Multi-GPU Setups: The Basics Of CrossFire And SLI

[falchard]

It really does not matter because neither company has perfected the drivers to work on all games with near 100% scaling. I can make ATI look good in this test just by choosing games their drivers scale better to then nVidia which makes the comparison moot. Until a time when they get the 2 video cards to work as 1 its nothing more then a gimick.

 

[zyzeast]

[citation][nom]vicsrealms[/nom]I haven't done a lot of research on SLI systems, but I have been looking at them of late. Can you use an SLI system on an AMD board or are they only setup for XFire? All the AMD boards I look at seem to be XFire.[/citation]
There are nvidia chipset AMD board out there that supports SLI only and not Crossfire, although they are somewhat rare. Sear AM3 SLI on newegg and you'll see the AM3 Boards that supports SLI, there is only a few of them though and none supports USB3.0 and SATA6GB/s

 

[Honis]

This arrangement doesn't really have an impact on performance, since all three cards will slow down to the speed of the Radeon HD 5830.

With my personal testing using the [hacked] Final Fantasy 14 benchmark I was able to get performance increases by overclocking my second card higher than my master card. My setup is an Radion HD 4870 in the master slot and HD 4890 in the second slot.

Can you guys double check this with ATI or in a follow up article?

 

[sheol]

Hm how does the NF200 chip count into it? As the i5 has the PCI-E controller built in with 16x lanes max, it is still connected to the GPU-s over a x16 link, not wider. Or maybe I am misunderstanding it?

 

[mindless728]

Hm how does the NF200 chip count into it? As the i5 has the PCI-E controller built in with 16x lanes max, it is still connected to the GPU-s over a x16 link, not wider. Or maybe I am misunderstanding it?

from what i understand is that it multiplexes the lanes between the cards basically giving each card the bandwidth when t needs it, as the cards are usually working through AFR (Alternate Frame Rendering) each card needs data from the cpu/ram through the PCI-E bus each time it needs to draw a new frame, though if both cards need the bandwidth at the same time (like in SFR or AA mode) then i would imagine the load is balanced between the two cards (each getting x8)

 

[Guest]

http://techreport.com/articles.x/18825/6

Techreport has also looked at scaling with CF 5870s at very high resolutions. Nothing was bottle necked by PCI-E lanes.

 

[demonhorde665]

[citation][nom]cknobman[/nom]Just asking: Why compare crossfire of ATI current gen top of the line against Nvidia SLI last gen cards??? Especially when you have already done several SLI benching in other articles on the new Nvidia gen cards?Also the point of SLI is super high res so why are you benching at 1080p? A single graphics card can handle that resolution nowadays.[/citation]


has it ocured to you that the 400 series was not out before they started doing the legf work of this article ??? (yes i know he mentions the 400 series at teh end ) but the actuall articlewas liekly written after doing the bench test and all , and it s highly likely there were no 400's at the time teh test started (or they were hard to get for this test)

 

[deuketc]

I have a Q6600 @ 3.2 running a xfx 5870. always wondered if my cpu is a bottlneck for it... any ideas? considering an upgrade

 

[phantomtrooper]

I wanna see CPU/SLI scaling with dual core vs. quad core vs. core i7

 

[elkein]

The P55 bottle-neck is mostly contrived, running two 9800 gtx+ on mine, yes they scale normally and properly in SLI.

 

[torbee]

I may be a little thick, but the article seems to just skip over the mix and match options possible with different types of cards. I have a single GTX285 and want to buy another card to set up in SLI but GTX285's are hard to find and are quite expensive when compared to say, a GTX460. Am I better off getting rid of the 285 for 2x460's or mix the 285 with a 460, or even use the 285 with a 5870?

 

[banderon]

I have a question as far as SLI and physx. I'm thinking of getting two 460s, which means that I'll only be able to have the two of them bridged since the 460s come with only one bridge connector. If I throw in a 260, intending it to do just the physx work, will my setup's potential speeds suffer from not having all three cards bridged? Or is it only important for the cards that are doing the graphics rendering to be bridged, with the physx card just having to be present in the system?

