Remember when you could install two 3DFX cards hooked together via SLI cable to Double your Speed?

[Luthien]

Oh, yeah I am an old timer. I think that Nvidea or ATI could capitalize by doing the same thing with each generation of its new cards giving those ultimate FPS to those willing to PAY for it. BTW, yes I did it mayself with the 3DFX cards. I remember wishing I could do it with three cards. Long time ago so dont remember the details very well. I think I was impressed with my 100FPS or something with Quake one. Just imagine two 6800U hooked together doubling their performance.

 

[BFG10K]

think that Nvidea or ATI could capitalize by doing the same thing with each generation of its new cards

That would be silly idea. For a start how are you planning on installing two AGP cards into the system at once? Secondly that cable would horrendously degrade image quality.

 

[Luthien]

It was done so it can be done again without horrible degredation of video quality couldnt it? How will it degrade performance? PCI and AGP cards... You, can buy dual AGP mobo's now albeit they are not in big cirulation and it wouldnt be enough for this. Two PCI express slots will be available within a year. Yes, it was before agp I think (not positive and dont feel like looking it up but it might have been one agp and one pci, cant remember). It wasnt a silly idea then so why is it silly today?

 

[BFG10K]

It was done so it can be done again without horrible degredation of video quality couldnt it?

The minute you added the cable it degraded IQ significantly.

You, can buy dual AGP mobo's now albeit they are not in big cirulation

I haven't seen or even heard of any.

Two PCI express slots will be available within a year.

It might work with PCI-E but it'll still degrade IQ.

Yes, it was before agp I think (not positive and dont feel like looking it up but it might have been one agp and one pci, cant remember).

Actually it was two Voodoo2s each sitting in a PCI slot of their own.

It wasnt a silly idea then so why is it silly today.

Because back then people had nothing better to increase performance.

 

[Luthien]

LOL, BFG I guess your dead set against it. So, best bet is two PCI express slots. Cant the cards have special instructions on the chips that offset degredations? You know if the degredation was slight it would sell like crazy.

 

[BFG10K]

We've already had multi-chip solutions from 3dfx and ATi and both failed in the grand scheme of things. History has proven that it's much faster, cheaper and easier to just keep increasing the performance of single chip solutions.

 

[sodcha0s]

We've already had multi-chip solutions from 3dfx and ATi and both failed in the grand scheme of things. History has proven that it's much faster, cheaper and easier to just keep increasing the performance of single chip solutions.

The only thing history proved was that 3dfx didn't aim high enough with the design of their chip, as in it wasn't powerful enough. SLI technology was a REVOLUTION (here we go again...) in the 3d gaming market, and in fact doubled performance. ATI or NV could use SLI technolgy with current chips I would think, although there is no need for dual boards. A single board solution would be the way to go. The question is, would anybody pay the huge cost for this kind of board.

edited for typo

 

[biostud]

I had two voodoo II 8mb in SLI

 

[Luthien]

Yeah that is what it was called SLI. At that time if you had that setup you it was like having a 6800U now.

 

[Bateluer]

Originally posted by: Luthien
Oh, yeah I am an old timer. I think that Nvidea or ATI could capitalize by doing the same thing with each generation of its new cards giving those ultimate FPS to those willing to PAY for it. BTW, yes I did it mayself with the 3DFX cards. I remember wishing I could do it with three cards. Long time ago so dont remember the details very well. I think I was impressed with my 100FPS or something with Quake one. Just imagine two 6800U hooked together doubling their performance.

Do they sell 900 watt PSUs? Or would you have to have twin 500s?

