The Big Benchmark

I'm not convinced this will help Xonotic. As the dev there states, AA is CPU bound, not GPU bound. Xonotic tends to be more GPU bound already so making this change may actually hit performance.

Another test:

Name: desktop-amd-7770, same setup as previous "desktop-amd", except GPU is now HD7770 and drivers are Catalyst 12.8

@edh: If you would have a look at the benchmark results then you´d quickly find out that the GPU isn´t that much of an interest. The CPU still seems to be the key part of your setup. Having an intel i7 and an old geforce 9800 (or similar) gives you extremely high FPS whereas using an AMD dual core (e.g. 2x2GHz) and a GTX560 isn´t that ideal.

@edh: If you would have a look at the benchmark results then you´d quickly find out that the GPU isn´t that much of an interest. The CPU still seems to be the key part of your setup. Having an intel i7 and an old geforce 9800 (or similar) gives you extremely high FPS whereas using an AMD dual core (e.g. 2x2GHz) and a GTX560 isn´t that ideal.

I think we have some different targets we are talking about here. If someone is getting 500fps on low details with the latest hardware, I'm not really bothered. This would be CPU limited on most systems. On the higher detail levels it is mostly GPU bound, otherwise you'd be getting the same 500fps all of the way up to Ultimate. When looking at older hardware you also see more of a GPU limit. Even comparing a Geforce FX5600 vs an FX5950 Ultra in a 2x1GHz Pentium 3 where each graphics card is way beyond the capability of the CPU, they each give drastically different performance which shows it to be GPU not CPU bound in such an environment. Similarly with a 6600GT in an Athlon XP 2400+ which is massively over powered on the GPU vs CPU there are very different framerates across the detail levels. This would not be the case if it was CPU bound. Making the game playable on older hardware is more of interest to me than someone getting 500fps on the newest hardware instead of 480fps. In addition there is a level of disparity I have reported on before between Windows and Linux which may show a level of GPU vs CPU bias:
http://forums.xonotic.org/showthread.php...2#pid35982

I need proof. No offence, but I think it's only fair to say what advantages there will be once people test it (gpu animations).

No proof but at least evidence:
Old system:
AMD Phenom II x3 @ 2.9ghz
ATI HD5750

Tried overlocking my gpu, gained 0% fps (average around 150). Then upgraded my CPU to AMD Phenom II x4 @ 3.4ghz -> 25% fps increase. On high quality configs the GPU might be the bottleneck, but on normal-low configs the CPU certainly has a huge impact. Maybe some systems are able to run Xonotic at 500fps, but I know quite a lot people who have problems on getting stable 125fps on a low-ish config.

Okay, maybe you're right, but there are still plenty of other things that need optimising in Xonotic that would affect the FPS a lot.

And a much more important issue is the fact that my GPU doesn't run on 2 cores while in Xonotic. This needs to be addressed.

http://forums.xonotic.org/showthread.php?tid=2861

Username: Halbyrd
System: WisdomLikeSilence
CPU: Intel Core i5-2300
GHz: 2.3
Cores: 4
RAM: 8gb
Vendor: ATI Technologies Inc.
Card: AMD Radeon HD 6900 Series
Driver: 4.2.11762 Compatibility Profile Context
OS: Windows 7
Arch: x64
Notes: Radeon HD 6970

OMG - MED: 10510 frames 43.2610000 seconds 242.9439911 fps, one-second fps min/avg/max: 154 249 462 (336 seconds)
Low - MED: 10510 frames 46.1140000 seconds 227.9134319 fps, one-second fps min/avg/max: 150 234 431 (336 seconds)
Med - MED: 10510 frames 49.8620000 seconds 210.7817577 fps, one-second fps min/avg/max: 138 219 431 (336 seconds)
Normal - MED: 10510 frames 52.2710000 seconds 201.0675135 fps, one-second fps min/avg/max: 127 210 403 (336 seconds)
High - MED: 10510 frames 58.4560000 seconds 179.7933488 fps, one-second fps min/avg/max: 112 188 373 (336 seconds)
Ultra - MED: 10510 frames 78.8500000 seconds 133.2910590 fps, one-second fps min/avg/max: 28 152 323 (336 seconds)
Ultimate - MED: 10510 frames 85.4700000 seconds 122.9671230 fps, one-second fps min/avg/max: 28 140 304 (336 seconds)

