16x 4x crossfire with suprising results

I

Idef1x

Limp Gawd
Joined
May 2, 2011
Messages
227
This is my first post in here, but I just had something I wanted to share.

Recently I bought my second Asus EAH6950 to crossfire on my Asus P6T. To my disappointment, the two PCI express ports are placed too close together, so the upper card would have a hard time getting ventilated.

I found two crossfire bridges of 100 mm, and placed the second 6950 in the third slot on the mobo. To be even further disappointed, I saw the last slot was a x16 but only with x4 speed.

I ran a 3dmark11 on the system and scored 8336 or something like that. I then tried to place the second 6950 in the second slot, to reach x16 speed, and ran the test again. It scored 8396! Those two scored are virtually the same in my book.

So a P6T crossfire x16 x4 with two Asus EAH6950 is roughly equal to the x16 x16 crossfire.

Any thoughts on this?
 
Not really sure why this is surprising? 16x 16x vs 16x 4x or 8x really is a minor difference and has to have a really hard pushing application to see statistical differences.

Sounds like a win win situation for you.
 
g-money said:
Not really sure why this is surprising? 16x 16x vs 16x 4x or 8x really is a minor difference and has to have a really hard pushing application to see statistical differences.

Sounds like a win win situation for you.
Click to expand...

Yeah, but I didn't have any experience with crossfire, so never really gave much thought to pci speeds, since it was never an issue. But you are completely right.

But yeah, I was just happy the difference between the setups was so minor :D just wanted to share if anyone else had the same complications as me.
 
A 4x slot card will incur 10% performance hit at 1080p and a 1x slot card will incur 30% hit, IIRC.
 
Raap said:
Doesn't this difference increase slightly with resolution?
Click to expand...

Yes it does. OP, run any tests or benchmarks that you run at native res. or as high as possible. Check out the difference there too. In [H] Surround evaluation, 8x became a bottleneck of around 10% at 5760 x 1200, but it's usually nothing major with a single display.
 
Yeah, maybe I should get a full version and do a test on the full resolution. I will try that tomorrow. I am gaming on a 27" 2560x1440.
 
How is the actual performance while gaming? Do you notice any strange stutter or anything?
 
Yeah it's been a long time that the PCI-E bus has been sitting around waiting to be taxed by anything, and for the most part nothing has managed to live up to doing so. Yet people still pay mad premiums for 16x/16x or 16x/16x/16x,:rolleyes:

The bright side is happy little accidents for people like OP. Though the unhappy accident of squashing the two x16 slots together is something I am confounded by quite often when looking at mobos.
 
I run 2X GTX 480's on my P6T motherboard as well. I've tried all the combinations and found that the performance does not make a difference even when SLI X16 and X4. I'm running a 30" Dell at 2560X1600. Benchmarks are virtually the same. This whole X16 buisness is just to sell you on something you don't really need.
 
Try running both gpu in x4 then :p
J/k, however it seems to matter at eyefinity surround :(. Oh well.
Posted via Mobile Device
 
Porter_ said:
obligatory link to [H] PCIe Bandwidth Performance Article
Click to expand...

That seems to cover it pretty well. It shown very little performance loss.

BababooeyHTJ said:
How is the actual performance while gaming? Do you notice any strange stutter or anything
Click to expand...

Nope, no stuttering, aritfacts or anything while gaming. It seems to perform quite alright.

Michael Turbo said:
What resolution are you gaming at?
Click to expand...
I'm gaming at 2560x1440, but the benchmarks I ran was only with the standard version of 3dmark11. I will perhaps get the full version to do a benchmark at full resolution. But as hardocp concludes in the article posted by Porter_, it doesn't have much impact before very large resolutions - like. 5760 x 1200
 
In related news, I tried Crossfire on a Intel P35 motherboard with 16x/4x PCI Express 1.0 ports and the results were dreadful. Everything I tried except for the Unigine benchmark was slower with Crossfire enabled :p
 
Well if I do 16x/16x crossfire, my two cards get really close together.

Is it anything to be concerned with? Mind you that EAH6950 only have the air-intake on the side, and not in the rear end. Or do you think the cards will manage fine?


NathanP2007 said:
http://www.youtube.com/watch?v=vfXALgE7mVM&feature=channel_video_title

From this video it seems real world application (video games) show a very real FPS difference between 4x and 16/8x when in CF.
Click to expand...

