Massive FPS gains in rf2 using PCI-e 3.0 x16 with higher end cards!

Right, you meant other benchmarks and not the rf2 live benchmark but you didn't specify that and the flow of this entire conversation as been fixated on rf2 performance, hence the confusion. But even with other benchmarks, i think it's also the case that SLI on PCI-e 3.0 x16/x16 vs x8/x8 (or PCI-e 2.0 x16/x16) also has no difference in performance in benchmark programs such as 3d mark. Since we know that 3d mark and many other benchmarks do not fully utilise PCI-e 2.0 x16 (same as PCI-e 3.0 x8) then sticking two in SLI with the same bandwidth for each card will not become bottlenecked.

As spinelli said, this has been shown to not be the case with rf2 which benefits from SLI (as he claims when using stereoscopic 3d there is near perfect scaling) but that same performance hit from a single card using PCI-e 3.0 x8 (or PCI-e 2.0 x16) and not PCI-e 3.0 x16 translates to the SLI performance as each card is hit by the same bandwidth deficit.

Spinelli has PCI-e 3.0 x40 motherboard and cpu so he doesn't have this issue (if i'm not mistaken). So he can use 3 PCI-e 3.0 lanes at x16/x16/x8.

 

where is PCI-E 3.0 @ 16 x16x16 and PCI-E 3.0 @ 16 x8x8 I am not talking about PCI-E 2.0?

 

Ari, PCI-e 3.0 x8/x8 is equivalent to PCI-e 2.0 x16/x16 in PCI-e lane bandwidth. Spinelli even mentions it in the post he quoted himself in. E.g....

 

Actually, having said all that i think i may know what Ari may be trying to point out. PCI-e 2.0 cpu's such as the i5-2500k only support 16 PCI-e lanes (according to Intel's website) so does that means it can only support pci-e 2.0 x8/x8 effectively?

I don't think this affects your results however since you have 40 PCI-e 3.0 lanes which is effectively equivalent to 80 PCI-e 2.0 lanes.

 

2500K only has 2.0 16x, so in SLI it would be running 2.0 @ 8x and 8x, which is equivalent to 3.0 @ 4x and 4x. I still don't see how this backs his statement about rF2 only showing 1-2% fps gains - when running SLI - in the PCI-E tests so many of us have done.

 

I don`t think so, the bandwidth advantage for PCIe 3.0 would mean that an 8x lane of PCIe 3.0 would be the same as a 16x lane of PCIe 2.0, but hes cards being PCIe 2.0 can't take advantage of that.

 

This is very interesting. Those with older systems running at 2.0 x8/x8 effectively in rf2 when even 2.0 x16/x16 is a bottleneck.

Even more interesting is how many (if not all) mainstream games show no disadvantage to running at PCI-e 2.0 x8 even. rF2 is really PCI-e dependent. I'm going to ask something very ignorant perhaps but is this a consequence of running dx9 instead of dx10+?

Perhaps he just honestly missed it but i get you.

 

did anyone ever benchmark a single GTX 980 with triple 1080p screens + multi view? I am very curious - currently have a GTX 680 and get decent performance with triple 1080p screens + multi view and graphic details turned down somewhat, but want to upgrade as I want better visuals and need to be around 100fps...

 

Yes you made a good point there Ari however if someone have a PCI-e 3.0 cpu and motherboard, even with only 16 PCI-e lanes, they have more than enough for PCI-e 2.0 x16/x16. Because PCI-e 3.0 x16 is equivalent in bandwidth as PCI-e 2.0 x32 total lanes which can be used as PCI-e 2.0 x16/x16 at full bandwidth for each card.

 

Yes I know but by going SLI you don't double the number of lanes, you are just splitting them between two cards. When a board has x16 x16, it can actively give one of the cards more bandwidth if it is using more, where as the normal set up is just static x8 x8. So you really won't see much of a performance difference.

 

I don't really understand what your trying to say Ari. But you say there isn't much of a performance difference yet this has been shown categorically false with Spinelli's testing.

Having said that, i guess we'd need to get someone with a 16 lane PCI-e 2.0 only system with two cards in SLI to test your theory (at the x16 is alternated between the cards in sli) and using rf2 as the benchmark since almost all other games don't seem bottleneck with a single card in only PCI-e 2.0 x8.

So someone with and i5/i7-2000 series or lower cpu with PCI-e 2.0 x16/x16 support on their motherboard with high end SLI cards would be needed.