OMG - MED: 10510 frames 43.2610000 seconds 242.9439911 fps, one-second fps min/avg/max: 154 249 462 (336 seconds)
Low - MED: 10510 frames 46.1140000 seconds 227.9134319 fps, one-second fps min/avg/max: 150 234 431 (336 seconds)
Med - MED: 10510 frames 49.8620000 seconds 210.7817577 fps, one-second fps min/avg/max: 138 219 431 (336 seconds)
Normal - MED: 10510 frames 52.2710000 seconds 201.0675135 fps, one-second fps min/avg/max: 127 210 403 (336 seconds)
High - MED: 10510 frames 58.4560000 seconds 179.7933488 fps, one-second fps min/avg/max: 112 188 373 (336 seconds)
Ultra - MED: 10510 frames 78.8500000 seconds 133.2910590 fps, one-second fps min/avg/max: 28 152 323 (336 seconds)
Ultimate - MED: 10510 frames 85.4700000 seconds 122.9671230 fps, one-second fps min/avg/max: 28 140 304 (336 seconds)

Looks a little slower than might be possible with this hardware. You might find this helpful:
http://www.howtogeek.com/howto/windows-v...ideo-game/

More results, system names: desktop-sb (for Celeron machine), desktop-amd-7770 (for Athlon machine)

Desktop-sb: 180 153 137 133 125 110 84
Desktop-amd-7770: 184 167 151 142 121 80 73

Hi there, just my first ( but not last post ! )

No AMD APUs here? Not to be underestimated, these are strong performers... at least the hig-end ones ( A8-5600K , A10-5800K )... so here are my results:

MED: 10510 frames 44.0158889 seconds 238.7774110 fps, one-second fps min/avg/max: 150 245 444 (336 seconds)
MED: 10510 frames 46.1597502 seconds 227.6875405 fps, one-second fps min/avg/max: 150 235 447 (336 seconds)
MED: 10510 frames 50.2499371 seconds 209.1544908 fps, one-second fps min/avg/max: 123 220 446 (336 seconds)
MED: 10510 frames 52.8321791 seconds 198.9317909 fps, one-second fps min/avg/max: 111 210 431 (336 seconds)
MED: 10510 frames 74.0338409 seconds 141.9621064 fps, one-second fps min/avg/max: 74 149 227 (336 seconds)
MED: 10510 frames 109.5647080 seconds 95.9250492 fps, one-second fps min/avg/max: 22 109 191 (336 seconds)
MED: 10510 frames 167.9587250 seconds 62.5748975 fps, one-second fps min/avg/max: 21 68 123 (336 seconds)

My Setup:

Platform:
=======
Processor : AMD APU A8-5600K ( 3.6Ghz , 4 x "Piledriver" integer cores, 2 x 256-bit FMAC units, 4 MB L2 Cache )
GPU : Radeon 7650D integrated on APU ( 256 shader units )
RAM: 8GByte HyperX 1600 ( at 667 Mhz ) memory, dual-channel.

Operating system info:
=================
OS : Slackware Linux 14
Arch : x86_64
Cores : 4
Memory: 7608580 ( 512 MB reserved for GPU )

GL information:
===========
Vendor: GL_VENDOR: ATI Technologies Inc.
Card: GL_RENDERER: AMD Radeon HD 7560D
Driver: GL_VERSION: 4.2.11931 Compatibility Profile Context

Extra info:
========
Re-compiled binaries with GCC 4.7 with flags: -march=bdver2 -msse4.2

Conclusions:
==========
Pretty nifty results , playing at 1280x768 on this is a blast! and the Keybench benchmark never drops of 60 FPS on Ultra settings ( Minus Decals, Minus Reflections, Particle Density = 0.7, but All shadowscheckboxes ON )

Considering harsh reviews coming from internet...AMD APUs ( higher end ones ) are a viable platform for gaming, unbeateable on price

ferndan, could you please include the log files too? And, since you recompiled the engine, tell us where you got the sources (git? if so, what revision?).

