GZDoom benchmark info needed

[_mental_]

My mistake, I loaded GZDoom with doom2.wad
Results for P:AR from the same GeForce 650M on OS X 10.9:

---

Without lights:
Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=1.291, Split = 0.000, Setup=0.549, Clip=0.765
F: Render=0.710, Setup=0.033
S: Render=0.738, Setup=0.188
All=10.442, Render=3.186, Setup=2.039, BSP = 0.499, Portal=0.000, Finish=5.170
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 7 upper, 0 lower, 0.005 ms
92 fps

With lights:
Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 13328 vertices)
Flats: 431 (9002 primitives, 43983 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=3.598, Split = 0.000, Setup=0.841, Clip=1.067
F: Render=3.474, Setup=0.085
S: Render=0.858, Setup=0.270
All=19.080, Render=8.753, Setup=2.987, BSP = 0.717, Portal=0.000, Finish=7.280
DLight - Walls: 1972 processed, 722 rendered - Flats: 8196 processed, 1446 rendered
Missing textures: 7 upper, 0 lower, 0.007 ms
58 fps

---

Without lights:
Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=3.432, Split = 0.000, Setup=0.555, Clip=0.840
F: Render=6.585, Setup=0.032
S: Render=1.127, Setup=0.188
All=15.443, Render=11.560, Setup=2.158, BSP = 0.538, Portal=0.000, Drawcalls=9.712, Finish=1.681
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 7 upper, 0 lower, 0.005 ms
62 fps

With lights:
Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=4.644, Split = 0.000, Setup=0.572, Clip=0.831
F: Render=8.947, Setup=0.033
S: Render=1.278, Setup=0.216
All=18.611, Render=15.299, Setup=2.214, BSP = 0.555, Portal=0.000, Drawcalls=11.663, Finish=1.044
DLight - Walls: 1972 processed, 748 rendered - Flats: 8196 processed, 1450 rendered
Missing textures: 7 upper, 0 lower, 0.006 ms
51 fps

[Graf Zahl]

Here's the link for the new test build:

http://www6.zippyshare.com/v/41974177/file.html

@mental: Thanks. It's really too bad that Apple skipped OpenGL 3.0. All my tests show that without persistently mapped buffers, immediate mode is still the fastest way to render stuff that consists of large amounts of low-vertex-count draw calls.
The new test build offers different methods:

- immediate mode (obviously not on MacOSX due to lack of compatibility profile)
- uniform arrays (that's what the last test build had)
- buffer uploads for each draw call
- map/unmap buffer for each draw call

After I have the results from the next round of tests I hopefully get some ideas how to optimize this for better performance on GL 3.x hardware.

[Enjay]

For info, the download doesn't include the GLEW32.DLL file.

Here are my test results. I noticed that you left your ini in the zip so, other than changing screen mode/resolution, I used your settings.

FrozenT

---

Code: Select all

Map MAP01: "entryway",
x = 4676.2854, y = 5956.6655, z = 369.0000, angle = 90.3790, pitch = -1.3348
Walls: 14573 (0 splits, 2007 t-splits, 110570 vertices)
Flats: 537 (24512 primitives, 107913 vertices)
Sprites: 708, Decals=0, Portals: 1
W: Render=6.018, Split = 0.000, Setup=3.944, Clip=5.919
F: Render=1.062, Setup=0.124
S: Render=0.252, Setup=1.301
All=31.957, Render=11.282, Setup=16.661, BSP = 5.371, Portal=0.078, Drawcalls=0.603, Finish=3.962
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
31 fps

Map MAP01: "entryway",
x = 4676.2854, y = 5956.6655, z = 369.0000, angle = 90.3790, pitch = -1.3348
Walls: 14573 (0 splits, 2007 t-splits, 110570 vertices)
Flats: 537 (24512 primitives, 107913 vertices)
Sprites: 708, Decals=0, Portals: 1
W: Render=6.304, Split = 0.000, Setup=3.953, Clip=6.028
F: Render=1.090, Setup=0.125
S: Render=0.258, Setup=1.358
All=31.299, Render=11.596, Setup=16.912, BSP = 5.447, Portal=0.081, Drawcalls=0.602, Finish=2.738
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
31 fps

