Questions regarding SLi

[Twinkie]

I ask this because someone on the another forums seem to suggest something entirely different.

Questions regarding SLi:

Does the 7950GX2 use the same technology behind SLi and i.e benefit from its advantages and suffer from its disadvantages?

This is what this person wrote about SLi

SLI doesn't work like that, the driver splits/alternates the frames between the two cards, then retrieves the information from the second card, sends it back to first card which then synthesizes the two pieces together to output the final finished video stream. Thats why although there are 2 cards rendering the image, the bandwith does not double, and neither does the frame buffer, because yes the rendering workload is split but the compiling is still done in the primary card.

And then fired off a statement that still puzzles me.

I hope this helps you understand why high resolution output with 8800GT SLI cannot perform well, and why it's a waste of money.

From my understanding SLi theoretically doubles shader power while effectively doubling bandwidth/fillrates. The only reason cards dont double the performance is because of various factors like driver overhead, AFR/SFR limitations, lack of better communcation between the two boards i.e increasing latency between the two cards etc etc.

Am i wrong to think that a 8800GT SLi is a better buy then a 8800GTX and it will perform better than a single 8800GTX when using high res and high AA?

 

[ShaidarHaran]

That person described SFR and called it AFR. I don't think they have a clue. Multi-GPU rendering excels @ high-resolution/high AA gaming. 2x 8800 GT should be faster than a single 8800 GTX in just about every instance.

 

[ChrisRay]

Technically SLI is most optimal where fillrate is your primary bottleneck. This reflects Z-fill/Pixel Fillrate. You will get scaling in shader limited situations. But even with AFR its very unlikely that you will achieve 80% + scaling. Typically SLI prefers higher resolutions and AA because the bottleneck tends to shift towards fillrate z-fill.

Now when the fillrate becomes a primary bottleneck. You will get more optimal use of the secondary GPU. This includes vertex/pixel scaling.

Chris

 

[Twinkie]

So basically thats why the fillrate "doubles" effectively. And this is why high AA/high res settings benefit alot from SLi because of the fact that single cards are severly fillrate limited at those settings?

Doesn't this mean bandwidth too increases by twice the amount? It seems that the most efficent method of Multi GPU rendering to date is AFR.

However i know that the frame buffers dont double because the output is being sent out from the master GPU. (the completed frame from the second GPU is sent to the first GPU via SLi bridge)

 

[ChrisRay]

"Double" may be a bad choice of words. Like I said it depends on how optimally SLI is scaling to begin with. In pure bandwith limited situations SLI doesnt double bandwith or scale very well period. Good examples of this would be HDR limited scenes without high resolutions/AA combined. But if the second card "is" being utilized well then you will get essentially double everything. But thats alot of "Ifs".

But assuming all things are optimal and the second card is distributing load evenly with the first.((as I said this usually requires a good fillrate bottleneck)) Then yes you technially double everything but framebuffer.

 

[Twinkie]

Alright thanks for the help Chris and everybody else aka Shaidarharan.

Last question: The 7950GX2 IS based on SLi technology and as ive stated suffers from SLi drawbacks but benefits from its advantages (excelling in fillrated scenarios)?

 

[ChrisRay]

Yup. The SLI switch is built onto the GPUs. ((the little interconnect between the 2 PCB)).

Chris