higher.

I also ran DX No Tess, 8xAA in crossfire and got 73.2/32.3; average framerate basically doubled and then some, but definitely within the margin of error I'd say. I usually expect 80% or more scaling. I suspect the drivers were crossfire-optimized for this particular benchmark.

Kind of off-topic, but thought I'd share since I did the benchies.

just brew it! wrote:

The future of OpenGL in the gaming segment will depend entirely on how much control Microsoft is able to exert over gaming in general going forward. At the end of the day, it doesn't really matter which one is "better" from a purely technical perspective.

Agreed.

Jason181 wrote:

Well, this certainly explains the performance issues I was having in Rage

Yet there were how many out of touch AMD complainers that placed the blame solely on Rage/Carmack, when the blame actually presided on AMD's bad OpenGL drivers, and AMD has always had bad OpenGL drivers. What made people think that Rage would be an exception, I dunno. Perhaps they were previously nvidia owners and didn't know, but now they know.

I definitely agree that AMD botched their driver release, but iD can't operate in a vacuum, and John Carmack implied as much in his Quakecon keynote. He took some responsibility for waiting on the driver situation too long, and it was definitely exacerbated by AMD releasing an old OpenGL driver. The wrong driver was definitely and inexcusably AMD's fault, although it appears that they might have been up against a deadline that they couldn't control.

Of course there were people that laid the blame entirely on iD, and that's just flat wrong. A software vendor it going to be accountable for the fact that the game didn't work for 40% of its potential customers. I agree that AMD's OGL drivers are atrocious (see above post for performance numbers), but a company ignores the environment into which the game is being released at its own peril.

just brew it! wrote:

The future of OpenGL in the gaming segment will depend entirely on how much control Microsoft is able to exert over gaming in general going forward. At the end of the day, it doesn't really matter which one is "better" from a purely technical perspective.

Agreed. The only myth here is that this OpenGL vs DX "debate" mattered at all on a technical level. The bottom line is if your development team is dead-set on OpenGL or DirectX, then your team will be dead-set on getting acceptable performance out of it. So the advantage to DirectX, during the DX6-9 days, was that nearly everyone was focused on improving performance with it. It's advantage was market momentum. OpenGL could have been 10x faster and easier to use, but it wouldn't have matter if nobody used it.

Though this momentum is changing today due to a more heterogeneous market (iOS, Android, Windows, Xbox, PS3, Wii vs. only Windows), DirectX is still riding its initial momentum on the desktop space. OpenGL is gaining momentum due to its cross-platform advantages. But the same truth remains: the technical details, speed, etc. does not matter. It's market position and momentum.

Most games are developed for consoles first. Since PS3 can't use Direct3D, why don't they already have a GL port of everything anyway?

Game_boy wrote:

Most games are developed for consoles first. Since PS3 can't use Direct3D, why don't they already have a GL port of everything anyway?

I could be mistaken, but I don't think PS3 supports OGL calls? PS3 does support OpenGL ES, evidently. PS3's hardware is a pretty far removed from x86 PCs though. Some of the programming is bare to metal to eek out all the performance possible.