PCI-Express x16/x4 CrossFireX/SLI Scaling

[Lmeow]

I'm wondering really, how much worse is the x4 when you are running two cards in CrossFireX or SLI?

On techPowerUp! the diffference between x16 and x4 wasn't major for single card setups (I'm referring to the GTX 470 and GTX 480, however on tom's, whether it be poor three way scaling or something else, a review with an i7 870 with HD 5870s? I believe or something else, with the lanes at x8/x8/x4 was worse than x8/x8.

I'm thinking of a GTX 470 later this year when I rake in some $, but wanted to know whether it would be alright possibly later in SLI with another GTX 470 in the x4 slot. (I have the EVGA X58 SLI LE, dont' really want two GTX 470s right next to each other)

Any extra evidence would be great.

 

[jyjjy]

I thought to be certified as SLI the second slot had to be at least x8? I'm pretty sure I remember reading that somewhere.

 

[Lmeow]

Hmm you might be right, although I heard one person ran GTX 480s in x16/x4 on a board like mine, he got 35000 in some 3DMark bench (not sure which one specficially) and asked whether it was good or not, like whether that was what he should be getting.

 

[JabbaWockeeZ]

yeah i have GIGABYE GA-MA785G-UD3H mb and this topic is very interesting for me too, bc i want to SLI 2 GTX 460`s one in pciex-x16 and other is pciex-x4. Will it work ok?

 

[Griffolion]

Whether the SLI certification issue is true or not, its highly unrecommended to do a x16 x4 SLI/Crossfire setup as the bottleneck in bandwidth is crippling. if you have a single GPU in the x16 slot and are looking to upgrade, with a x16 x4 motherboard its best to sell your existing one and get a more powerful single one.

 

[Mousemonkey]

yeah i have GIGABYE GA-MA785G-UD3H mb and this topic is very interesting for me too, bc i want to SLI 2 GTX 460`s one in pciex-x16 and other is pciex-x4. Will it work ok?

Do the SLi hacks work on an AMD crossfire chipset?

 

[Lmeow]

^ I think it works on some AMD chipsets.

How come a card by itself is alright at x4, but in CFX/SLI mode, x16/x4 seems to show bandwidth issues, does it?

 

[Maziar]

I'm wondering really, how much worse is the x4 when you are running two cards in CrossFireX or SLI?

On techPowerUp! the diffference between x16 and x4 wasn't major for single card setups (I'm referring to the GTX 470 and GTX 480, however on tom's, whether it be poor three way scaling or something else, a review with an i7 870 with HD 5870s? I believe or something else, with the lanes at x8/x8/x4 was worse than x8/x8.

I'm thinking of a GTX 470 later this year when I rake in some $, but wanted to know whether it would be alright possibly later in SLI with another GTX 470 in the x4 slot. (I have the EVGA X58 SLI LE, dont' really want two GTX 470s right next to each other)

Any extra evidence would be great.

Tom's recently wrote a great review comparing x16x4 with dual x8/x16
http://www.tomshardware.com/reviews/pcie-geforce-gtx-480-x16-x8-x4,2696.html
Its not recommended to run 2 cards in SLI/CF mode at x16x4 speeds,because it doesn't perform as well as dual x8/x16 in almost all games and the performance gap is quite significant.

 

[jyjjy]

Tom's recently wrote a great review comparing x16x4 with dual x8/x16
http://www.tomshardware.com/reviews/pcie-geforce-gtx-480-x16-x8-x4,2696.html
Its not recommended to run 2 cards in SLI/CF mode at x16x4 speeds,because it doesn't perform as well as dual x8/x16 in almost all games and the performance gap is quite significant.

That article actually has no testing of 16x/4x, just dual x16 vs dual x8. It does test a single card at x4 and overall it lost 20% as a result but it looks like it is very specific to which game as 3 of the 5 they tested showed almost to difference. Honestly I don't know what to conclude from that article about a possible 16x/4x GTX 470 setup.

 

[Maziar]

Yes you are right, it only compare SLI/CF on dual x8/x16 mode.
Here is an old but good review which compares them all(It doesn't include the GTX 470 but it should give you the idea)
http://www.tomshardware.com/reviews/crossfire-pci-express,2095.html

 

[jyjjy]

Yeah, I've seen that one too. Unfortunately none of the boards they tested are actually both PCIE 2.0 and x16/x4. I know PCIE 1.0/1.1 at x16/x4 was not good but 2.0 doubles the bandwidth and I haven't seen any articles that actually test sli/crossfire on such a set up.
Here is the summary of another article about PCIE scaling but (again) it only shows x4 for a single card;
http://www.techpowerup.com/reviews/AMD/HD_5870_PCI-Express_Scaling/25.html
They do much more extensive battery of benchmarks however. It seems that for a single HD5870 you only lose 5% performance on average. From that one might assume that the penalty of using a 2.0 x4 slot in crossfire is very small as well. I'm not certain if that is the case but they seem to think so in the conclusion of that article.
The only actual testing I've seen is from someone on this board asking if he should bother installing a second HD4870 he got from a friend considering his second slot was 2.0 x4. As he had the card I suggested he install it and run some benchmarks. The results were right in line with an HD4870x2 but the HD4870 needs less data than something like a GTX 470.

 

[Maziar]

In some games,the difference between x16x4 and dual x8/x16 may not be huge,but in some title especially the ones which are GPU-limited like Crysis, the performance is quite noticeable.
Anyway i haven't found any new articles comparing x16x4 with other CrossFire modes, I'll report back if i find any.

 

[jyjjy]

In some games,the difference between x16x4 and dual x8/x16 may not be huge,but in some title especially the ones which are GPU-limited like Crysis, the performance is quite noticeable.

Have you seen results for Crysis with 2.0 slots?
That techpowerup article shows only a 5-10% loss with a x4 slot on a single card in Crysis. Not optimal but far from something that would make me decide not to use crossfire/SLI on a 2.0 x4 board.

 

[Maziar]

True but that's tested in single mode and not CrossFire.
But it seems that PCI-E 2.0 has improved the performance of x4 speed compared to PCI-E 1.0

 

[Lmeow]

I've finally found out the reason for why the x8/x8/x4 scaling was incredibly poor on the Asus Maximus III Formula PCI-E scaling review done by tom's.

The x4 slot is not a PCI-E 2.0, it's either PCI-E 1.0 or 1.1 which means it would have the bandwidth of a PCI-E 2.0 x2 lane, I know it doesn't exist but you get the idea, and although PCI-E 2.0 x4 didn't bottleneck the data too much, PCI-E 2.0 x2 must have.