Pcie 2.0 3-way sli/crossfire vs pcie 3.0 2-way sli

I'm looking to purchase a new motherboard and with the emergence of pcie 3.0, I'm wondering if I would see better performance with a 2-way sli/crossfire on a pcie 3.0 board or a pcie 2.0 board with 3-way sli/4-way crossfire. The only thing set in stone is that I have a socket 1155 cpu. I could use either nvidia or amd but chances are it will be in the $300 range. Ideally I would like to be able to drop in another video card as games get more demanding. Having the option to put in an ivy bridge cpu later on is a huge bonus IF they come in 1155. Any insight would be appreciated.

Reply to Hatesu

more cards is better, the slots dont make that much difference.

Reply to farrengottu

PCI-E 3.0 isn't going to make much of a difference by now unless your running some kind of GPU computing app that can take advantage of the extra bandwidth. 3-way sli or crossfire would be the way to go.

If you have any documentation or technical data to back that up it would be appreciated and ty for the response.

Reply to Hatesu

PCIe is not a bottleneck (unless less than PCIe 2.0 8x lane)

http://www.tomshardware.com/review [...] ,2696.html

And I don't think I have to provide support for more card in crossfire/SLI equal better performance.

------------------------------ Constant vigilant! Keep spammers away from Tom's Hardware forum.
Reply to Pyree

Also you will need IB or SBE (socket 2011 and later IBE) to have pcie 3.0.

Reply to k1114

I know that the sandy bridge cant take advantage of pci gen 3 but I've read that they will be making ivy bridge core i7's that are socket 1155. Is that gonna matter? Would a six-core system with two gpu's be better than a quad core with 3-4 gpu's.

Reply to Hatesu

For Microsoft Flight probably go with a hex core, but for the rest, go with quad core with more GPU. By the way, what is your screen setup that require tri/quad SLI/crossfire?

Message edited by Pyree on 01-12-2012 at 02:39:11 AM

Reply to Pyree

I haven't picked up any monitors as of yet but 3 would be ideal. Deus ex is capable of five, but that would be some serious cash. I'm really just curious about some of the sli setups and if gen 3 pci would make much diff with a ivy bridge and 2 cards compared to sandy bridge and 3-4 cards. Even the above article mentions the cpu being a issue with that particular setup.

Reply to Hatesu

Here's pcie scaling with sb. http://www.tomshardware.com/review [...] ,2910.html Pcie2 is fine and SB is plenty powerful so IB and pcie3 isn't necessary. If you really want IB I would suggest you wait til april and also grab z77 chipset.

Reply to k1114

Appreciate the input. Think I'll stick with 2.0 for now. Thx again.

Reply to Hatesu

1000 Words:

It's all about your GPU and PCIe 3.0 won't benefit anyone Gaming for quite some time...

Message quoted 1 times

Reply to jaquith

jaquith wrote :

It's all about your GPU and PCIe 3.0 won't benefit anyone Gaming for quite some time...

That's not entirely true.

http://hardocp.com/article/2010/08 [...] _vs_x8x8/2

First he's talking about using x8 lanes on 1155. You're comparing x16 lanes which nobody has ever thought were saturated.

That article is over a year old using 2x480 in SLI and you can see x8 bottlenecks at multi-monitor resolution. Think about how much more is created when using 2x 7970.

There is a place for PCI-e 3.0 and moreso the 40 lanes of pci-e offered by NF200/1155 and 2011 chipsets. If you're considering a multi-monitor solution the answer is far from clearcut.

------------------------------ CM Storm Scout || Gigabyte D3H || i5-2500K @ 4.55 || 8GB Patriot Sector 5 @ 1600 || Sapphire 6950 2GB OC || OCZ Vertex 3 SRT || Seagate 1.5 Green || Silverstone 700w || CM 212+ || Dell 2209wa
Reply to a4mula

I hate to debate folks, I'm not going to post Troll info. In HIGH resolutions the problem is not the x8 or x16 lanes -- it's a vRAM Bottleneck (1GB vs 2GB vs 3GB).

Example of x4 vs x8 vs x16; note the x4 suggests the x4 is barely saturated let a lone x8/x16:

±1FPS~±3FPS is well within the margin of error, and if you ran the benches several times you'll never have the same FPS or 3DMark11 or any benchmark perfectly match-up. Folks make mountains out of molehills and mountains into pebbles to suit their agenda.