 

[Mingon]

not heard of the new creative 3dlabs card then ? this is going to use both a pci express 16x slot AND an adjacent normal slot for what seems to be a geometry processor. Rather than have dual cards I think we will see more setups like this espescially from nvdia. The rampage/sage core from 3dfx was rumoured to have an on card chip that made similar calculations, an we know who owns the tech now.


more info

more blah blah for the linkified lazy peeps

What is Wildcat® Realizm? technology?
Wildcat Realizm technology, the latest professional graphics architecture from 3Dlabs, combines the strength of programmable Visual Processing Units (VPUs) with a unique Vertex/Scalability Unit (VSU) for the industry's first dual-VPU configurations. Wildcat Realizm-based products will offer single and dual-VPU solutions and will support 16-lane PCI Express and AGP 8x bus standards . Performance is raised an astounding 100% over any other VPU solution with onboard memory capabilities of up to 1 GB. Wildcat Realizm is specifically designed to provide unmatched graphics productivity for CAD, DCC, video editing and visualization professionals and is a significant advance in effortlessly accelerating industrial-strength shader programs written in high-level shading languages such as OpenGL ® Shading Language and Microsoft® DirectX® 9.0 HLSL.

What is a VPU?
A VPU is a V isual P rocessing U nit and was first introduced to the professional graphics market by 3Dlabs. The Wildcat Realizm VPU is an extreme, high-performance graphics device that couples end-to-end compiler-friendly shader programmability with complete hardware acceleration.

What is a VSU?
VSU stands for Vertex/Scalability Unit. The Wildcat Realizm VSU, the first of its kind, is designed to manage single and dual-VPU configurations. It provides additional hardware acceleration for enhanced geometry processing and supports DirectBurst? memory capabilities with complete shader programmability. The Wildcat Realizm VSU takes full advantage of the bandwidth of PCI Express with a full x16 PCI Express interface.

What are the advantages of Wildcat Realizm's scalable architecture?
The Wildcat Realizm VSU receives graphics commands at full bandwidth from an x16 PCI Express interface and processes vertices with an incredible 67 billion floating point operations per second of high-precision floating point power in a SIMD (Single Instruction Multiple Data) array of highly optimized vector processors. The VSU is then able to drive two VPUs at full bandwidth over a 8.4 GB/sec interface while optimally distributing graphical primitives between the two VPUs to achieve a doubling of both geometry and fill-rate performance .

What benefits does Wildcat Realizm technology provide to end-users?
Wildcat Realizm technology combines the ground-breaking programmability of Wildcat VP with the unmatched geometry processing capability of Wildcat4 to provide industry-leading productivity for both traditional applications and new-generation shader-based software. Wildcat Realizm technology delivers an unbeatable combination of sheer performance, advanced shader programmability and professional-grade display quality.

Will products based on Wildcat Realizm technology support advanced features pioneered by other Wildcat products?
Absolutely. All the features that make Wildcat the professional's choice, such as large physical memory and virtual memory support; advanced shader programmability; SuperScene full-scene antialiasing; 64-bit accumulation buffers; DirectBurst memory and Framelock & Genlock are all supported.

What is new with Wildcat Realizm technology?
Wildcat Realizm technology combines several graphics innovations into one powerful solution. The new technology includes virtual shader memory that stores up to 256 K shader instructions onboard and executes at full hardware speed; seamless conversion between 16-bit and full 32-bit floating point values to optimize both processing precision and storage requirements for floating-point images; direct display of floating-point framebuffers; simultaneous access of up to 32 textures for sophisticated single-pass shaders; and extended 256 GB virtual address space to take full advantage of new 64-bit platforms.

What is DirectBurst? memory technology?
The Wildcat Realizm VSU supports DirectBurst memory, a capability unique to 3Dlabs' products. DirectBurst is onboard cache memory, which the host uses to cache command buffers, display lists, vertex arrays, and vertex objects locally on the graphics accelerator, significantly boosting the performance of any geometry-intensive application. DirectBurst uses a dedicated memory reserve, so the caching does not decrease framebuffer access speed, as do competing solutions.

Can Wildcat Realizm technology drive professional-class displays?
Wildcat Realizm technology can drive dual high-resolution screens at high refresh rates for the ultimate in display quality, including 9.2 Megapixel displays at full 3480 x 2400 resolution with full-scene antialiasing enabled.

What graphics APIs are supported by Wildcat Realizm technology?
Wildcat Realizm technology will support OpenGL 1.5, which includes OpenGL Shading Language and DirectX 9.VS 2.x/PS 3.x, which support the Microsoft HLSL. Graphics accelerators based on Wildcat Realizm technology will also fully support OpenGL 2.0 when the new API is released in mid-2004.