Map MAP01: "entryway",
x = 4676.2854, y = 5956.6655, z = 369.0000, angle = 90.3790, pitch = -1.3348
Walls: 14573 (0 splits, 2007 t-splits, 110570 vertices)
Flats: 537 (24512 primitives, 107913 vertices)
Sprites: 708, Decals=0, Portals: 1
W: Render=6.167, Split = 0.000, Setup=4.007, Clip=5.888
F: Render=1.161, Setup=0.125
S: Render=0.253, Setup=1.300
All=32.650, Render=11.502, Setup=16.760, BSP = 5.438, Portal=0.077, Drawcalls=0.681, Finish=4.311
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
31 fps

PAR

---

Code: Select all

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 9144 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=0.641, Split = 0.000, Setup=0.645, Clip=0.619
F: Render=0.264, Setup=0.040
S: Render=0.088, Setup=0.131
All=16.390, Render=1.659, Setup=1.912, BSP = 0.472, Portal=0.000, Drawcalls=0.112, Finish=12.791
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 7 upper, 0 lower, 0.004 ms
59 fps

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 9144 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=0.639, Split = 0.000, Setup=0.641, Clip=0.623
F: Render=0.265, Setup=0.040
S: Render=0.087, Setup=0.131
All=16.360, Render=1.657, Setup=1.910, BSP = 0.471, Portal=0.000, Drawcalls=0.114, Finish=12.765
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 7 upper, 0 lower, 0.004 ms
60 fps

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 9144 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=0.644, Split = 0.000, Setup=0.654, Clip=0.628
F: Render=0.262, Setup=0.040
S: Render=0.086, Setup=0.130
All=16.354, Render=1.658, Setup=1.933, BSP = 0.476, Portal=0.000, Drawcalls=0.112, Finish=12.736
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 7 upper, 0 lower, 0.004 ms
60 fps

Computer specs as before (with the GTX 760).

[Graf Zahl]

Well, didn't I say that I don't need GL 4.x results? You essentially tested the 'good' render path 3 times. What I need is the tests for the 'bad' render path for older hardware.

BTW, you left vsync on making the results mostly useless. The left-in .ini was an accident. Please ignore.

[Enjay]

I did wonder if vsync was left on or not (especially when I saw the PAR results) but didn't actually check. Given that you don't need my results though, I'll not bother re-running the tests (although on a quick check I'm getting around 245 fps with the PAR save).

[Graf Zahl]

Yes, that sounds more like it. Your results don't help me because they very closely mirror my own setup. What I need is info from different hardware. Currently most important are:

- first and second generation GL 3.x cards from NVidia (Geforce 8xxx, 9xxx or GTX 2xx)
- first and second generation GL 3.x cards from AMD (Radeon 4xxx and 5xxx, if I'm not mistaken)
- Intel GMA 3000.

Before going on I need to know how the different render methods perform on these systems.

[VGA]

Here it is on my c2d e8400 @ 3.3ghz, gtx260, 4gb RAM
Nvidia driver version is 337.88
I used default settings at my native resolution of 1680x1050

Here are the bench results for gl_rendermethod at 0,1,2,3:
Frozen Time save:

---

Code: Select all

Map MAP01: "entryway",
x = 4676.2854, y = 5956.6655, z = 369.0000, angle = 90.3790, pitch = -1.3348
Walls: 14573 (0 splits, 2007 t-splits, 36216 vertices)
Flats: 537 (24512 primitives, 107913 vertices)
Sprites: 708, Decals=0, Portals: 1
W: Render=4.574, Split = 0.000, Setup=3.285, Clip=6.665
F: Render=5.234, Setup=0.115
S: Render=0.323, Setup=0.986
All=30.892, Render=12.639, Setup=16.552, BSP = 5.499, Portal=0.070, Drawcalls=3.221, Finish=1.635
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
31 fps

Map MAP01: "entryway",
x = 4676.2854, y = 5956.6655, z = 369.0000, angle = 90.3790, pitch = -1.3348
Walls: 14573 (0 splits, 2007 t-splits, 36216 vertices)
Flats: 537 (24512 primitives, 107913 vertices)
Sprites: 708, Decals=0, Portals: 1
W: Render=4.809, Split = 0.000, Setup=3.261, Clip=6.648
F: Render=5.280, Setup=0.111
S: Render=0.311, Setup=1.019
All=42.334, Render=12.977, Setup=16.518, BSP = 5.477, Portal=0.048, Drawcalls=2.052, Finish=12.770
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
23 fps

