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

Would you be able to maybe be a little more specific?

It should say something like, for eg. PCI-E 3.0 16x @ PCI-E 3.0 16x, or PCI-E 3.0 16x @ PCI-E 2.0 8x, or whatever. Something like that...

 

Sorry to barge in... just to put the resolution thing to bed... I ran the test in the other thread (correctly in the end) and then compared to 2160p

The 2160p results are in red

PCI-E @ 2.0 x16
Avg: 125.090 - Min: 105 - Max: 160
Avg: 80.550- Min: 66 - Max: 92

PCI-E @ 3.0 x16
Avg: 165.660 - Min: 119 - Max: 181
Avg: 89.850 - Min: 76 - Max: 100

With the synthetic 'live' bench with no AI, AA forced to 4x and pre-render frames to 1, there is a difference between PCIe2 and 3, but not a lot compared to that at 1080p

In real simulation racing though, but with AI control with 20 AI and at grid pos 10, level 4 (8x) AA, fxaa on there is barely 1fps difference, Min-max and average.

So we are all right... but to be honest, what matters to me is in racing against others, whether AI or Human... and that's where the performance counts... so 2.0 to 3.0 x PCI bus speed is a moot point at that level... oh well.. it was worth a go... OC'in the GPU and VRAM though.. that makes LOADS of difference caveat (if you can avoid throttling due to power draw which I can't on air)... I am going to put on the waterblock tomorrow, ran out of 'being arsed' tonight.

 

Lol! You're joking right?

You just gained about 10 fps and over 10% fps in everything - Avg, min and max framerates. How you consider that difference a moot point but overclocking a big point is beyond me. Lol!! What do you get from overclocking? 40 extra fps? Lol 12% is considered a very respectful difference in gaming framertes. 12% is sometimes the difference between a $600 and $300 GPU.

And then on top of all that the comments about racing against other opponents is what matters more? Lol, that will place even more demand on your GPU and possibly even widen the performance difference between the two PCI-E modes, just like the gap was widened in your 4K tests.

I think you may have had too much to drink today

 

??? Online is quite CPU intensive as well.

Also, a situation where the CPU is not as crucial means that the CPU is less likely to be a bottleneck, which means the GPU will have more potential demand placed on it, which means the PCI could potentially widen even further in a situation with little CPU load.

 

hey man... I gained 10% in a synthetic benchmark that I will never race in... I have had no drink at all. I spent my evening trying to put something down, and it looks to me like you have not read it again man...lol

I gain 1fps in my bench which is closer to what I race with... 1fps... that's not 10%...and you didn't even copy that bit in the quote for some reason.. trying to make others look stupid doesn't always work you know.

I didn't want confuse the benches, hence why I stated what I had gained (1 fps) rather than put another table up.. but you missed that bit....lmao

I'll leave you to it.. I'm not here to argue... just to share info ... I don't mind even being wrong... but I won't put up with being made look stupid for no reason Spinelli.

EDIT... No offense meant.my average CPU load in the bench was < 5%

it was around 22% in my own bench with 20AI...

 

Lol, I did read that part and that could be because of increased CPU usage with all the A.I. bottlenecking you more. The fact is, when the GPUs are allowed to breath and do their "thing" you also got quite respectible gains, not to mention HUGE gains in 4K. Also, do you honestly think that this benchmark is the only situation where these differences will show themselves? No way

 

I agree.

 

Left it running for 5 minutes. Its says exactly this at the bottom left:

Amd Radeon 7900 series: PCI-E 3.0 16x @ 16x 3.0
Amd Radeon 7900 series: PCI-E 3.0 16x @ 16x 3.0

 

Ok great. Ya it's reading it wrong for whatever reason. It should say something like "PCI-E 3.0 16x @ 8.0x 2.0" for each GPU, or if just running one GPU and nothing else in any other PCI-E slots, then "PCI-E 3.0 16x @ 16x 2.0".

But oh well, who cares what it says, you know what it really is now

 

Yeah I'm not worried, every test or benchmark gives me a result or score I should be getting system combind give or take. Plus I don't run a heavy overclock. Its hard enough keeping this card cool as it is. On top of that. I can run the sim maxed out with solid FPS with 8xAA, I can't complain. I just wish crossfire would work. Even at 2.0 it would give me an even bigger gain.

How about you guys dig into the mult-GPU issues next? lol!

 

Where are you results from the live benchmark rogue?

 

On the first page off the live Benchmark thread.

 

Ah ok, thanks.

So many graphics cards and people posting results...forget where to look. I was thinking you have an nvidia that we could compare your cards performance to see if you're actually getting fault bus interface readings for certain. I still maintain it can't possibly be reading correct since the i7 920 is at least 1-2 generations before PCI-e 3.0 support was placed into intel CPU's.

I think TechAde can help you out tomorrow (i.e. today) if he has the time, testing his 7990 in pci-e 3.0 to confirm.

 

Sweet, tell him to clock his CPU down to 3.5 as well, be interesting to see how this old dog rolls.

