 |
|
| Quake2 |
|
|
By rudolfc - Posted on December 25, 2004 - 14:06:01 (#15167)
Current version when comment was posted: 1.0.0 |
|
|
BTW2:
(sorry)
Quake2 runs at 100.5fps (audio working with default setting), in 640x480 fullscreen, running with it's own software rendering mode.
(timedemo 1)
Bye!
|
|
| GL teapot |
|
|
By rudolfc - Posted on December 25, 2004 - 14:03:13 (#15166)
Current version when comment was posted: 1.0.0 |
|
|
BTW: the teapot is running at 200-300FPS with MESA6.2 compiled with the latest GCC from Oliver Tappe.
Quake2 openGL software rendering is at 4.8fps. (timedemo 1)
Rudolf.
|
|
| PCIe |
|
|
By rudolfc - Posted on December 25, 2004 - 14:01:02 (#15163)
Current version when comment was posted: 1.0.0 |
|
|
Hi,
Just tested this personally in 32bit mode on a PCIe GeForce FX6600GT, which is a real PCIe chipset: no AGP bridge exists for this card: speed is native PCIe speed. MTRR-WC is activated, nVidia driver is running (no 2D acc though yet, but that should not matter for for this test!)
CPU=3.2Ghz P4 FSB800.
Throughput reports 310Mb/sec.
I would be very curious what happens with raw data instead of drawing a bitmap ;-)
Bye!
Rudolf.
|
|
| Some Variance in the numbers |
|
|
By dmcpaul - Posted on November 7, 2004 - 01:29:59 (#14665)
Current version when comment was posted: 1.0.0 |
|
|
Heya,
I get some variance in the numbers
392Mb/s up to 419Mb/s
AMD 2Ghz with a Nvidia 5700 Ultra.
Cheers
David
|
|
| PCIe |
|
|
By stippi - Posted on November 4, 2004 - 06:45:16 (#14628)
Current version when comment was posted: 1.0.0 |
|
|
Hi,
I only published this thing because Rudolf had written in his newest driver release, that PCIe boards would be working. Rudolf, you must know this from someone, so please, write that someone an email and tell him/her to run this app! :-)
-Stephan
|
|
| ok, ok.. |
|
|
By rudolfc - Posted on November 4, 2004 - 05:11:33 (#14627)
Current version when comment was posted: 1.0.0 |
|
|
Once more me:
Just installed the patch. Quake2 now gets 88fps (so a bit lower as expected), but throughput is lower than I expected: now on this system 333Mb/sec. I guess the same applies here that also applies for the laptop: there is something else (software like) being the bottleneck (not the bus).
Rudolf.
|
|
| speeds / 2140Mhz timing issue |
|
|
By rudolfc - Posted on November 4, 2004 - 04:53:21 (#14626)
Current version when comment was posted: 1.0.0 |
|
|
Hi,
One last remark:
I am underclocking my P4 to 2140Mhz because above that, the system will malfunction: the media runs too fast, speed recogniztion is wrong: etc.
(maybe I should try that patch someday :)
Anyway: I underclock this CPU via lowering the FSB (frontside bus) speed of the CPU. The PCI and AGP clocks are connected to that, so these are lowered as well, lowering my benchmark results.
As I just pointed out, I should get 450Mb/sec when I increase the speed to be nominal. Also Quake2 would get the 92FPS (72/102) * 133 = 92 (more or less).
Unfortunately, BeOS messes this up also. Instead of a performance increase, you see a big-time decrease in speed if I do that. Both quake2 and this trhoughput app drop speed considerably.
(throughput gives me a poor 250Mb/sec speed now).
I guess the speed calcs suffer from this as well. Quake2 calc maybe wrong, but the actual frames being displayed go slower as well: you can see that by plain looking at it.
OK, this concludes what I have to say about the subject I think :-)
Rudolf.
|
|
| AGP / speeda |
|
|
By rudolfc - Posted on November 4, 2004 - 04:34:32 (#14625)
Current version when comment was posted: 1.0.0 |
|
|
Hi,
I have a small addition:
->Note that the P4 would probably get around 450Mb/sec with AGP and FW if the FSB would be 133Mhz(!!)