Map MAP01: "entryway",
x = 4676.2854, y = 5956.6655, z = 369.0000, angle = 90.3790, pitch = -1.3348
Walls: 14573 (0 splits, 2007 t-splits, 36216 vertices)
Flats: 537 (24512 primitives, 107913 vertices)
Sprites: 708, Decals=0, Portals: 1
W: Render=5.407, Split = 0.000, Setup=3.305, Clip=6.680
F: Render=8.077, Setup=0.112
S: Render=0.357, Setup=0.928
All=313.256, Render=16.465, Setup=16.505, BSP = 5.478, Portal=0.070, Drawcalls=6.428, Finish=280.216
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
4 fps

Map MAP01: "entryway",
x = 4676.2854, y = 5956.6655, z = 369.0000, angle = 90.3790, pitch = -1.3348
Walls: 14573 (0 splits, 2007 t-splits, 36216 vertices)
Flats: 537 (24512 primitives, 107913 vertices)
Sprites: 708, Decals=0, Portals: 1
W: Render=61.850, Split = 0.000, Setup=3.277, Clip=6.695
F: Render=43.629, Setup=0.112
S: Render=10.957, Setup=0.931
All=316.774, Render=122.296, Setup=16.489, BSP = 5.473, Portal=3.246, Drawcalls=110.546, Finish=177.854
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
3 fps

Voxel map save:

---

Code: Select all

Map map01: "entryway",
x = -736.0000, y = 352.0000, z = 41.0000, angle = 315.0000, pitch = 0.0000
Walls: 25 (0 splits, 0 t-splits, 72 vertices)
Flats: 24 (34 primitives, 148 vertices)
Sprites: 292, Decals=0, Portals: 0
W: Render=0.007, Split = 0.000, Setup=0.006, Clip=0.005
F: Render=0.016, Setup=0.003
S: Render=0.247, Setup=0.075
All=6.893, Render=0.307, Setup=0.093, BSP = 0.004, Portal=0.000, Drawcalls=0.026, Finish=6.484
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
142 fps

Map map01: "entryway",
x = -736.0000, y = 352.0000, z = 41.0000, angle = 315.0000, pitch = 0.0000
Walls: 25 (0 splits, 0 t-splits, 72 vertices)
Flats: 24 (34 primitives, 148 vertices)
Sprites: 292, Decals=0, Portals: 0
W: Render=0.006, Split = 0.000, Setup=0.006, Clip=0.005
F: Render=0.034, Setup=0.002
S: Render=0.217, Setup=0.049
All=6.826, Render=0.297, Setup=0.066, BSP = 0.004, Portal=0.000, Drawcalls=0.035, Finish=6.456
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
141 fps

Map map01: "entryway",
x = -736.0000, y = 352.0000, z = 41.0000, angle = 315.0000, pitch = 0.0000
Walls: 25 (0 splits, 0 t-splits, 72 vertices)
Flats: 24 (34 primitives, 148 vertices)
Sprites: 292, Decals=0, Portals: 0
W: Render=0.016, Split = 0.000, Setup=0.015, Clip=0.015
F: Render=0.023, Setup=0.005
S: Render=0.271, Setup=0.127
All=11.835, Render=0.354, Setup=0.170, BSP = 0.006, Portal=0.000, Drawcalls=0.079, Finish=11.276
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
85 fps

Map map01: "entryway",
x = -735.9209, y = 351.9110, z = 41.0000, angle = 309.9023, pitch = 0.0000
Walls: 26 (0 splits, 0 t-splits, 76 vertices)
Flats: 24 (34 primitives, 148 vertices)
Sprites: 292, Decals=0, Portals: 0
W: Render=0.052, Split = 0.000, Setup=0.006, Clip=0.005
F: Render=0.077, Setup=0.002
S: Render=1.241, Setup=0.070
All=10.349, Render=1.415, Setup=0.088, BSP = 0.004, Portal=0.000, Drawcalls=0.908, Finish=8.789
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 0 upper, 0 lower, 0.000 ms
93 fps

On the voxel map everything got distorted at values 2 and 3, here are some screenshots:

---

---

I don't see a savegame for P:AR so I didn't test that one.

[Graf Zahl]

What IWAD did you load it with? doom.wad or doomu.wad. The savegame requires doom.wad, if you use doomu.wad it won't work.

As for the voxel map visual problems, I'll wait with an investigation until I get results from other systems. By now it seems to be clear that both methods are too slow, at least on NVidia hardware.

[VGA]

Oops i was trying with doom2.wad. My Ultimate Doom is doom.wad, but no probs.

