elianda wrote:

Here is an quite interesting thread about CPU scaling of various Riva128, V1, Rush, V2 , V2 SLI, V3:
http://www.voodooalert.de/board/index.php?pag … 2749&highlight=

This has been always a hot topic 😀

looking at these charts I am inclined to swap the P3-533 for a P3-800 on my Voodoo2-SLI retro box as it clearly can handle more cpu horsepower. The 800 is also period correct; 1999ish.

Man those V2's in SLi really wake up when paired with a fast CPU. It appears that CPU's in the 1.2ghz range about max them out, though. Faster than that and you should probably be considering another video card as the V2's will start to become a bottleneck to the system.

yes 1.2ghz looks like the sweet spot - these comparative benchmarks are very usefull. I have a P3-1400 tualatin powering a V5500 [AGP] in another box - I wonder how much more cpu that card can take before it plateaus like the V2-SLI's do @ 1200?

Well, notice the block of full length cards?

Let's put the last one into a test system:

With all these WRAMs...

...this must be a Ticket to Ride for two...

Guess it's time for another revolution...

More coming soon...

WTF! A dual GPU Number Nine Revolution!!!! When the hell did they make those and where did you get it?

This is infact an Avid Meridien III Cactus Board with 3D Effects Board (center double card sandwich), Pinnacle Genie Pro (front) and Display Adapter (back).
The Display Adapter consists of a Dual Number Nine Revolution PCI with 8 MB WRAM each.

Quirks:
The Hawkeye 4 driver with DirectX support in Win9x installs both adapters but only the main one is useable. System freezes on shutdown, seems like driver crash. The Hawkeye 4 driver does not seem to support a Dual GPU configuration properly.

In WinXP the driver that comes with Windows detects and installs two Revolutions and you have the common two monitor options with Clone, Dual View etc. But this driver does not support D3D.

NT4 driver is just a miniport driver that probably supports two monitors but no OpenGL 3D.

My first D3D tests show strong artifacts in 16 bit modes, but in 32 bit modes it seems to be ok. Still I am not sure if I can set some options for 16 bit like forcing a certain texture format to fix this. Speed on first tests seem to be between Riva128 and Virge.

Very interesting. Nice find.

Here some footage: https://www.youtube.com/watch?v=8n4xeWWC9eY

And here also some BIOS message image of the Imagine 128 in the front. The chip should have some basic 3D functionality, but I don't know which application supports some output to show this. It consists of a CL-GD5422 with 512kB for VGA output and the Imagine 128 with 4 MB VRAM for GUI 2D. There should be 128 bit buses on each datapath...

More stuff, Trident 3DImage 9750 from various half way working benches / games.
http://www.youtube.com/watch?v=_3rxzzolFzY

I can't recommend this card for Unreal. 😜

This is maybe interesting, while testing a new Voodoo I used the i740 as main card the CPU was an Athlon 800 MHz now (500 MHz before). So I redid 3DMark2001SE to check how the i740 scales:

Athlon 500 MHz / 800 MHz
Score 682 / 783
Car Low 13.4 /14.2
Car High 2.0 / 3.0
Drag Low 11.4 / 13.8
Drag High 4.7 / 5.8
Lob Low 17.8 / 17.8
Lob High 6.2 / 7.5
Fill Rate Single 53.6 MT / 53.7 MT
Fill Rate Multi 53.6 MT / 53.7 MT
Poly 1L 1.5 MT/s / 1.5 MT/s
Poly 8L 0.7 MT/s / 1.1 MT/s
Vertex 6.7 / 6.7
PSprites 0.3 / 0.3

So I would conclude that the i740 is with an Athlon 500 MHz already at it's limits regarding rendering.
You gain with a faster CPU still geometry setup speed for scenes with a lo of polys. The cap CPU speed for this might be estimated from the Poly 8L increase, 300 MHz more add 0.4 MT/s. To reach 1.5 MT/s another 300 MHz more would roughly be required, so somewhere close to a virtual 1.1 GHz Slot-A Athlon or a 1 GHz Thunderbird.

elianda wrote:

So I would conclude that the i740 is with an Athlon 500 MHz already at it's limits regarding rendering.

Did you consider that the i740 only does AGP texturing (ie. no local textures) so it might easily be limited by AGP performance? I don't know which platform would be best here, but IIRC the Slot A Mainboards where somewhat infamous in regard of their AGP implementation.

idspispopd wrote:

elianda wrote:

So I would conclude that the i740 is with an Athlon 500 MHz already at it's limits regarding rendering.

Did you consider that the i740 only does AGP texturing (ie. no local textures) so it might easily be limited by AGP performance? I don't know which platform would be best here, but IIRC the Slot A Mainboards where somewhat infamous in regard of their AGP implementation.

Slot 1/Socket 370 would probably be best. They are more stable.

Also, there was a PCI version of the i740. It uses a bridge chip between the GPU and onboard memory so it doesn't have to dip into system RAM to store and fetch textures and that card would sometimes be noticeably faster than the AGP version.