->Note that theoretically, the max througputs for the busses are:
PCI bus: 133Mb/sec, shared with all other PCI devices;
AGP bus in PCI mode: 266Mb/sec dedicated;
AGP in AGP1x: 266Mb dedicated;
AGP in 2x: 532;
AGP in 4x: 1064;
AGP in 8x: 2128;
PCIe with nVidia in AGP '8x' mode: 4256.
(nVidia uses overclocked AGP8x bridges: so AGP16x: this will thus be the bottleneck on PCIe)
Note that these speeds will not be reached: FW only uses that speeds if burst-writes are issued. Other transfers go at (much) lower speeds.
Burst-writes are issued on more or less sequential accesses to graphicscard RAM.
Rudolf.
|
|
| AGP / results |
|
|
By rudolfc - Posted on November 4, 2004 - 04:22:29 (#14624)
Current version when comment was posted: 1.0.0 |
|
|
Hi there,
First off, I want to let you know I looked at the source of this app and decided to remove the Sync() command from the loop, as it's redundant AFAIK and lets your performance results drop about 5-15%.
Then, I should tell you probably that a real good benchmark would be writing raw data to the framebuffer instead of using app_server / API functions.
That being said, I think that results are probably reasonable OK if you test in 32bit mode, without apps running.
OK, Here are my results now: (tested in 32bit unless otherwise noted)
->dual P3/500, R5.0.3, nVidia card, PCI mode and AGP mode without FW:
175Mb/sec (orig app), 195Mb/sec (sync removed)
->Celeron400 laptop with Neomagic PCI card (dano):
50Mb/sec in 16bit mode (orig app), 52Mb/sec in 16bit mode (sync removed)
(note that in 32bit it would probably have about 80Mb/sec speed with this app, but taht's not supported on this card)
->P4@2140Mhz, 102Mhz FSB (underclocked), nVidia card (dano):
-PCI mode 150Mb/sec (sync removed)(note that this is exactly the same as the P3 if you correct for FSB speed 133Mhz)
-AGP 4x + FW mode @ FSB 102Mhz (orig app) 327Mb/sec, 350Mb/sec (sync removed)
->laptop PentiumM 1.6 at orig FSB (so AGP running at orig clock) with nVidia card, AGP4x +FW
-352Mb/sec (orig), 360Mb/sec (sync removed).
(note that on this laptop there's probably another bottleneck that results in these relatively bad results(!):
Quake2 timedemo1 (software rendering) gives me 92fps on this system, while on the P4 I only get around 72fps. benchmarking with sending raw data directly to the framebuffer would probably increase results visibly here)
OK, that's it.
Now if someone could test on nVidia PCIe, with and without AGP manager in place(!!), then we would probably have a jaw-dropping experience ;-)
Bye!
Rudolf.
|
|
| new results |
|
|
By rala - Posted on November 4, 2004 - 02:41:05 (#14619)
Current version when comment was posted: 1.0.0 |
|
|
now after fresh boot without running apps:
5.0.3: 284,46 mb/s
rc3: 276,6 mb/s
|
|
| more results |
|
|
By rala - Posted on November 4, 2004 - 02:29:47 (#14618)
Current version when comment was posted: 1.0.0 |
|
|
it's 260 mb/s now with beos 5.0.3
zeta rc 3 with haiku agp and nv 0.30 264 mb/s
|
|
| good results |
|
|
By rala - Posted on November 4, 2004 - 02:27:30 (#14617)
Current version when comment was posted: 1.0.0 |
|
|
msi 694 d pro with geforce2mx400 with fw
dual pIII @ 933
haiku agp and nv 0.30
248,61 mb/s
|
|
| Re: Euan. No, you don't :-P |
|
|
By BiPolar - Posted on November 4, 2004 - 01:05:20 (#14616)
Current version when comment was posted: 1.0.0 |
|
|
LOL, don't you dare to disqualify me :-)
First time I win on something!