Why is OpenGL Shading Language support so important to design professionals?
The OpenGL Shading Language is an open industry standard enabling any programmable graphics hardware to be controlled via a C-like shader language to deliver high-quality cinematic and photorealistic interactivity. OpenGL Shading Language will be available on all vendors' hardware and on multiple platforms, including Microsoft Windows® and Linux® . OpenGL Shading Language is a significant step beyond languages such as Cg and HLSL, which simply translate a C-like program into another language such as OpenGL or Direct3D. The OpenGL Shading Language integrates a full compiler into the driver itself, providing the direct compilation of high-level shader programs into VPU machine code for the highest performance, unlimited functionality, and unrestricted architectural innovation.

What bus interfaces does Wildcat Realizm technology support?
Products built with Wildcat Realizm technology will take full advantage of both AGP 8x or x16 PCI Express buses.

 

[nRollo]

Originally posted by: BFG10K
We've already had multi-chip solutions from 3dfx and ATi and both failed in the grand scheme of things. History has proven that it's much faster, cheaper and easier to just keep increasing the performance of single chip solutions.

I can't believe this. I've finally found something BFG and I disagree on.

Apparently you weren't around in the old days of SLI or AFR, BFG.

1. There was nothing that was even close to faster than the sli BFG. I don't remember there being a lot of image degradation either.

2. There was no IQ degradation at all with the AFR, only some problems synching the rendering for a smooth framerates. The MAXX was the second fastest card you could buy behind the GFR DDR. If it would have had DDR, it may well of been as fast.
http://www.anandtech.com/video/showdoc.html?i=1116&p=13

Loved my MAXX and SLI, these were truly "enthusiast" solutions. I'd buy a 9800XT MAXX today if one were available for $500..
(I bet a LOT of people would)

 

[oldfart]

There was image degradation when you went to SLI. The image became more "grainy". Reducing the refresh rate to 75 Hz (the V2 drivers had a refresh rate independent of the 2D mode) helped but didn't completely eliminate it. 85 Hz and up was very noticeable.

 

[Soulkeeper]

let's see the two v2 idea turned into a v5 which was a better idea imo (putting both chips on one board)
with the transitor counts on current single chip gpu's tho i doubt seriously that this would be economical

basically the performance doubles every die shrink as it is with single chip solutions
they just double every individual peice they need more of in the chip design
no need to add more gpu's to the mix when yur chips already handle 16 pixels at a time

it all comes down to cost

 

[Luthien]

Well, havent any of you noticed the parallel this subject has with multiple cpu motherboards? At some point before another technological leap multiple cpu's will be a standard and most likely the same will apply in the graphics arena.

 

[Keysplayr]

Originally posted by: Soulkeeper
let's see the two v2 idea turned into a v5 which was a better idea imo (putting both chips on one board)

with the transitor counts on current single chip gpu's tho i doubt seriously that this would be economical



basically the performance doubles every die shrink as it is with single chip solutions

they just double every individual peice they need more of in the chip design

no need to add more gpu's to the mix when yur chips already handle 16 pixels at a time



it all comes down to cost

Well, they cant die shrink to infinity now can they. Eventually, cpu's and vpu's will have to go multiple (possibly in 1 core) because they cant shrink any more.

 

[Sunbird]

Hey,they could make twice as much money because people would buy twice the cards,
thats a good thing I would think.

 

[reever]

Instead of buying a car you could go out and buy a Simfusion system, complete with up to 32 R300 gpu's

 

[Pudgygiant]

Heh, I was doing this last week actually with an s3 virge and a v2 1000 just for shits and giggles while I was waiting for my new card.

 

[Creig]

SLI was a good idea that should've caught on. If a consumer only wanted X 3D performance they could purchase one card. If you wanted top-notch performance, just slap another in your machine. If you had an SLI setup back then and one card died, you could simply pull it out and RMA it while still being able to use your machine. You can't do THAT with a single card solution.

A motherboard SLI implementation would be even better. That way you could use any video card right off the shelf. Have each individual card render alternate frames (SLI was only capable of alternate lines) and send them out the DSUB connectors to a breakout box that would combine them and ship the correctly ordered frames to the monitor.