Re-compiled binaries with GCC 4.7 with flags: -march=bdver2 -msse4.2

Hey ferndan, did you check the influence of these flags on your game performance? (Benchmark with a recompiled version, but without tuning for a special architecture.) I was always wondering how much impact these flags have in "everyday" applications

Yeah, I've forgot the logs, here there are.

About recompiling... the engine source code is already on the Xonotic folder , so:

1. Go to the sources/darkplaces folder

2. Locate the "makefile" file.

3. Edit it and locate a line with the flag CPUOPTIMIZATIONS, it looks like:

CPUOPTIMIZATIONS= -mfpmath=both -fno-math-errno -finite-math-only -fno-rounding-math -fno-signaling -fno-trapping-math

4. Add the flags you need, save and next execute :

make

5. It will generate THREE binaries:

nexuiz-dedicated
nexuiz-glx
nexuiz-sdl

6. Copy them on the Xonotic root folder... but rename them to the corresponding "xonotic" naming , depending on architecture ( backup the original binaries first ):

xonotic-linux32-dedicated or xonotic-linux64-dedicated
xonotic-linux32-glx or xonotic-linu64-glx
xonotic-linux32-sdl or xonotic-linux64-sdl

Hope that helps... now I will run with the original stock binaries, I'll let you know soon

I've tested on a few other systems from the last decade or so.

1. Sempron 2800+ @ stock 2.0GHz
1Gb RAM
ATI Radeon 9550
GL_VERSION: 2.1.8545 Release
Windows XP Pro 32-bit

Results: 67/63/52/24/19/-/-

The Windows driver appears to have some graphical errors with warpzones which make things a bit wierd. It also doesn't support offset mapping so ultra nd ultimate were not tested. Can't recommend this setup even if it is a playable speed because of the graphical issues.

2. Athlon XP 2000+ Palomino @ stock 1.666GHz
768Mb RAM
Geforce FX5900 128Mb BIOS modded to FX5950 Ultra
GL_VERSION: 2.0.3
Windows XP Pro 32-bit

Results: 50/48/41/21/15/-/-

This card does not support offset mapping so again ultra and ultimate were not tested.

3. Athlon XP 2000+ Palomino @ stock 1.666GHz
768Mb RAM
Geforce FX5900 128Mb BIOS modded to FX5950 Ultra
NVIDIA driver
GL_VERSION: 2.1.2 NVIDIA 173.14.36
Arch Linux x86

Results: 52/48/41/20/15/-/-

As before but under Linux. Performance is about the same, still no offset mapping.

4. Athlon XP 2000+ Palomino @ stock 1.666GHz
768Mb RAM
Geforce FX5900 128Mb BIOS modded to FX5950 Ultra
Nouveau, Gallium 0.4
GL_VERSION: 1.5 Mesa 9.0.1
Arch Linux x86

Results: 49 on OMG, then graphical errors after this.

Not bad speed compared to the propreitary driver on OMG but on Low things all fell apart. Can't recommend Nouveau for Geforce FX still.

5. Athlon XP 2000+ Palomino @ stock 1.666GHz
768Mb RAM
Geforce4 MX440
GL_VERSION: 1.5.7
Windows XP Pro 32-bit

Results: 55/35/25/-/-/-/-

Surprising performance perhaps as it actually appears faster than a Geforce FX5950 Ultra on OMG! However this is skewed by it not supporting OpenGL 2.0 so what it is rendering is not strictly the same. Can't recommend such old hardware.

6. Athlon XP 2400+ Thoroughbred @ 2.025GHz
1Gb RAM
Radeon 9550 (250/400), BIOS modded to 9600 (325/400) then overclocked to 405/472MHz, it would give 536/530 but I backed it off for longevity!
GL_RENDERER: Gallium 0.4 on ATI RV350
GL_VERSION: 2.1 Mesa 9.0.1
Arch Linux x86

Results: 56/45/38/21/16/-/-

The free driver works rather well. No graphical issues seem with the ATI Windows driver but still offset mapping is not supported.

Uploaded all of these results in single ZIP file.