 

Hello,
my 2cents...
I've been reading a lot on PCI2 vs PCI3 (16x, 8x) and actually the difference is barely noticeable on single GPU.
here's only one of the many tests you will find.
check out this one: http://www.pugetsystems.com/labs/ar...Speed-on-Gaming-Performance-518/#1080pResults

When going to 2 cards, this might make a bigger difference. (still if you're on PCI2 (x, it's still not that much)

So where do the differences some from:
If you actually compare PCI2 and PCI3 systems, including the CPU, the PCI3 systems will obviously perform better... but not necessarily because of the increased PCI-lanes bandwidth.
As some people suggested to test this properly: you need PCI3 people with new CPU that reduce the PCI-BUS speed to PC2.
I'm pretty sure, there won't be any significant performance difference.

What makes a bit difference, especially in multi-GPU is the CPU...
I'm running 2 7970 with an "old" 965BE clocked @3,8 and it's about 20-30% slower on most benchmarks if you compare to newer i5 CPUs

 

To Cracheur
PCIX3 X8
Example results:

CPU: i5-4670K @ 4.4GHz
rFactor 2 Build: 860
Graphics driver version: 335.23

GPU: EVGA 780Ti (manual overclock) - 1240 core/1880 mem (+220/+280)

Time: 67601ms - Min: 111 - Max: 177 - Avg: 145.604

PCIX3 X16
Example results:

CPU: i5-4670K @ 4.4GHz
rFactor 2 Build: 860
Graphics driver version: 335.23

GPU: EVGA 780Ti (manual overclock) - 1240 core/1900 mem (+220/+300)

Time: 67401ms - Min: 142 - Max: 209 - Avg: 179.332

 

Ok, two things worth of note. You have an AMC cpu and they do not have any PCI-e 3.0 support. So if you're able to switch the pci-e lanes to 3.0 mode, they may not be actually working in that mode. Have you checked with a program such as GPU-z what the reading is? Also, it would appear that the reading can be false (as we suspected based on rogue22's HD 7990 reading in gpu-z but await confirmation of this with TechAde's results for comparison in a system we know will work in PCI-e 3.0 mode due to his CPU haveing integrated pci-e 3.0 support, required for the mobo pci-e 3.0 mode to actually work.

Secondly, even if you had pci-e 3.0 mode working on your motherboard, it would be running both lanes in PCI-e 3.0 x8 mode and not x16 mode in each. And since PCI-e 3.0 x8 has the same bandwidth as PCI-e 2.0 x16 (at least on paper but i've seen or heard somewhere this may not be entirely true an 3.0 x8 may be a little more bandwidth than 2.0 x16 but even if true i wouldn't expect a 20-30% difference as observed in yours to other benchmarkers with i5's if this is true), you would be effectively comparing the same thing which would ofc yield no difference in performance with a graphics card(s).

edit: sorry, i may have read your post wrong.

Have you checked to see if your cpu utilisation graphs are maxed out to 100% in rf2 during the live benchmark? If not, then at least with clock speeds, you're cpu is not bottlenecking rf2.

Also, have you done the same cpu utilisation checks for those other benchmarks where you say that you consistently performed 20-30% slower than compared to the same graphics cards in newer i5's?

From what you're reporting, i would wager the chances of the 20-30% performance loss being more likely a result of the PCI-e 3.0 not being even on in the first place.

But having said that, those newer i5's should not have performed better if their on PCI-e 3.0 either since i5's don't have motherboards with two PCI-e 3.0 x16 lanes at any one time. Based on what i've learnt from Spinelli, you need 40x PCI-e 3.0 lanes (what is required to have 32x available for true two x16 slots for sli/crossfire at PCI-e 3.0) on your motherboard but there are no i5 motherboard with more than 16x PCI-e 3.0 lanes in total if i'm correct? The moment you plug in two cards, each of the two PCI-e 3.0 lanes becomes x8 only (which would effectively be PCI-e 2.0 x16 each, right?).

 

Cracheur, I agree; in 99% of tests - and there are tons and tons of them online - PCI-E 3.0 @ 16X is only around 1-5% faster than 3.0 @ 8x / 2.0 @ 16x. That is why the rF2 results are so shocking (12-20+ % differences). We have quite a few results now from enough different users to prove that this isn't just some fluke result.

AMD CPUs still aren't capable of PCI-E 3.0, although many AMD CPUs offer more than 16 lanes. However, dual 16x 2.0 lanes are the same thing as dual 8x 3.0 lanes which any Ivy-Bridge or Haswell desktop processor can do i5, i7, and even i3 (I thini).

Rogue's 7990 is running PCI-E 2.0. He's on a Sandy Bridge CPU. Physically impossible for him to be on 3.0.

Yes, Dr.R, you're correct; you need SB-E, IB-E, or H-E to get true dual simultaneous 16x 3.0. It's because of the CPU, not the motherboard. The motherboard could support 64 lanes but if the CPU only has 16 then you're only capable 16. Remember, the PCI-E controller for those 16 (or 40) lanes is on the CPU itself. Same with the PCI version - a Sandy Bridge user is stuck with 2.0 even though they might use a 3.0 capable motherboard, for eg. the Z77 motherboards that came out for Ivy Bridge which also work with Sandy Bridge.