• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

What bottlenecks SLI cards?

H

Hiyruu

2[H]4U
Joined
Jul 28, 2001
Messages
3,611
Like Nvidia's new SLI, why doesn't

2 cards= 200% ?

Why does the second card only give an extra '80%' gain or whatever? What is the bottleneck?

Also, will SLI be limited to only 2 cards?
 
The 2nd card has to pre-render everything it sends over (latency) while the main card has to just render in mostly realtime and then merge the buffers.

So you answer is mainly latency not a bottleneck (well besides the rest of the system). Just like how a processor running at a certain speed will always (or mostly) be faster than a dual processor at half the speed.
 
what? what is it?

Latency or CPU?

Also, why does the second card have to "send over" the data? Can't they both do it directly?
 
the second card only has an 8x PCIe where the main card has 16x.

then theres some overhead from syncronizing the 2 cards.
 
chrisf6969 said:
the second card only has an 8x PCIe where the main card has 16x.
Click to expand...

Hmm, why are they delibretly only giving it 8x PCIe? Seems they're creating their own problem there.
 
The current design doesn't support more than one 16x slot. Please correct me if I am wrong.


***{Something to do with available bus width total being 32 lanes. So you got 16+8+(8x1)=32?} NOT 100% sure on this.
 
The 8x slot is not a bottleneck considering it is only getting around half of the signal.
 
Overhead will cause some loss, and the rest is limited by the cpu. End of discussion, that is it.
 
Correct. Current PCIe implementation supports 32 lanes, so two 16x PCIe slots would be all of the lanes (e.g., there wouldn't be any 1x slots). They use a 16x and an 8x to get around this, and 8x isn't a bottleneck as it still supports 2GB/s (AGP 8x, which is no where near used with current technology, is 2.1GB/sec).

I imagine some of the cycles are lost to load-balancing and other overhead-requiring tasks. It really isn't that big of a deal.

If you get 56FPS in FarCry with one card, adding a second @ 80% increase is 100.8FPS
If it were to get the full 100% benefit, it would be 112FPS. I really don't think an 11.2FPS difference is all that extensive, especially at higher framerates.
 
So it's down to the CPU? I never would of guessed that.

So as time goes by, and faster CPUs come out, this gap will then close.


which is no where near used with current technology
Click to expand...

can you explain? Are you saying that current technology is holding the gfx cards back?
 
it would be nearly impossible to get a staight up 200% gain.
with the little bridge connector and it having to process more information, it amazing they get the preformance increse they do.
 
current technology isn't holding the graphics cards back, current technology isn't being utilized by today's graphics cards.

The fastest of fast video cards (the x800xt PE and 6800u EE) do not max out the 8x AGP bus.

Graphics cards using PCI-e will not significantly faster than their AGP counter parts, and potentially will be slower due to both graphics card manufacturers using a PCI-e bridge on their cards. (Yes I know ATI claims not to, but nvidia "blew the lid" so to speak by x-raying the card and pointing out the area which is indeed a bridge). The ONLY advantage in the next few years that PCI-e cards will see is the ability to run SLI. Other than that, AGP is more than "good enough".
 
Well, all i know is that im looking into a dual proc xeon SLI setup, that is the platform that Nvidia is seeing 190% performance increases, and in some cases they "claim" near 200% over one card
 
DigitalEmperor said:
(Yes I know ATI claims not to, but nvidia "blew the lid" so to speak by x-raying the card and pointing out the area which is indeed a bridge). .
Click to expand...
not true, they were wrong, it turned out to be a buffer or something and not a bridge
 
advanced101101 said:
not true, they were wrong, it turned out to be a buffer or something and not a bridge
Click to expand...

link? never afraid to admit being wrong but I haven't read anything like that.
 
advanced101101 said:
Well actaully if you go down this description is available:

giga.bmp
Click to expand...

I thought the x8 was irrelevant anyway.
 
1. There's only 1 16x and an 8x because yes, the current PCI-e standard only allows a maximum of 32 lanes. nVidia's nForce 4 will (supposedly) feature two 16x lanes.

2. The PCI-e is not the limiting factor. AGP 4x is enough for current games as AGP 8x offers no improvents over AGP 4x, if just a small half percent increase in framrate.

3. The limiting factor is the CPU.

End of story....
 
anyone plannin on a dual xeon dual sli setup should understand that it depends on teh software.

nvidia is using a different sli implent then 3dfx did. so that is also a factor.
 
I'd have to say that the algorithm that splits the screen also bottle necks it, it can't perfectly split the shader/fillrate load so the card with less work is forced to wait.
 
there is no bottleneck....it''s only 80% because you NEVER, i repeat, NEVER get 100% efficiency with parallel processing. Just like SMP Computers, you don't get double the performance, you get betweek 50%-80%
 
You must log in or register to reply here.
Back
Top