 

[liquidsnake718]

Why not also compare this to X58 crossfire and SLI configs? if its just P55 its for a segment albeit large but not all... even those who have the x48?'s or Nforce680 boards?

 

[tehramen]

GTX 280 performance in Alien vs Predator goes down as the CPU clock speeds increase. Wtf?

 

[daedalus685]

You guys need to clarify the mixed crossfire section. The cards do NOT 'slow down' to match the weakest card in crossfire. Crossfire works asynchronously (unlike SLI) and all cards will run at whatever clocks you give them. There will be no case when the cards will attempt to clock up or down on their own.

Additionally, there have been tests lately that show 5770+5750 being proportionally faster than 5750+5750.... why on earth would you not put 5850CF in the chart for comparisons sake??? The 4000 series would always perform like a CF of the slowest two, even though it is asynchronous and the possibility was there for improvement there were no improvements likely on the driver level. The faster card simply does a lot more waiting than the slower, but it is important to note that it is not slowing down.. in a game that may not use CF you still get the full speed of the faster card.

Reviews like this:
http://www.xbitlabs.com/articles/video/display/asus_eah5770-powercolor_pcs_hd5750_8.html#sect1

Show the idea but it really needs to be investigated further.. is it new with the 10.x drivers? Is it just the 5700s? We really need to see more detail.. and the charts you provide woudl be the perfect place to find some if you simply included a few normal crossfire runs for comparison.

At any rate, please be much clearer when you write about something as technically complicated as crossfire/sli... the amount of misinformation does not need to be added to. Enough folks seems to think crossfire down clocks like SLI already, more don't need to be added to the list. (though something should be said for the scaling bonus that SLI gets perhaps due to the perfect matching it forces).

 

[eddieroolz]

Nice article! I learned quite a few things from this.

 

[jobs012]

Great article!! clears a lot of confusion on the x8 & x16 pcie lanes

 

[malnute]

Again "TH" leaves the obvious comparison out, why didn't they compare these cards on a board with dual 16x pcie?

 

[makwy2]

This article could not have been published at a better time! Just ordered a second gpu to CrossFire! I will read and re-read this a couple times while the item is shipped to me.

 

[williehmmm]

Torbee - a GTX285 will be unable to sli with a GTX460. With Sli both/all cards need to be the same chip, 285 with 285, 460 with 460.

Banderon - there is a mention in another article that having a third card outwith the sli configuration for dedicated physx, this gave no performance increase at all.

 

[Ubrales]

Good article!

 

[rubycube]

2x HIS ice-Q HD4670 cards are supposed to work in cross-fire, but the cards purchased don't have bridge connectors. When installed on the motherboard they didn't work together in the "mainstream" configuration either. HIS seem unable to help, so watch out if you're intending this configuration. Any suggestions as to how to get this setup to work would be much appreciated. (motherboard: Foxconn P45 intel chipset, CPU: C2D E6750)

 

[Notleh]

Great article. I knoew graphics had been pushed more and more to the gpu, but not as much as the tests show. The i5 went up a couple of notches for gamers.

 

[Alvin Smith]

I'm still trying to figure out how/if SLI affects running four monitors from two GPU cards (for NLE/Editing) ... How will adding the bridge affect config and (render) performance when running 3 or four monitors from TWO ZOTAC GT240 cards, for instance, running popular pro-sumer NLE Edit Suites ?

= Hmmmm .... =

 

[OvrClkr]

yea I agree, the 5850's should have been tested on an 890FX board (x16/x16). Not sure what happened there..

They should have tested the NV cards with the P55 and the AMD cards with the AM3 just to get a better idea of the difference between gaming with an AM3 CPU vs. an Intel CPU when CF and SLI are considered.