First with loading only PAR.wad:

---

Code: Select all

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=0.658, Split = 0.000, Setup=0.569, Clip=0.735
F: Render=0.977, Setup=0.039
S: Render=0.118, Setup=0.144
All=5.706, Render=2.272, Setup=1.992, BSP = 0.500, Portal=0.000, Drawcalls=0.718, Finish=1.401
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 7 upper, 0 lower, 0.004 ms
167 fps

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=0.617, Split = 0.000, Setup=0.604, Clip=0.845
F: Render=0.811, Setup=0.040
S: Render=0.109, Setup=0.124
All=8.698, Render=2.149, Setup=2.109, BSP = 0.491, Portal=0.000, Drawcalls=0.416, Finish=4.398
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 7 upper, 0 lower, 0.005 ms
110 fps

Map e1m6: "Central Processing",
x = 607.2454, y = -1819.9005, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2894 (0 splits, 70 t-splits, 7548 vertices)
Flats: 432 (5126 primitives, 24889 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=0.834, Split = 0.000, Setup=0.623, Clip=0.894
F: Render=1.438, Setup=0.041
S: Render=0.116, Setup=0.128
All=106.958, Render=2.914, Setup=2.191, BSP = 0.502, Portal=0.000, Drawcalls=1.319, Finish=101.808
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 6 upper, 0 lower, 0.004 ms
10 fps

Map e1m6: "Central Processing",
x = 618.0023, y = -1850.7850, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=4.683, Split = 0.000, Setup=0.601, Clip=0.928
F: Render=7.417, Setup=0.040
S: Render=1.374, Setup=0.132
All=106.524, Render=14.024, Setup=2.209, BSP = 0.504, Portal=0.000, Drawcalls=12.160, Finish=90.184
DLight - Walls: 0 processed, 0 rendered - Flats: 0 processed, 0 rendered
Missing textures: 7 upper, 0 lower, 0.005 ms
10 fps

Then with lights.pk3 and brightmaps.pk3 and PAR.wad:

---

Code: Select all

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=1.333, Split = 0.000, Setup=0.593, Clip=0.835
F: Render=1.752, Setup=0.040
S: Render=0.181, Setup=0.149
All=6.831, Render=3.791, Setup=2.099, BSP = 0.477, Portal=0.000, Drawcalls=0.758, Finish=0.885
DLight - Walls: 1973 processed, 703 rendered - Flats: 8196 processed, 1447 rendered
Missing textures: 7 upper, 0 lower, 0.005 ms
139 fps

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=1.401, Split = 0.000, Setup=0.650, Clip=0.905
F: Render=1.641, Setup=0.042
S: Render=0.170, Setup=0.161
All=9.689, Render=3.748, Setup=2.255, BSP = 0.492, Portal=0.000, Drawcalls=0.442, Finish=3.634
DLight - Walls: 1973 processed, 728 rendered - Flats: 8196 processed, 1449 rendered
Missing textures: 7 upper, 0 lower, 0.005 ms
99 fps

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=1.486, Split = 0.000, Setup=0.890, Clip=1.332
F: Render=2.392, Setup=0.090
S: Render=0.189, Setup=0.265
All=109.387, Render=4.632, Setup=3.254, BSP = 0.671, Portal=0.000, Drawcalls=1.413, Finish=101.392
DLight - Walls: 1973 processed, 717 rendered - Flats: 8196 processed, 1453 rendered
Missing textures: 7 upper, 0 lower, 0.006 ms
9 fps

Map e1m6: "Central Processing",
x = 618.0078, y = -1850.7798, z = 169.0000, angle = 109.2316, pitch = 0.5988
Walls: 2921 (0 splits, 70 t-splits, 7612 vertices)
Flats: 431 (5163 primitives, 25066 vertices)
Sprites: 296, Decals=0, Portals: 0
W: Render=4.399, Split = 0.000, Setup=0.620, Clip=0.930
F: Render=10.382, Setup=0.041
S: Render=1.930, Setup=0.133
All=107.289, Render=17.276, Setup=2.208, BSP = 0.479, Portal=0.000, Drawcalls=13.218, Finish=87.696
DLight - Walls: 1973 processed, 737 rendered - Flats: 8196 processed, 1445 rendered
Missing textures: 7 upper, 0 lower, 0.005 ms
9 fps

[Blue Shadow]

Going by VGA's results, the two new rendering methods seem to be ridiculously slow when compared with the other two. The "immediate" method is significantly better.

Interestingly, the two new methods are performing relatively better in the voxel test than in FrozenT and PAR.