 

Please try to understand that this is not right:" PCI-e 3.0 x8/x8 is equivalent to PCI-e 2.0 x16/x16 in PCI-e lane bandwidth" because the bandwidth advantage for PCIe 3.0 would mean that an 8x lane of PCIe 3.0 would be the same as a 16x lane of PCIe 2.0, but cards being PCIe 2.0 can't take advantage of that.

What you need to test is PCI-e 3.0 x16@ x16x16 vs PCI-e 3.0 x16@ x8x8 to find out if my claim is true or not.

 

Not sure why you're talking about PCI-e 2.0 only cards. That is ofc an exception to the rule but unless you have a graphics card pre-December 2011, this is not an issue.

Yes but in the mean time do you have any other evidence that hints that your theory might be true?

 

Pci-e 3.0 @ 8x/8x is equivalent to pci-e 2.0 @ 16x/16x Ari. Look at my tests, I'm not understanding you...

 

Who is talking about PCI-e 2.0 only cards? this means "but cards being PCIe 2.0 can't take advantage of that." that cards can not run PCI-e 2.0 @ x16x16 because cards can not take advantage of x16x16 only x8x8 is possible when PCI-e 2.0

Theory, no it is a fact.

 

I think there is a language barrier miscommunication problem occurring between us here.

PCI-e 2.0 x16/x16 is possible if you have the correct CPU (and motherboard) that supports at least 32 lanes of PCI-e 2.0. Such cpu's included the i7-920 and X58 chipset boards with 2x PCI-e 2.0 x16 slots.

But for most cpu owners such as the i5-2500k only have PCI-e 2.0 with 16 lanes max. If someone has this cpu and even if they have 2x PCI-e 2.0 x16 slots on their motherboard, they can only run their SLI/Crossfire in PCI-e 2.0 x8/x8.

I see that you have an i7-4770K meaning you can only run your 780Ti's at PCI-e 3.0 x8/x8 (same as PCI-e 2.0 x16/x16).

update: Use this post by TechAde comparing his 780Ti in two different systems yielded different results. At the time of reporting these results he didn't realise it was the PCI-e 3.0 vs 2.0. http://isiforums.net/f/showthread.p...-for-rFactor-2?p=310450&viewfull=1#post310450

GTX 780Ti

PCI-e 2.0 x16 = 134.185 fps
PCI-e 3.0 x16 = 168.981 fps (26% more fps)

And if you were using SLI in rf2 (provided SLI works in rf2) then that same 26% would translate to SLI performance from 2.0 x16/x16 to 3.0 x16/x16 (if you have a cpu + motherboard that can do that).

 

Why are you talking about PCI-e 2.0 all the time stop that.

What you need to test is PCI-e 3.0 x16@ x16x16 vs PCI-e 3.0 x16@ x8x8

Yes I have i7 4770K OC @4798 MHz, EVGA GTX 780Ti Classified OC @1357 MHz 2-way SLI and my Fire Strike benchmark score is 20450 with only PCI-e 3.0 x8/x8 and I doupt that Spinelli can beat that with PCI-e 3.0 x16/x16

 

I'm talking about PCI-e 2.0 because 2.0 x16/x16 is the same as 3.0 x8/x8.

Yeah, we know this already. Only a handful of games show a benefit from 3.0 vs 2.0, rf2 is definitely one of them. I even posted a video of Fire Strike performance with a high end card in PCI-e 1.1/2.0/3.0 and performance was pretty much identical.

 

No SLI don`t work like that mate you can not compare SLI with single card .

This is how it is, by going SLI you don't double the number of lanes, you are just splitting them between two cards. When a board has x16 x16, it can actively give one of the cards more bandwidth if it is using more, where as the normal set up is just static x8 x8, so you really won't see much of a performance difference.

 

I don't understand why you believe this, why a board can't be in x16/x16 according to you. Like i said, i agree with you this being the case if you have a cpu with only 16 lanes of PCI-e support but if you have a cpu with 32 lanes or more then i see no reason why you shouldn't be able to use SLI/Crossfire in full x16/x16. The only exception would be if you're motherboard does not support x16 in both slots when using SLI/Crossfire. In fact, if x16/x16 is not even possible as you claim then why do manufactures even make motherboards that they clearly state as saying things like "SLI/Crossfire PCI-e x16/x16"?

Also the percentage performance gain in rf2 from PCI-e 2.0 x16 to PCI-e 3.0 x16 with a single card has been shown to be the same with SLI PCI-e 2.0 x16/x16 to PCI-e 3.0 x16/x16 by Spinelli.