System Name: Gateway LT4004u Netbook
CPU: Intel Atom N2600 @ 1.6 GHz
2 cores, 4 threads
2 GB of RAM, running at DDR3/800 (underclocked by CPU)
Windows 7 Starter 32-bit
Intel GMA 3600
8.14.8.1065 driver (OpenGL 3.0 support)
Benchmark ran at 1024x768 (I have registry hack enabled for downscaling display)

I ran the benchmark before the registry hack at 640x480 and was able to complete all 4 runs on high settings at 14 fps (Ultra didn't do so well). At 1024x768, I got through 2 runs on high before Windows restarted.

Hey QorL here;-)

Many of u here are posting results without resolution settings!
-OMG, 1920 X 1080
10510 frames 60.4010000 seconds 174.0037417 fps, one-second fps min/avg/max: 111 185 341 (336 seconds)
-Normal, 1920 X 1080
10510 frames 62.7810000 seconds 167.4073366 fps, one-second fps min/avg/max: 87 177 360 (336 seconds)
-Ultimate 1920 X 1080
10510 frames 150.3070000 seconds 69.9235565 fps, one-second fps min/avg/max: 15 79 160 (336 seconds)

zero (0) Anysotropy,

System:
-Gigabyte GA-890GPA-UD3H
-Gigabyte GV-R685OC-1GD
-AMD Phenom II x4 965 3.4 ghz,
-Corsair 8gb 1600 mhz (unganged) 1333 mhz
-Intel 330 ssd 180 gb
-gl_version:6.14.10.12002
-Win 7 ultimate, 6.1 - 7601 sp1

Many of u here are posting results without resolution settings!

I think you're a little confused here on the Big Benchmark. What you have run is the big keybench with your own settings.

The Big Benchmark runs the big keybench with automated multiple runs on multiple settings, some of which are specific to The Big Benchmark and which won't apply if you run the big keybench manually. All of these tests are done at 1024x768 as that is a resolution which almost any modern display can support. Therefore no one is listing the resolution unless they are running something different for technical reasons, like on a netbook with a lower resolution.

Your results can not be counted as they are incomparable.

Please reread the first post in this thread, then rerun The Big Benchmark as described and post your results.

Did it right now,
http://ompldr.org/iaGx5Ng

grep '^MED' the-big-benchmark.log
MED: 10510 frames 43.9460000 seconds 239.1571474 fps, one-second fps min/avg/max: 140 247 484 (336 seconds)
MED: 10510 frames 47.4300000 seconds 221.5897112 fps, one-second fps min/avg/max: 135 231 477 (336 seconds)
MED: 10510 frames 52.2270000 seconds 201.2369081 fps, one-second fps min/avg/max: 121 213 463 (336 seconds)
MED: 10510 frames 55.5210000 seconds 189.2977432 fps, one-second fps min/avg/max: 108 201 443 (336 seconds)
MED: 10510 frames 61.3340000 seconds 171.3568331 fps, one-second fps min/avg/max: 100 182 403 (336 seconds)
MED: 10510 frames 96.0460000 seconds 109.4267330 fps, one-second fps min/avg/max: 21 134 356 (336 seconds)
MED: 10510 frames 104.3120000 seconds 100.7554260 fps, one-second fps min/avg/max: 18 121 263 (336 seconds)



239 | 221 | 201 | 189 | 171 | 109 | 100

Yet another benchmark on unplayably old hardware:

Pentium 3 550MHz
320Mb RAM
GL_VENDOR: NVIDIA Corporation
GL_RENDERER: GeForce4 MX 440/AGP/SSE
GL_VERSION: 1.5.8 NVIDIA 96.43.23
Arch Linux x86-64

16/9/6/-/-/-/-

So Geforce4 MX440 is better than a Radeon 9000 but still not playable. Really can't recommend hardware of this age as it does not support OpenGL 2.0 and hence maps that use warpzones are somewhat difficult.

Swift shader,
Will this software work on xonotic?
http://www.youtube.com/watch?v=C-DtXjjkH9c

Swift shader,
Will this software work on xonotic?

Why not try it? Better that way rather than getting someone else to view a Youtube video then hazard a guess. I don't tend to click on links to videos without any explanation - a one paragraph description uses much less bandwidth.

