Hi Hank,

I would like to ask you a question about VSA-100 architecture.There is one thing that leaves me puzzled. Why 3dfx changed their texture/pipelines ratio? out of Voodoo 1 , all next chips were like
2TMU on each pipeline, which allows single pas multitexturing(for example Voodoo 3 = 1 pipeline/2TMU). So why VSA-100 didn´t follow this scheme ? it might have been 2pipes/each with 2TMU´s = 333Mpix/666Mtex. I think it would hepl performance of VSA 100 dramatically.

In this interview back in 1999 Scott Sellers says:

Quotethat it appears based on their spec sheets and functionality that the GeForce's 256 raster engine is likely 2 TNT2 pixel pipelines put in parallel. Clearly, the geometry capability is brand new for GeForce, but it appears as though the raster functionality is virtually identical to the TNT2 raster engine (except it's obviously doubled in width and they've also added cube environment mapping...

http://www.guru3d.com/tech/editorials/210999/index2.shtml

So why 3dfx did not "stitched" two Voodoo 3 pixel pipelines adding 32Bit color, High-rez textures support and T-Buffer thing

So was there any reason to change the scheme ? perhaps size of chip, or incompatibility with T-buffer, or what else

I will do appreciate any suggestion about that.

thank you
regards Daniel

yes it was Q3D board but definately not Voodoo 5600.... At that time 3dfx had only first revision with Intel bridge chip and that revision was non-functional. I guess they used Quantum 3D Mercury with hardware FSAA for Quake III Arena presentation and they just change name in bios for Voodoo 5600...



http://www.thedodgegarage.com/3dfx/q3d_mercury_faq.htm

http://www.anandtech.com/show/413/6

QuoteWhat it all boils down to is that 3dfx does not actually have working silicon back at this point in time. All they could tell us is that they are "eagerly awaiting" their first samples. At Comdex, they were still showing T-Buffer demos using Quantum 3D's 8-way Voodoo2 SLI setup.

Hello,

I remember article about 3dfx Voodoo 5600 early preview back in year 1999 -2000. They showed "Voodoo 5600" with Pentium III running Quake 3 Arena at 1600x1200 with 32 bit colour, maximum texture detail and trilinear filtering - all the bells and whistles.The game was also running with four sample anti-aliasing enabled at 60fps. But then later somebody proved, that it was tweaked bios and inside was actually Quantum3D board, because 3dfx didn´t finish board design yet and with Intel bridge chip the card was unfunctional.

I am not sure if it was at Comdex or ECTS computer games trade show.My question is, what graphic card was used instead of Voodoo 5600. I just can´t remember. All I konw it was some Quantum 3D board with 3dfx chips on it.

Perhaps AAlchemy 4116 ( it uses 4 Voodoo 3 chips in sli, but Voodoo 3 wasn´t able of 32 bit output and they claimed it was 32bit color in Quake test)




thanks for info


Daniel

Amigamerlin 3.1 R11 driver-set is based on SFFT Alpha 29 Direct3D core, contains a new OpenGL library and it's the first Amigamerlin release with a new dll. The installation program will correctly install to the system a dxtn.dll. This dll, which we have discuss here, it will be only a container lacking of functions (dummy file). The user should replace it with the real one compiling the source from Daniel Borca's site. Any way drivers will works well also without the replacement. We chose this solution to prevent copyright problems. Amigamerlin 3.1 R11 supports 3dfx Voodoo4 4500, Voodoo5 5500, and Voodoo5 6000 cards on Microsoft Windows 2000/XP Operative Systems.

where I can download this new file ? I have been looking on Daniel Borca web pages and I couldnt find any file with dxtn.dll name.

and should I replace it after instalation original Amiga drivers or before instalation ?

thanks

Daniel

QuoteOriginally posted by Ragnarok72

QuoteOriginally posted by del42sa


I think Quantum 3D used this idea in AAlchemy, they calling it "Performance Trilinear"

The Enable Performance Trilinear (PT) option uses special Quantum3D-developed
hardware functionality that enables rendering 2 pixels per clock rather than just one, while still
performing trilinear texture blending. Without Performance Trilinear, you would have to
choose between 2 pixels per clock operation and trilinear mipmapping. Using Performance
Trilinear will automatically force the maximum color depth to be 22 bits per pixel, but will
increase your fill rate. Applications that require the maximum fill rate should use this feature.

