Motherboards All you need to know about running multiple GPU's in x16 x8 x4 slots and the outcome.

OK results are in. I've been benching all day, and run each test at least 8 times to achieve consistency, and I'll post up the fastest 2 or 3 of each combination.

Heaven settings used for all tests.




PC clocks and details.




Single GTX580 in 8x

Run 1


Run 2



Single GTX580 in 16x

Run 1



Run 2




2 GTX580 SLI 16x 8x

Run 1



Run 2



Run 3




2 GTX580 SLI 16x 16x

Run 1



Run 2



Run 3




I was expecting the SLI difference between 16x 16x and 16x 8x to be wider than it was for 1080p, but it seems the higher the res, the less impact.

 

yeah that's the way it works.. see it's always been like that

the cpu can feed only so much data to the gpu.. that's why you see at lower resolutions this big gap in performance between computers with different cpu speeds

the gpu can output as fast as the cpu can feed at say 1024x768.. so your bottleneck at that res will be the cpu and bus speed

it's why when you see comparisons at low res gaming.. they don't mean anything cause most don't game at those resolutions and the gpu is able to output more frames than the monitor usually.. the higher res is where the gpu starts to become the choke

the cpu can easily feed more data than the gpu can output at this point.. it's why 8x and 16x argument shows little difference.. once the gpu is spitting out as many frames as it can and the cpu can keep up- your seeing the best performance you can see with the setup

you see it in the video.. I don't think the guy in the video actually knows what's going on though- he's talking like the 4x lane means it's feeding the 6990 the bandwidth of pcie 2.0 2x to each gpu..

if you look at what's actually happening.. you see the cpu is still able to give the gpu enough data to render what it can at that high resolution even in a 4x lane..

 

i get confused here...because in the other thread you said this as well, but he doesn't lose fps, he gains, from 16.something to 22.something...his score increases as well...

 

For the record, I have gone back to single card in my system. I was getting less max fps with crossfire in the Benchmark than single card, and games seem stuttery and worse. Dirt3 gives more FPS in the bonnet cam, but inside goes down to 20fps, but 55-60 with single card.

Gash

 

Something's not right there. I get over 95% scaling with my cards, but that's SLI not Xfire.. but even so... clearly wrong that. You can't call dual card systems gash because you have a problem

 

there are lots of things that could limit your max fps...but that number is useless...what is your average fps? what is your minimum fps? as long as your xfired min fps is >= single min fps and xfired average fps 1.5-2.0 x single average fps...it's win...

i think you and true_gamer got off a little and focused on the wrong number...

as far as the stuttery game play, some people have those issues and some don't...havent done much research on it, but google it...sure there is advice on what the cause may be...

 

Single card avg fps = 33
SLI avg fps = 58

almost twice the frame rate. I wasn't looking at max frame rate.. was going off avg.

 

Check the other thread, I posted the benchmark. Min was 9 with single card, 10 with both. Max was 40 with 1, 30 with 2?

edit: http://www.apple-source.co.uk/junk/crossfire-bench-16x.jpg

 

are you true gamer? lol...i wasn't talking about your post brother...it was a reply to alex and true...DAMNIT POOKEY!

 

Oh and I wasn't calling dual card systems gash - just my system. My plan is to buy a 2nd 580 and go SLI with those instead

 

Funny there was no mention of x8 x8 which is the only option for 1155 or 1156 chipsets without NF200.

I run x8 x8.

 

Am I the only one here running crossfirex at a high resolution?

 

As feathers says, (pre-Gen 3 PCI-E) multi-PCI-E slot 1155 or 1156 chipsets can only do x8 x8 without an additional chip.

Even the Asus Maximus Iv Extreme...

Source: http://www.asus.com/Motherboards/Intel_Socket_1155/Maximus_IV_Extreme/#specifications

 

i'm glad i didn't sell my x58 now

 

GEN 3 PCIe.

 

Very true, clarified my earlier post.

 

looking at some articles, tends to lead me to believe the micro stutter issue is partially caused by cards not clocked evenly...more vram on one card...differing vga bios'...

 

I'm not sure if it is relevant but I used to run 2 identical 5770's (the serial numbers were even very close) in CrossfireX (at 1680x1050) with Dirt 2 on ultra settings. I got a small micro-stutter that was annoying.

There was one post processing setting within the config that I turned down to high and it cured it. I can't remember the name of the setting, if I remember I will check later tonight.

 

what fps were you getting? making some assumptions, that kinda makes sense...being that it was a more budget-oriented card...certain post processing settings might hit a setup like that harder than a higher end gpu...maybe it hits the memory bandwidth harder or the card just doesn't have enough grunt to get through certain sections of the game without a decent drop in min fps. and the fact it cleared up by lowering a setting to high...interesting stuff regardless, if you think about it from that angle.

 

Over 75 FPS: http://forums.bit-tech.net/showpost.php?p=2584636&postcount=3

I recalled this article, it is worth the read: http://www.tomshardware.com/reviews/radeon-geforce-stutter-crossfire,2995.html.