All it is is a software renderer. These have existed for a long time.

It is only better if there is no dedicated GPU and you only have really terrible onboard graphics but a massively fast CPU.

Their claim of 620 on 3DMark06 with an i7 really shows that you're not going to get much performance. From memory I used to get a few thousand in 3DMark06 with a 6600GT which is a reasonably playable card for Xonotic on normal details. I would therefore guess that SwiftShader will not get anything playable with pretty much any system.

This is worse performance wise that the integrated software renderer in Xonotic. It is already of some use for those without a dedicated GPU but it does require SSE2. You can try this if you want from the Xonotic console:

Code:

vid_soft 1
vid_restart

Sorry to break the news but there are no magic beans to improve performance like this. Their '100x faster' boast is only from something very, very terrible.

I'm just some regular dude that plays xonotic;-)
My belief in all this graphic stuff that works good or bad is here just because of money, so I think that the answer lies in software.
I remember times when i was plying quake 1 in 1999 and was working fine on 8mb graphic, today all of this stuff is abolished by some
shader models. I'm not a professional in programming or whatever but i could see a scam right away.
Thx edh;-)
I'm QorL not a normal dude;-)

@up

Not even close to being a scam, you are VASTLY underestimating, what GPUs now need to deal with. Sheer amount of data to process has gone up by several orders from magnitude (and if ray tracing becomes a practical thing, it will grow even further). You get better image in quality in return (more vertices means more detailed models, bigger textures too, AA/AF to deal with artifacts of rendering) Also, Quake 1 in 1999 wasn't exactly news, not even Quake 2, Q3A was released by the end of that year. And good luck playing Q3A on the card you're talking about.

Shader models or new OGL versions are not abolishing anything, in fact they are adding ways to achieve new effects not available at the time before. And as noted before, you need more computing power to deal with it.

My belief in all this graphic stuff that works good or bad is here just because of money, so I think that the answer lies in software.

Xonotic uses OpenGL which runs in hardware on the graphics card. No magic software will improve that situation if you have a decent graphics card and driver that provides hardware acceleration.

To demonstrate, here are two comparisons:

1. Take a Core i7 920 with onboard graphics that provide no hardware acceleration (as they did in their testimonial) and install this software and you will experience massive improvements as they quote for 3DMark06. However, it still only gets 620 which is on a par with a Geforce FX perhaps.

2. Take the same system with the highest end gaming graphics card you can find and try installing this software. How much difference will it make? None at all. This is because the graphics card already provides hardware acceleration which will be much, much faster than any software path unless you really do have a supercomputer.

Software emulation can not match hardware acceleration.

I remember times when i was plying quake 1 in 1999 and was working fine on 8mb graphic

The graphics card is irrelevant to the original Quake so long as you have 2Mb video RAM and VESA 2.0 compliance. Both DOS Quake and WinQuake run entirely in software. The only difference that might be seen are when running a VESA 1.0 card as opposed to VESA 2.0 or running on a card with less than 2Mb RAM. VESA 2.0 cards provide faster 2D rendering (all the card does in Quake, the 3D is done in the engine) and you need greater than 1Mb to render the top resolution: 1280x1024.

I've been there before. Swapping a Geforce2 MX into an old Pentium and getting no performance improvement at all versus the Matrox that was in there before. Why? WinQuake only used software rendering. GLQuake and it's derivatives will make use of hardware so that it where the improvement was to be had and it looked better. Therefore rather than running 512x384 at 21fps I was able to run 1024x768 at 32fps.

i could see a scam right away.

It's not that it's a scam. It's a wonderful piece of software if you don't have a graphics card/driver which provides hardware acceleration. This may be an acceptable solution for light gaming uses like the Sims, compositing window managers and similar but for a graphically complex game like Xonotic where a dedicated graphics card is an accepted recommendation, this will achieve nothing, especially when we already have a software rendering path.

A bottom of the range graphics card is not a major expenditure and will outperform this substantially.

Besides, the software costs money and is closed source, hence even if it was to improve performance (which it won't for 99.99999% of people who could actually run Xonotic anyway), it can not be integrated into Xonotic.