[VGA]

Maybe because the rendering ... broke!

[GFD]

Here's a whole bunch more benchmarks from my GeForce 9200. Hopefully they're useful. Just used default settings again (except not in fullscreen).

Bench's startup log
FrozenT: latest dev, rendermethod O, 1, 2, 3
PAR: latest dev, rendermethod O, 1, 2, 3
PAR (lights): latest dev, rendermethod O, 1, 2, 3
zdoom_vox: latest dev, rendermethod O, 1, 2, 3

For some reason, putting URL tags around a 0 changes the 0 into the URL?? Whatever.

[Graf Zahl]

Going by VGA's results, the two new rendering methods seem to be ridiculously slow when compared with the other two. The "immediate" method is significantly better.

Interestingly, the two new methods are performing relatively better in the voxel test than in FrozenT and PAR.

Those new methods are what is supposed to be the way things should be done in GL 3.x, i.e. using buffers to store the vertices.

This is truly insane. GZDoom has such a low vertex to drawcall ratio and at the same time a relatively high amount of draw calls plus an unstable set of vertices that it's hopelessly inefficient to transition 1:1. I already knew that the current implementation is useless at least on NVidia, but before moving on I really need some data for AMD and Intel to decide how to proceed.
Does nobody here use AMD and Intel? I already got all the numbers I need for NVidia but none for the other two.

The reason it has such a lower impact on the voxel map is that voxels, as they are models, can be stored in a static vertex buffer (let's disregard for now that there's a bug somewhere in there) which do not break down with buffer upload delays. But even there with the low amount of map geometry it clearly shows its problems.

@gamefreakdude:

I have to admit that these numbers really do bother me. It means that even under best circumstances the new version is slower on such old hardware than the old one. Arrrgghhh.

[_mental_]

I just wanted to do a quick check, but the latest benchmark fails on fragment shader compilation.