I didn't experienced stability issues.
For Slot-1 I currently have only a setup of 440LX with 133 MHz P2. If you have a suitable test system running, please report the results for comparison. Preferably some 440BX at 100 MHz FSB.
Or has maybe someone already made some comparison of various boards AGP performance?
How different is the performance in reality? What would be a good test?

I could setup a KT266A board though if the i740 fits.

You really expect that the i740 is still limited by AGP speed and not by the chips own performance?

Well I tried on an Elitegroup K7S5A with an AthlonXP 2400+ and it seems that the i740 does not work well with it. As soon as I try to use 3D modes, so basically AGP transfers it freezes. I cross checked with a Matrox G200 AGP and it works fine.

elianda wrote:

This is infact an Avid Meridien III Cactus Board with 3D Effects Board (center double card sandwich), Pinnacle Genie Pro (front) […]
Show full quote

This is infact an Avid Meridien III Cactus Board with 3D Effects Board (center double card sandwich), Pinnacle Genie Pro (front) and Display Adapter (back).
The Display Adapter consists of a Dual Number Nine Revolution PCI with 8 MB WRAM each.

Quirks:
The Hawkeye 4 driver with DirectX support in Win9x installs both adapters but only the main one is useable. System freezes on shutdown, seems like driver crash. The Hawkeye 4 driver does not seem to support a Dual GPU configuration properly.

In WinXP the driver that comes with Windows detects and installs two Revolutions and you have the common two monitor options with Clone, Dual View etc. But this driver does not support D3D.

NT4 driver is just a miniport driver that probably supports two monitors but no OpenGL 3D.

My first D3D tests show strong artifacts in 16 bit modes, but in 32 bit modes it seems to be ok. Still I am not sure if I can set some options for 16 bit like forcing a certain texture format to fix this. Speed on first tests seem to be between Riva128 and Virge.

😳

What's the "3D effects card" used for??

elianda wrote:

Well I tried on an Elitegroup K7S5A with an AthlonXP 2400+ and it seems that the i740 does not work well with it. As soon as I try to use 3D modes, so basically AGP transfers it freezes. I cross checked with a Matrox G200 AGP and it works fine.

Does the G200 use AGP texturing? The i740 uses ONLY AGP texturing, and that's the problem. The 8mb of onboard RAM on an AGP i740 is only for 2D framebuffer. Everything else is stored and fetched in main memory over the AGP bus. That's why the PCI version is so much better. The bridge chip fools the GPU into thinking it's storing and fetching texture data in main memory when it's really stored in the texture memory on the card. No slow transfers across the AGP bus and no collisions with the CPU when accessing main memory.

The Kyros were benched on the same system, so AGP works fine. If there is no AGP driver installed 3DMark2001SE reports OutOfVideoMemory if you try to run a bench. It seems to be more of a hardware incompatibility with the i740 and the Universal AGP.

So can anyone recommend a tool that reports real AGP transfer rates for a certain setup?

This way I could check if the Athlon 800 MHz on the AMD750 chipset is blocking the performance of the Intel i740.

http://www.thg.ru/mainboard/20000214/print.html

3DMark99 and 2000 have large texture tests to force AGP texturing. Run one that is well beyond the video card's local RAM capacity.

For D3D7 cards they use:
3DMark2000 'High Polygon, 1 Light' benchmark
NVIDIA SphereMark (archive.org link)
SPEC viewperf

There is an "Intel Platform Test v1.2" that has a DRAM to AGP test but I can't find it.

Before I start with the AGP stuff, I had some intermezzo with the SavageMX:
http://mail.lipsia.de/~enigma/neu/pics/s3_sav … mx_spectrah.jpg

Here some benchresults: ftp://78.46.141.148/pictures/S3_SavageMX/
As this card has just 4 MB RAM, 3DMark2000 complains and refuses to run. 3DMark99Max just freezes on start, same with Ati Dawning. So just Final Reality and 3DMark2001SE and some info pages. The card seems to be strongly fillrate limited, also framerate in Unreal varies a lot if reflections, coronas or fog gets visible (7 fps to 55 fps in 640x480).

idspispopd wrote:

elianda wrote:

So I would conclude that the i740 is with an Athlon 500 MHz already at it's limits regarding rendering.

Did you consider that the i740 only does AGP texturing (ie. no local textures) so it might easily be limited by AGP performance? I don't know which platform would be best here, but IIRC the Slot A Mainboards where somewhat infamous in regard of their AGP implementation.

ok here is some result matrix for 3DMark2000 HighPoly 1 light, 640x480x16:

the order is Kyro I / Kyro II / i740 in kPolys/s

AthlonXP 2400+, K7S5A : 5488 / 5924 / not working
Athlon 800, GA7-IX: 4500 / 4643 / 1287

So the Athlon 800 seems to be too slow for the Kyros but good enough for the i740.