PCI Express And SLI Scaling: How Many Lanes Do You Need?

There are quite a few folk who have been waiting for this.

This would even add more value to the i5-750 in your best CPU for the money article.

I love how people always bash on x8 x8 and how it sucks, when in reality x16 x16 is only 4% better.

You spend unnecessary $$$ on a x58 platform while I save money that I can put towards a GPU upgrade with my p55 platform

i'm curious how other games are "dependent" on bandwidth while others are not... does that mean that the games that aren't dependent on bandwidth isn't using the full potential given the the size advantage of x16 over the x8? i wish every game would utilize every inch of your hardware in the future.

For all your fps hunters, I bet you only have a screen at home with 50 or 60Hz. So just for your information, everything above 50 or 60fps is just useless... In this aspect, a PCIx 4x is actually enough... ;-)

so a 4850 crossfire setup will hardly be bottlenecked by an 8 lane motherboard.

anyone know if 4850's are going to be unavailable any time soon? You could get the 3000 series for quite awhile after the 4000's released so I'm crossing my fingers until i can afford a cpu upgrade and another 4850

cpu is currently a 7750BE and so im pretty sure it would bottleneck the 4850's. I think it does with just one actually.

@ carlhenry

It's not the game's fault. The GPU can only go as fast as it was made to go. So in simple terms you could say that GPUs these days aren't "fast" enough to use all the bandwidth PCI Express offers.

Very nice review but I have to ask, why did you not test with 5970's?

On a card for card basis they are still quite a bit more powerful than the GTX 480 and should require the most bandwidth of any current card for maximum performance.

This review tells me that if you already have an i7 and at least 2xPCIe 16x lanes on your motherboard then in order to play modern games, all you are going to be upgrading for the next several years is your graphics.

Even though I'm not interested in SLI I am glad to finally see a benchmark comparing PCI-E x16 and x4. My motherboard only allows me to use my video card in my x4 slot for some reason. I've been wondering how much performance I've been losing over that.

Wow. I wouldve swore that they already did an article on this. Geez times are rough over at toms arent they?

I'm not an enthusiast, so a single powerful card would be more than enough for me...

outlw6669Very nice review but I have to ask, why did you not test with 5970's?On a card for card basis they are still quite a bit more powerful than the GTX 480 and should require the most bandwidth of any current card for maximum performance.

The first article tested CrossFire scaling with three 5870's:
http://www.tomshardware.com/reviews/p55-pci-express-scaling,2517-2.html
1.) It appears that the GTX 480 runs out of CPU faster than the HD 5870. 2.) It also appears that the biggest difference between games is how hard they hammer the GPU, based on details, lighting effect, etc.
3.) The result is that you're seeing an FPS cap from either the board or the CPU as the load shifts away from the GPU to other components. The good news is that this "cap" is higher than the "minimum playable" frame rate most people can tolerate, in most games.
So, what does this have to do with your question? The HD 5970 uses a PLX Bridge: http://www.tomshardware.com/reviews/radeon-hd-5970,2474-2.html

Both GPUs get the same data, and the PLX Bridge simply doubles it from one set of lanes to two GPUs. So, an x16 slot turns into two identical x16 sets, or an x8 slot turns onto two identical x8 pathways. The PCIe "bottlenecking" data you get for two 5870's should therefore be identical to the PCIe data you get from one HD 5870 x2, such as the Asus ARES, which is actually a faster card than the HD 5970.

Ah, that does make sense now.
Thank you for the explanation Crashman

would the same pcie scaling performance differences still be there with a slower CPU?

I'd be interested in seeing more games tested. In this test, only CoD seemed to show any noteworthy decrease in performance for a single VGA card in the 4x PCIe slot and even that at FPS rates where it didn't matter.
I'm considering using the 4x on my mainboard for airflow reasons in my case and as of now, nothing seems to speek against it. I'm only using a lowly GTX 460 anyway.

I would have liked to see what would happen if the i7 wasn't overclocked.

iam2thecrowewould the same pcie scaling performance differences still be there with a slower CPU?

The slower your CPU, the more the limit shifts from other components to the CPU. That means the maximum FPS will get dragged down even farther, making the 1680x1050 results look closer to the 1920x1200 results.

th_atI'd be interested in seeing more games tested. In this test, only CoD seemed to show any noteworthy decrease in performance for a single VGA card in the 4x PCIe slot and even that at FPS rates where it didn't matter. I'm considering using the 4x on my mainboard for airflow reasons in my case and as of now, nothing seems to speek against it. I'm only using a lowly GTX 460 anyway.

Is it a PCIe 2.0 slot? Please read the CrossFire article to see how bad PCIe 1.1 x4 is, and don't use it.

Very educational article. I usually recommend against the X58 platform when I'm asked for help, and seems like this article validates my recommendations.

Seems that by the time PCIe 2.0 x16 is needed we will be using PCIe 3.0 anyway. This is also good in the sense that not you can look for x8 - x8 boards with other features such as USB3 and SATA6GB/s rather than having them gimped in favour of x16 - x16. I can guess the general result of this but how much performance drop is there using one of the p55 x16 - x4 links? Im assuming the article was using both cards at x4 -x4?