---

OS: Windows 7 (NT 6.1) Build 7601 Service Pack 1
...
CPU Speed: 1397 MHz
CPU Vendor ID: GenuineIntel
Name: Intel(R) Core(TM)2 Duo CPU U9400 @ 1.40GHz
Family 6, Model 23, Stepping 10
Features: MMX SSE SSE2 SSE3 SSSE3 SSE4.1
...
GL_VENDOR: ATI Technologies Inc.
GL_RENDERER: ATI Mobility Radeon HD 4330
GL_VERSION: 3.1.8918
GL_SHADING_LANGUAGE_VERSION: 1.40
GL_EXTENSIONS: GL_AMDX_vertex_shader_tessellator GL_AMD_draw_buffers_blend GL_AMD_performance_monitor GL_AMD_texture_texture4 GL_AMD_vertex_shader_tessellator GL_ARB_color_buffer_float GL_ARB_compatibility GL_ARB_copy_buffer GL_ARB_depth_buffer_float GL_ARB_depth_texture GL_ARB_draw_buffers GL_ARB_draw_buffers_blend GL_ARB_draw_instanced GL_ARB_fragment_program GL_ARB_fragment_program_shadow GL_ARB_fragment_shader GL_ARB_framebuffer_object GL_ARB_framebuffer_sRGB GL_ARB_geometry_shader4 GL_ARB_half_float_pixel GL_ARB_half_float_vertex GL_ARB_instanced_arrays GL_ARB_map_buffer_range GL_ARB_multisample GL_ARB_multitexture GL_ARB_occlusion_query GL_ARB_pixel_buffer_object GL_ARB_point_parameters GL_ARB_point_sprite GL_ARB_provoking_vertex GL_ARB_shader_objects GL_ARB_shader_texture_lod GL_ARB_shading_language_100 GL_ARB_shadow GL_ARB_shadow_ambient GL_ARB_texture_border_clamp GL_ARB_texture_buffer_object GL_ARB_texture_compression GL_ARB_texture_compression_rgtc GL_ARB_texture_cube_map GL_ARB_texture_env_add GL_ARB_texture_env_combine GL_ARB_texture_env_crossbar GL_ARB_texture_env_dot3 GL_ARB_texture_float GL_ARB_texture_mirrored_repeat GL_ARB_texture_non_power_of_two GL_ARB_texture_rectangle GL_ARB_texture_rg GL_ARB_texture_snorm GL_ARB_transpose_matrix GL_ARB_uniform_buffer_object GL_ARB_vertex_array_object GL_ARB_vertex_buffer_object GL_ARB_vertex_program GL_ARB_vertex_shader GL_ARB_window_pos GL_ATI_draw_buffers GL_ATI_envmap_bumpmap GL_ATI_fragment_shader GL_ATI_meminfo GL_ATI_separate_stencil GL_ATI_texture_compression_3dc GL_ATI_texture_env_combine3 GL_ATI_texture_float GL_ATI_texture_mirror_once GL_EXT_abgr GL_EXT_bgra GL_EXT_bindable_uniform GL_EXT_blend_color GL_EXT_blend_equation_separate GL_EXT_blend_func_separate GL_EXT_blend_minmax GL_EXT_blend_subtract GL_EXT_compiled_vertex_array GL_EXT_copy_buffer GL_EXT_copy_texture GL_EXT_draw_buffers2 GL_EXT_draw_instanced GL_EXT_draw_range_elements GL_EXT_fog_coord GL_EXT_framebuffer_blit GL_EXT_framebuffer_multisample GL_EXT_framebuffer_object GL_EXT_framebuffer_sRGB GL_EXT_geometry_shader4 GL_EXT_gpu_program_parameters GL_EXT_gpu_shader4 GL_EXT_multi_draw_arrays GL_EXT_packed_depth_stencil GL_EXT_packed_float GL_EXT_packed_pixels GL_EXT_pixel_buffer_object GL_EXT_point_parameters GL_EXT_provoking_vertex GL_EXT_rescale_normal GL_EXT_secondary_color GL_EXT_separate_specular_color GL_EXT_shadow_funcs GL_EXT_stencil_wrap GL_EXT_subtexture GL_EXT_texgen_reflection GL_EXT_texture3D GL_EXT_texture_array GL_EXT_texture_buffer_object GL_EXT_texture_compression_latc GL_EXT_texture_compression_rgtc GL_EXT_texture_compression_s3tc GL_EXT_texture_cube_map GL_EXT_texture_edge_clamp GL_EXT_texture_env_add GL_EXT_texture_env_combine GL_EXT_texture_env_dot3 GL_EXT_texture_filter_anisotropic GL_EXT_texture_integer GL_EXT_texture_lod GL_EXT_texture_lod_bias GL_EXT_texture_mirror_clamp GL_EXT_texture_object GL_EXT_texture_rectangle GL_EXT_texture_sRGB GL_EXT_texture_shared_exponent GL_EXT_texture_snorm GL_EXT_texture_swizzle GL_EXT_transform_feedback GL_EXT_vertex_array GL_EXT_vertex_array_bgra GL_IBM_texture_mirrored_repeat GL_KTX_buffer_region GL_NV_blend_square GL_NV_conditional_render GL_NV_copy_depth_to_color GL_NV_explicit_multisample GL_NV_primitive_restart GL_NV_texgen_reflection GL_SGIS_generate_mipmap GL_SGIS_texture_edge_clamp GL_SGIS_texture_lod GL_SUN_multi_draw_arrays GL_WIN_swap_hint WGL_EXT_swap_control
Max. texture size: 8192
Max. texture units: 16
Max. fragment uniforms: 1024
Max. vertex uniforms: 1024
Max. varying: 64
Max. combined shader storage blocks: 64
Max. vertex shader storage blocks: 64

Execution could not continue.

Unable to load shader Default:
Init Shader 'Default':
Vertex shader:
Vertex shader was successfully compiled to run on hardware.
WARNING: 0:116: Compiler warning: varying is deprecated since GLSL1.3, we suggest to use out


Fragment shader:
Fragment shader failed to compile with the following errors:
ERROR: 0:5: error(#239) Declaration must include a precision qualifier or the default precision must have been previously declared
ERROR: 0:7: error(#239) Declaration must include a precision qualifier or the default precision must have been previously declared
ERROR: 0:15: error(#239) Declaration must include a precision qualifier or the default precision must have been previously declared
ERROR: 0:16: error(#239) Declaration must include a precision qualifier or the default precision must have been previously declared
ERROR: 0:17: error(#239) Declaration must include a precision qualifier or the default precision must have been previously declared
ERROR: 0:18: error(#239) Declaration must include a precision

Drivers are very old, I know. But because of crappy switchable graphics I cannot update them. Actually, these drivers are the latest available from vendor.

[Graf Zahl]

WTF!?!

Since when does full openGL need a precision qualifier? I thought they were solely for GLES. I know that my driver would fail to compile if I added them, I already experienced that with work stuff.
Also no idea where the 'varying' comes from. None of the shader sources still contains that word.