That test doesn't show anything about 16x/4x - since he uses a 6990. Please correct me if I'm mistaken.
 
Last edited:
Idef1x said:
Well if I do 16x/16x crossfire, my two cards get really close together.

Is it anything to be concerned with? Mind you that EAH6950 only have the air-intake on the side, and not in the rear end. Or do you think the cards will manage fine?




That test doesn't show anything about 16x/4x - since he uses a 6990. Please correct me if I'm mistaken.
Click to expand...

From how i understand it, instead of using one 6970 in 16x and another one in 16x in CF, then finding a board and doing the same in 8x 8x, then finding another board that has two 4x PCI-E slots...he used a6990's which is a just CF'ed 6970s but on one card. He then put the 6990 in a 16x slot and did his testing. He then put it into a 8x and did his testing. And he then put it into a 4x and did his testing. The results being that when in 16x and 8x the 6990 (CFired 6970's) essentially got the same frames per second, but when in a 4x slot there was a significant drop in FPS. Why that video came specifically to mind is he did the exact same test and got the same result with a synthetic benchmark (3D mark). Which was there was basically no difference in performance when using his 6990 in 16x/8x/4x. He then switched over to real world application (games) because not only is a synthetic benchmark just that...synthetic, but he felt it didnt push the card the same way a real game does, and he was right, with the 16x vs 4x results being drastic. In BFBC2 the results being 120FPS vs 80FPS essentially.

Now, i do get that a 6990 is not identical to using two real 6970's in crossfire and testing them both at 16x then testing them both at 4x. However i think its as close as you will get considering i dont think any motherboards being manufactured right now come with two 4x PCI-E slots that are crossfire ready. I could be wrong.
 
hdgamer said:
The 6990 is 2 gpu's on one card. Of course that's gonna bottleneck a 4X PCI slot. But run one gpu in a 4X and one in a 16X and you won't see much difference.
Click to expand...

He is just trying to show the performance impact that a slower slot *could* have. Its not the same as a real 16x/4x test, however it highlight the fact that 4x slots can cause a noticeable slowdown with some configs.
 
This ancient article was actually very helpful today. Thank you OP and all responders. :)
 
I was going to give you [H]ell over necroing this thread, but I have to say it was a good read.
Wish I had known about it and the other article listed here, back when I could have used it in 2013, right before the current generation consoles were being released.
 
Right, that was a revelation with Sandy Bridge: the southbridge was upgraded to PCIe 2.0, which made inexpensive motherboards with CFX 16/4 (or 8/8/4) much more appealing. Previously, the x4 slots on the P55 and P45 boards were a disappointment!

It was a huge win for the constant Sandy versus x58 wars that would go on back in 2011, but this was the straw that finally broke the camel's back :D
 
defaultluser said:
Right, that was a revelation with Sandy Bridge: the southbridge was upgraded to PCIe 2.0, which made inexpensive motherboards with CFX 16/4 (or 8/8/4) much more appealing. Previously, the x4 slots on the P55 and P45 boards were a disappointment!

It was a huge win for the constant Sandy versus x58 wars that would go on back in 2011, but this was the straw that finally broke the camel's back :D
Click to expand...

If i understand it correctly they were unofficially pci-e 3.0. Well at least my 3960x is.

I had a problem with my R4E board originally with dual 7970-DCU2s. The top slot always ran at x4 and picked up to x8 on occasion which i could never understand why. At first i was gaming as usual until i saw it in gpuz while doing something or rather with it. RMAd the board. Have to say i never noticed any difference though as in performance problems.
 
No he is talking about the ports off the southbridge on regular Sandy Bridge, which are PCIe 2.0.

Processor
SandyBridge: 2.0 CPU / 2.0 SB
SandyBridge-E: 3.0 CPU / 2.0 SB (3.0 is unofficial, but works)
IvyBridge/Haswell: 3.0 CPU / 2.0 SB
IvyBridge-E/Haswell-E: 3.0 CPU / 2.0 SB

Skylake will finally bring PCIe 3.0 to the southbridge, and Skylake-E will bring PCIe 4.0 to the CPU lanes.
 
You must log in or register to reply here.
Back
Top