PD 5.41 are the famous DX7 drivers !?

[Brama]

I received today a newsletter from Matrox Developers Group in which the new driver for Win9x is listed as DX7.

[Himself]

Current G200/G400 DX7 driver - build 5.41.008:
ftp://ftp.matrox.com/pub/mga/archive/win_9x/1999/w9x_541.exe

Yep, surprised me as well, oh well, guess I can stop waiting for more DirectX performance then. Makes sense actually, since 3DMark2000 was much faster using DX7 than the internal AMD routines on my system.

(Current results 1626 vs 1720 with new motherboard, previous was 1465 aprox)

[Buuri]

It cannot be. Must be a bug.

_
B

[Buuri]

I just browsed through your old posts about that, and I'm positive it's because of your Win95.
While there was no 3DMark! when 95 came out, is it feasible to support it under 95? Even if it is, that must be why you're getting your twisted results.

_
B

[Buuri]

Actually I think that the info given by 3DMark2000 is misleading. The "3DNow! mode" as well as the "DX7 mode" BOTH use DX7 for the geometry, but with the hardware/software separation.
So in software mode they only toggle the 3DNow! support from DX off.

_
B

[Meek]

Interesting explanation, Buuri, but I thought otherwise. Is your information from the makers of 3dmark2000?

I actually thought that if you have a hw-T&L board present, you get an extra option to use that for the benchmark. I thought the "Directx 7 software T&L" was using the built-in software rendering routines of DirectX 7, and thereby containing also built in 3dnow!" and SSE optimisations.

And the SSE optimised and the 3dnow-optimised renderers actually were using 3dmark's (Remedy's) own sw-rendering pipeline for the geometry, including transform & lighting.

On my Athlon system actually all the 3 possible scores are pretty close to each other.

M.

------------------
Athlon500/MSI6167/128M/10GIBM/6GSamsung/1GSCSI IBM/AHA2940/CL2xDVD/FastMM2+MJPEG/HPPSScanner/HPPSPrinter/G400DH32M/ZIP100SCSI/Mitsumi2xCD-R/Samsung17GL/MX300/AKA Electric Classic

[Himself]

Ok, I just reinstalled Win98 (had to for the new mb anyway since I use win98 to test code). As you can see the OS you use makes very minor speed difference here, from 1720 to 1753 is nothing too exciting in my book, nor is the AMD results from 1626 to 1670 all that interesting either.

SYSTEM
Software Versions:
Windows Version Windows 4 , Build 1998
DirectX Version 4.07.00.0700
Bios Version Award Modular BIOS v4.51PG
Bios Date 11/30/99

Memory:
Total Physical Memory 128MB
Free Physical Memory 73MB

PROCESSOR
CPU Information:
Processor Type AMD-K6(tm) 3D+ Processor
Processor Speed 448MHz
Processor Caps 3DNow!, MMX

Cache Memory:
L1 Cache Size 64KB
L2 Cache Size 256KB

DESKTOP
2D Display Adapter:
Name Matrox Millennium G400 - English
Driver Date 11-26-1999

Desktop Size and Color Depth:
Width 1280
Height 960
Color Depth 16 bit

TEMPLATE
Display:
Type: Internal
Width: 1024
Height: 768
Depth: 16-Bit
Buffering: Triple
Z-Buffering: 16-Bit
Refresh Rate: VSync Off
Texture Format: 16-Bit

<H1>CPU Optimization: D3D Software T&L</H1>

RESULTS
Platform: Internal
3DMark Result: 1753 3D marks
CPU Speed: 88 CPU 3D marks
Game 1 - Helicopter - Low Detail: 38.6 FPS
Game 1 - Helicopter - Medium Detail: 23.6 FPS
Game 1 - Helicopter - High Detail: 8.6 FPS
Game 2 - Adventure - Low Detail: 44.0 FPS
Game 2 - Adventure - Medium Detail: 20.1 FPS
Game 2 - Adventure - High Detail: 11.1 FPS
Fill Rate (Single-Texturing): 264.7 MTexels/s
Fill Rate (Multi-Texturing): 262.2 MTexels/s
High Polygon Count (1 Light): 1257 KTriangles/s
High Polygon Count (4 Lights): 847 KTriangles/s
High Polygon Count (8 Lights): 618 KTriangles/s
8MB Texture Rendering Speed: 214.9 FPS
16MB Texture Rendering Speed: 190.7 FPS
32MB Texture Rendering Speed: 114.0 FPS
64MB Texture Rendering Speed: 55.7 FPS
Bump Mapping (Emboss, 3-pass): 87.3 FPS
Bump Mapping (Emboss, 2-pass): 130.0 FPS
Bump Mapping (Emboss, 1-pass): 219.8 FPS
Bump Mapping (Environment): 94.4 FPS

<H1>CPU Optimization: AMD 3DNow!(tm)</H1>

RESULTS
Platform: Internal
3DMark Result: 1670 3D marks
CPU Speed: 81 CPU 3D marks
Game 1 - Helicopter - Low Detail: 33.7 FPS
Game 1 - Helicopter - Medium Detail: 19.4 FPS
Game 1 - Helicopter - High Detail: 5.4 FPS
Game 2 - Adventure - Low Detail: 46.1 FPS
Game 2 - Adventure - Medium Detail: 22.2 FPS
Game 2 - Adventure - High Detail: 12.3 FPS
Fill Rate (Single-Texturing): 264.7 MTexels/s
Fill Rate (Multi-Texturing): 262.2 MTexels/s
High Polygon Count (1 Light): 1082 KTriangles/s
High Polygon Count (4 Lights): 972 KTriangles/s
High Polygon Count (8 Lights): 862 KTriangles/s
8MB Texture Rendering Speed: 228.2 FPS
16MB Texture Rendering Speed: 204.9 FPS
32MB Texture Rendering Speed: 127.3 FPS
64MB Texture Rendering Speed: 63.5 FPS
Bump Mapping (Emboss, 3-pass): 80.4 FPS
Bump Mapping (Emboss, 2-pass): 119.5 FPS
Bump Mapping (Emboss, 1-pass): 218.2 FPS
Bump Mapping (Environment): 94.7 FPS

[Buuri]

Meek,

Sounds like a very good explanation. Yet, it troubles me they give an impression that "DX7 software" mode wasn't using 3DNow!/SSE at all.

_
B

[Himself]

I get the same thing under Win98. Don't have any numbers handy but the same pattern was there. You can boost the scores up a fraction, that's about it. And really, DirectX 7 is the same under both, the video card drivers are the same under both, it would be rather unlikely for it to be an OS thing. The OS difference would affect both varients of scores at the same time. It could very well be 3DMark2000 at fault sure, that's my opinion as well, (3DMark99Max for example was extremely screwed up with AMD cpus), but DX7 also has 3D-Now! support, so it's really comparing DX7's 3DNow support vs DMark2000's 3D-Now! support. In fact if you looked at my numbers, 3DMark2000 had spots of faster performance faster in AMD mode, but overall DX7 mode was faster.