Example of two of the SAME GPU's but different vRAM (1.5GB vs 3GB); little squiggle on the very bottom:

ref - http://www.tomshardware.com/forum/ [...] 5#t2011085 note the HARDOCP link.

I build 4-WAY rigs all of the time, and the only thing I care about is saturation and lanes. Yeah, I prefer LGA 1366 or LGA 2011. However, the LGA 1155 is perfectly fine when equipped with the NF200 *where nVidia* provides the BIOS Keys for their GPUs to run 2~3-WAY. AMD/ATI allows 4-WAY CF on the LGA 1155. The 3-WAY vs 4-WAY is nVidia's (BIOS Key) control thing more than it's the lanes themselves. AMD/ATI is open and works just fine if you like their GPUs...I don't. The LGA 1155 could, if nVidia allowed it to, run 4-WAY SLI on LGA 1155 and IMO with very little overall impact.

Message quoted 1 times
Message edited by jaquith on 01-13-2012 at 12:42:34 AM

Reply to jaquith

jaquith wrote :

I hate to debate folks, I'm not going to post Troll info.

Nobody here denied that VRAM plays a part of large resolution gaming. Staying on track is if 1155s x8 lanes would cause bottleneck issues at large resolutions.

The Hardocp article is using the same exact video card in both 2560x1600 and 5760x1200. The only difference is x16 vs x8 lanes. While there was no noticeable effect at 2560x1600 there was visible bottlenecking while using 3 monitors.

This was also done with a video card that is performing at a fraction of what the 7970 is capable of.

Now if you can show me 7970 crossfire data at eyefinity/surround resolutions that compares 1155 x8 lanes to 2011 (or even 1366) x16 lanes then I'm more than open to see it. This is something I've been dying to see as not only is my current rig running Eyefinity, but my current build is revolved around it as well.

Message quoted 1 times
Message edited by a4mula on 01-13-2012 at 12:43:20 AM

------------------------------ CM Storm Scout || Gigabyte D3H || i5-2500K @ 4.55 || 8GB Patriot Sector 5 @ 1600 || Sapphire 6950 2GB OC || OCZ Vertex 3 SRT || Seagate 1.5 Green || Silverstone 700w || CM 212+ || Dell 2209wa
Reply to a4mula

a4mula wrote :

Nobody here denied that VRAM plays a part of large resolution gaming. Staying on track is if 1155s x8 lanes would cause bottleneck issues at large resolutions.

The Hardocp article is using the same exact video card in both 2560x1600 and 5760x1200. The only difference is x16 vs x8 lanes. While there was no noticeable effect at 2560x1600 there was visible bottlenecking while using 3 monitors.

You are adding 1 + 1 = 3; the HARDOCP with the GTX 480 is 100% vRAM bottleneck and 0% PCIe issue. The GTX with 1.5GB of vRAM is ill equipped and ditto with any 1.5GB GPU to handle multiple monitors in many games. I have 3-WAY GTX 470's and my resolution is 5900x1080 (bezel correction), and I started running into major drop-off's with several games; 100% vRAM issue. My plan is to keep my i7-980X and simply once available replace them with GTX 600's and not even a thought of PCIe 2.x vs PCIe 3.x.

a4mula wrote :

Now if you can show me 7970 crossfire data at eyefinity/surround resolutions that compares 1155 x8 lanes to 2011 (or even 1366) x16 lanes then I'm more than open to see it.

Take your energy to Google and look. I'm sure the HD 7970 only improvement is the GPU cores themselves and it won't even touch x8 PCIe 2.x. The saturation is as much a function of the sheer data from the Game. Nothing out there is sending 8 X 500MB/s of data!

Message quoted 1 times
Message edited by jaquith on 01-13-2012 at 01:06:20 AM

Reply to jaquith

jaquith wrote :

You are adding 1 + 1 = 3; the HARDOCP with the GTX 480 is 100% vRAM bottleneck and 0% PCIe issue. The GTX with 1.5GB of vRAM is ill equipped and ditto with any 1.5GB GPU to handle multiple monitors in many games. I have 3-WAY GTX 470's and my resolution is 5900x1080 (bezel correction), and I started running into major drop-off's with several games; 100% vRAM issue. My plan is to keep my i7-980X and simply once available replace them with GTX 600's and not even a thought of PCIe 2.x vs PCIe 3.x.