(sorry, couldn't resist)
|
|
| I'm quite upset that the SIS card is beating me. |
|
|
By Euan - Posted on November 3, 2004 - 20:12:48 (#14614)
Current version when comment was posted: 1.0.0 |
|
|
Must be because the card uses system ram. So isn't actually copying from RAM-to-AGP. Rather RAM-to-RAM.
That's cheating, and that result should be disqualified.
I win! :D
|
|
| Radeons |
|
|
By s_d - Posted on November 3, 2004 - 17:39:08 (#14613)
Current version when comment was posted: 1.0.0 |
|
|
Radeon IGP 320, 2GHz Celeron at IBM Thinkpad R40:
127.9
Radeon-noname DDR 32. PIII-550 MHz. Intel/Via chips.
159.8 (Be Inc. driver from OGL Beta package)
|
|
| voodoo 3 3000 |
|
|
By michael_s - Posted on November 3, 2004 - 17:29:28 (#14611)
Current version when comment was posted: 1.0.0 |
|
|
axp 2500+
mobo kt133a
2x128 pc166
exp/dano 5.1d0
1024x768x85 32 b/p
146 mb/s
|
|
| Radeon9000Pro |
|
|
By BePage - Posted on November 3, 2004 - 16:19:22 (#14610)
Current version when comment was posted: 1.0.0 |
|
|
101Mbyte on Radeon9000Pro/128Mbyte
Dual PIII 1Ghz
|
|
| benchmarking |
|
|
By rudolfc - Posted on November 3, 2004 - 15:05:49 (#14608)
Current version when comment was posted: 1.0.0 |
|
|
Hi,
Make absolutely sure you all test in 32bit mode: the app is greatly depending on the same colordepht being set on all systems: otherwise you cannot compare results.
32bit depth is the fastest with this test, and this one should be actually right. The other depths do some extra conversion or so, which seemingly slows transferrate down (more than a factor of 4!)
Hope this helps a bit :)
Rudolf.
|
|
| meh |
|
|
By Euan - Posted on November 3, 2004 - 14:05:07 (#14607)
Current version when comment was posted: 1.0.0 |
|
|
I get 202.11MB/S
System:
ATI Radeon X800XT AGP
Radeon Driver 4.1
Rudolfs AGP driver not installed
XP2500 @ 2GHz
1GB RAM
ABIT NF2 mobo.
|
|
| too slow for me! |
|
|
By xeD - Posted on November 3, 2004 - 12:47:12 (#14604)
Current version when comment was posted: 1.0.0 |
|
|
Abit VP6
Matrox G550.
only 56 Mb/s.
I think is very very slow for such a videocard.
Any idea how to get better performances? :)
|
|
| SiS730 Results. |
|
|
By BiPolar - Posted on November 2, 2004 - 14:21:34 (#14593)
Current version when comment was posted: 1.0.0 |
|
|
Just to share my results...
Hardware specs:
- PCChips M810L
- AMD K7 @ 900 MHz
- SiS730 onboard video card with 8 MB of shared mem. (AGP 4x, no FastWrite, AFAIK)
- 256 MB SDRAM @ 133 MHz
OS/Software config:
- BeOS R5 PE, using Exp/Dano's "sis" video drivers.
- Rudolf's AGP bus module/driver (doesn't helps for this card)
- Desktop res.: 800x600x32 @ 85 Hz
Results:
- Aprox. 38.2 MB/s [*1]
- Aprox. 228.3 MB/s [*2]
[*1] Results with VESA disabled.
[*2] Results with VESA enabled at boot (the "vesa" file reads: "mode 640 480 8").
As you can see, for this particular video card, you gain _a lot_ by forcing BeOS to do a "VESA init" of your card at boot, even while later the "sis" driver kicks in. Same happens under Exp/Dano, so I guess the same applies to Zeta.
|
|
|
|
|