Again, this way a consumer doesn't HAVE to purchase two cards if they don't want to, but have the option of doing so if they want higher performance. And since each card would be rendering alternate frames you should see a real-world frame rate of 2X over the single card.

This would also increase the lifespan of your video card. Say that you have a mid-level card that's having trouble with the latest game you purchased. Instead of having to plunk down $250-$300 for a top of the line card, you can simply purchase another mid-level card identical to the one already in your machine and you would have framerates equal to (or above) the single top of the line card.

 

[Pudgygiant]

Originally posted by: Creig
SIf you had an SLI setup back then and one card died, you could simply pull it out and RMA it while still being able to use your machine. You can't do THAT with a single card solution.

Not with the v2's. Those were purely 3d accelerators and needed a standard 2d card.

 

[Creig]

True. I was referring to the 3D SLI cards, not the 2D card. You could pull one 3D card out for RMA and still be able to use the other one that's left.

 

[dWhisper]

Technically, it has been done since in one form or another. The major problem with SLI wasn't just the image degrade, it was the method that it did the image at. Each card did half of the rendering, which shared PCI bandwidth. So you never got twice the speed, you got a performance boost from two cards. The old obsidion cards got around the image degrade by having a monstorous PCI card which had SLI integrated on one board.

It also just isn't needed with the AGP bus. The idea worked well because AGP hadn't stuck it's head up in force just yet, so the extra memory really helped. The predecessor to the ATi Radeon cards did a similar solution, except it had two Graphics Renderers built into the board. And instead of doing an every-other-line like SLI did, it did a top-and-bottom format. But the Rage cards sucked, so it didn't ever matter.

Graphics cards have taken a very large architectural change since the GeForce cards first came out, and introduced a graphics processing unit. The GPU offloaded the processing of 3D instructions from the CPU, and increased dramatically what a 3D card could do. With the 3dfx cards, the chip was something utilized by the CPU to calclate functions. Since then, there have also been other increases like pipelines, shaders, etc, that have increased what a GPU is capable of, independant to the CPU.

Basically, it means that adding another GPU wouldn't really help you out in the long run. The current bottleneck for most cards isn't the processing, it's the rendering and functions used on the cards. We've been seeing much smaller pushes in clockspeed increases, and much more in the number of included pipelines. Look at the jumps taken between the Radeon 8500, 9700, and 9800. Big jump, then a smaller jump. Then, look at the jump from the 128meg 9800 to the 256meg 9800. If you squint really hard, you might see the difference.

True, it would add a lot if we had a 32-pipeline card(s) out there, but the extra memory, GPU and the like would not have nearly the effect they had with the 3dfx cards.

 

[Schadenfroh]

i got 3 voodoo2 12mbs, 2 of which can be used in SLI, 1 of the voodoo2s is currently in use, i have 2 voodoo5 5500s, one of which is in use. SLI was a great technology, but as BFG said, the cable would degrade quality, unless some motherboard designer found a way to use some sort of bridge that allowed 2 pci express boards to communicate with each other a VERY high speeds with a digital connection between them. the best way to do SLI is a Single board solution, but that will fail because of the horrindous cost (if you put dual 6800 ultras on a single board that would cost around $1000+) Plus, if you were to use modern cards in SLI, you would run into problems. the massive heat and power requirements plague the nwer cards, just imagine if you had twice the NV40s blowing heat into your case

 

[Elcs]

What about that Obsidian card? Cant remember what its name was but didnt it do 2 Voodoo 2 chips on 1 board effectively doing SLI?

I was a bit young to get into specifics around then so Im not 'with it'. Since I dont know much more, I cant comment further but surely its possible to make an effective dual chip card without a tremendous cost or even an SLI arrangement.

SLI was where 1 card did 1 line, the other card the next right? Wouldnt this present a problem with AA and AF? Although I dont think 'introducing a cable' would degrade IQ to a degree by which it was visible or at least without great scrutiny.... I can see that with all the advances in graphics technology behind the basics that producing a 2 chip or 2 card design would encounter some difficulties.

CPU's are looking to go dual core rather than dual CPU.... perhaps dual core GPU's rather than dual GPUs?

I think its 'wait and see' time again.