Have you even read the article? I'll state once again, the only difference in the 5760x1200 testing was 1156 vs 1366. The VRAM of the 480s didn't somehow shrink or grow. This was a direct comparison of x8 lanes versus x16 pci-e 2.0 lanes. I'm hard pressed to understand how you seem to think that the VRAM is going to play any role in an apples to apples comparison. If VRAM was the culprit then you would have seen identical results between x8 and x16!

Quote :

Take your energy to Google and look. I'm sure the HD 7970 only improvement is the GPU cores themselves and it won't even touch x8 PCIe 2.x. The saturation is as much a function of the sheer data from the Game. Nothing out there is sending 8 X 500MB/s of data!

Watch, I can do that too. Nothing out there is sending 4 X 500MB/s of data! Yet we know x4 lanes are clearly saturated by even modest gpus. So at what point do the x8 lanes become saturated? You can find tons of examples of x8 lanes seeing degraded performance, usually it's a small amount ~1-3% but these examples are typically done at 1920x1080 or even worse 1680x1050. What happens to that ~1-3% when you multiple the pixel count by 4? What happens when you increase the bandwidth of the gpu substantially?

I have looked, I'm actively looking for a review that samples the 7970 crossfired on both 1155 and 2011 at multi-monitor resolutions. It's my hope that there isn't a major bottleneck as it means I can move forward with my build on Z68. Until that review happens however, not you, not I, not anyone knows.

I've stated all I'm going to state on the issue. There can be no argument that x8 lanes do impact performance. The question remains, how much that impact means at large resolutions with the latest gpus. Personally I look forward to finding out.

Message quoted 1 times

------------------------------ CM Storm Scout || Gigabyte D3H || i5-2500K @ 4.55 || 8GB Patriot Sector 5 @ 1600 || Sapphire 6950 2GB OC || OCZ Vertex 3 SRT || Seagate 1.5 Green || Silverstone 700w || CM 212+ || Dell 2209wa
Reply to a4mula

Ironically enough I found that hard ocp site last night and although thier new test rig uses a diff asus board, the one I was considering is also an asus board (Maximus IV Extreme)

The whole reason I asked this question was because originally I had picked an MSI board (z68a-gd80). The Msi board was pcie 3.0 but I didn't realize it didn't support 3-way sli. Soooo here we are now, I have this i7 2600k and I want to carefully consider my remaining options before moving further. Say what you want about msi or asus but I have always had great success with thier boards in the past. When others were having issues with devices and stability, my machine was running flawless.

While I agree that cpu, lanes and gpu ram all have somewhat of an effect on the outcome, the real question is how much and can I make use of it any time soon.

Reply to Hatesu

I should add that even though I sent the msi board back, there is no gaurentee that they will accept the return. If thats the case and do get a pcie 2.0 board then I will be sure to post the results.

Reply to Hatesu

a4mula wrote :

Have you even read the article?

Nope, I glanced at it now and yesterday. I can point-out countless Articles with poorly contrived or misleading conclusions; you need to see overlapping information before assuming a conclusion; Theory vs Reality. I noted a methodology of 'tapping' which makes me wonder unintended happenstance. I do often see a +1~+3 FPS going from x8/x8 to x16/x16, I have my 'theories' but I know it's not a saturation issue. One of my 'theories' is how drivers look at the number of available lanes.

There's a lot of variables that affect frame rates including: CPU, RAM, drivers, methodology in testing, chipet vs direct path, MOBO design, drives -- you name it.

Here's a couple things to look and glance at:

CPU bottlenecking -> http://i1013.photobucket.com/album [...] ecking.jpg
LGA 1366 vs LGA 1155 and NF200 (plenty of benchmarks w/graphs) - http://www.tomshardware.com/review [...] ,2910.html

--

The current heated debate is over the SB-E CPU; as it turns out the older reviews on the SB-E might be wrong, Intel quietly redid it's lithography and the 'Reviews' used: 1st engineering CPU's, then the C1 (bad/retail/1st release), and now we need to see the C2 (re-released as of Jan 20th SB-E). Reminiscent of B2 recall.

Reply to jaquith

Forum Motherboards & Memory : Motherboards Pcie 2.0 3-way sli/crossfire vs pcie 3.0 2-way sli

Motherboards

Memory

Chipsets & Bios

Abit

Aopen

ASRock

Asus

Chaintech

Corsair

DFI

Elitegroup

Epox

FIC

Foxconn

G.SKILL

Gigabyte

Jetway

MSI-Microstar

Shuttle

Soyo

Supermicro

Tyan