"On previous AAlchemy systems, enabling trilinear filtering forced the 3dfx VSA-100
chips on board into a single-pixel-per-clock mode of operation. By modifying the
Parallel Rendering Architecture FPGA and driver software in these new AAlchemy
models, we've enabled the 3dfx VSA-100 chips to provide trilinear filtering while
concurrently operating in a two-pixel-per-clock or 'Performance Trilinear' mode,"
stated Ross Q. Smith, Founder and VP of Sales & Marketing at Quantum3D. "With
Performance Trilinear, we've effectively doubled the fill rate performance on these
new models. AAlchemy was already the performance and price/performance leader
in the industry—now we've opened the gap between the competition and us even
further."

http://www.quantum3d.com/press/pdf/archive/pr_11-27-00-2x.pdf

The AA5™ family of advanced realtime 3D graphics subsystems is available exclusively for deployment in the Quantum3D® AAlchemy™ family of scaleable, open architecture PC-based Image Generators (PC-IGs). AA5 subsystems feature Quantum3D's Parallel Rendering Architecture that combines the power of 8 or 16 3dfx VSA-100 graphics chips. The AA5 supports Quantum3D's unique Performance Trilinear™ technology which enables 2-pixel per clock trilinear rendering, 0, 2, 4 and 8-way Scan Line Interleaving (SLI) and 4 or 8-sample T-buffer™ full scene anti-aliasing to enable vis-sim integrators to "fine tune" the PC-IG for desired performance and image quality levels.

Hmm....they very well may have.  Not to mention, if they are still only maxing out at 8x FSAA mode when they are using 8 or 16 chips, then it sounds like those extra 4 or 8 chips are probably only there so that their TMUs can be hijacked while the other chips are running in SLI....

I think, it is not problem for them to use higher sample antialiasing, but I guess beyond 8x FSAA the difference isn´t so big, but the performance penalty is heavy.

I am not sure, how they enabled two pixel per clock, while they still doing trilinear texture filtering at the same time, but I guess they must use other TMU´s from second board by their Parallel Rendering Architecture FPGA and driver software moddification or it must be another trick. Notice they saying it will automatically force the maximum color depth to be 22 bits per pixel(post filter), but will
increase your fill rate.........

you should ask ross smith from Quantum3D, mail him a feel free to ask him : rsmith@Quantum3D.com


Dan

QuoteOriginally posted by Ragnarok72

I just had a crazy idea.

I don't know if this would be more of a driver-design or hardware-design issue, or if this was already suggested, but while a V5 is running in Single Chip mode, would it be possible to use the TMU pipelines from the other chips during multitexture operations?

I.e., a V5 5500 in Single Chip mode uses the 2nd chip's TMU to gain an effective 4-texture-per-pass operation?  Now I'm no hardware technician like Hank is, but I would assume that since the card is a multiple-chip design, that there would be at least some underlying circuitry already on the PCB that would allow for some sort of communication between the chips, which could perhaps be utilized for this task without the need for hardware modification.  But since I don't know if there is, that would be where you guys come in.

Of course, on the driver end we'd probably have to re-code it to treat the card as a 1 VSA-100 with an additional off-die TMU unit, when it comes to Single-Chip Mode....I might be mistaken, but didn't the original Voodoo1 have that type of architecture, with an off-die TMU?

I think Quantum 3D used this idea in AAlchemy, they calling it "Performance Trilinear"

The Enable Performance Trilinear (PT) option uses special Quantum3D-developed
hardware functionality that enables rendering 2 pixels per clock rather than just one, while still
performing trilinear texture blending. Without Performance Trilinear, you would have to
choose between 2 pixels per clock operation and trilinear mipmapping. Using Performance
Trilinear will automatically force the maximum color depth to be 22 bits per pixel, but will
increase your fill rate. Applications that require the maximum fill rate should use this feature.

"On previous AAlchemy systems, enabling trilinear filtering forced the 3dfx VSA-100
chips on board into a single-pixel-per-clock mode of operation. By modifying the
Parallel Rendering Architecture FPGA and driver software in these new AAlchemy
models, we've enabled the 3dfx VSA-100 chips to provide trilinear filtering while
concurrently operating in a two-pixel-per-clock or 'Performance Trilinear' mode,"
stated Ross Q. Smith, Founder and VP of Sales & Marketing at Quantum3D. "With
Performance Trilinear, we've effectively doubled the fill rate performance on these
new models. AAlchemy was already the performance and price/performance leader
in the industry—now we've opened the gap between the competition and us even
further."

http://www.quantum3d.com/press/pdf/archive/pr_11-27-00-2x.pdf

The AA5™ family of advanced realtime 3D graphics subsystems is available exclusively for deployment in the Quantum3D® AAlchemy™ family of scaleable, open architecture PC-based Image Generators (PC-IGs). AA5 subsystems feature Quantum3D's Parallel Rendering Architecture that combines the power of 8 or 16 3dfx VSA-100 graphics chips. The AA5 supports Quantum3D's unique Performance Trilinear™ technology which enables 2-pixel per clock trilinear rendering, 0, 2, 4 and 8-way Scan Line Interleaving (SLI) and 4 or 8-sample T-buffer™ full scene anti-aliasing to enable vis-sim integrators to "fine tune" the PC-IG for desired performance and image quality levels.

Daniel

http://pub43.ezboard.com/fx3dfxfrm1.showMessage?topicID=16717.topic

is it true that he owns SLI


Dan