Then i apologies if you did but i'm pretty stumped by your results if they're correct. Let me pose the question to you another way. Which do you think is more likely to be true? That your test, the only one that supposedly showed absolutely no affect on performance is correct or that the consistently same reporting that there is a very real/tangible affect on performance by all the others who have also tested (which is around 10-20 individuals...not to mention one of them being TechAde, who has about 20+ different GPU's and 3-4 different systems at his disposal because he works for rFPro) is more likely to be correct? Do you not find this odd and does it not make you pause for at least a moment and ponder whether your results could be erroneous when so many others have consistently confirmed observing the opposite to be true? You could be right and we could all be wrong. I just think the odds of that being the case to be practically infinitesimal given our data sample size and the amount of testing that was done reporting back a tangible affect on performance in rf2 with different PCI-e modes....even 2.0x16/3.0x8 vs 3.0x16 mode. Can you just confirm one last time that you tested your 780 ti in both PCI-e 3.0 x16 mode and PCI-e 3.0 x8 (or 2.0 x16 mode which is the same thing) and that performance was not affected in rf2?
I did run test yesterday and I have just run new test today. My system is i7 4790k, Assus Maximus MBO, EVGA GeForce GTX 780Ti Classified Kingpin. With my PC PCI-E 3.0 x16 vs 2.0 x16. There is no +20% difference with build 998. I did my test like this guy made his PCI-E 2.0 16x vs 8x but my test is made with PCI-E 3.0 (rFactor2 / Heaven) If you don't want to see the hole video here are hes results View attachment 17850 rFactor2 is not Nvidia supported game and Gmotor2 is updated a lot since 09-30-14, 09:01 PM , cpu core updates etc.
Just want to add - its not necessarily the case that somebody needs to be wrong or right, both could be right. There can be other factors (like suggested before) that makes the difference. FPS loss will be noticeable only if PCI-E bandwidth is "filled". We need other people to test PCI-E 2.0 vs 3.0 to see if anything changed lately. Either way rF2 is very sensitive to PCI-E mode/bandwidth.
To my knowledge only few quad core supported game engines can fill hole "PCI-E bandwidth" and take benefit from this. rFactor2 don`t have quad core support.
Define "PCI-E bandwidth". PCI-e 2.0 x16 is limited at ~8GB/s, pci-e 3.0 16x at ~16GB/s. For example my PC (2600k, 8GB DDR3 1600MHz) is able to put ~24GB/s on memory tests. As far as I know new intel CPUs/RAM can do up to 50GB/s total. So do you really think that 4cores are needed to fill more than 8GB/s for pci-e 2.0 vs 3.0 to make a difference? As I understand what matters is RAM/CPU bandwidth vs PCI-E bandwidth and if that bandwidth is used by a game. In case of rF2 it is very sensitive to PCI-E mode/bandwidth. I have found benchmark that might matter: https://www.techpowerup.com/reviews/NVIDIA/GTX_980_PCI-Express_Scaling/21.html Take for example 1080p resolution. You can see that overall performance summary shows very low difference comparing 3.0 16x vs 2.0 16x, the difference is pretty much unimportant. Other speeds show larger differences, but still nothing close to rF2. 2.0 4x vs 2.0 16x is the only test I can replicate with my system. According to techpowerup tests there is ~15% performance difference when comparing 2.0 4x vs 2.0 16x Now interesting part. My test in rF2 going from pci-e 2.0 4x to pci-e 2.0 16x showed almost 2 times FPS difference, which is far higher than any other gpu/cpu bound games that I have seem. The actual values were: pci-e 2.0 16x Frames, Time (ms), Min, Max, Avg 24519, 120000, 186, 222, 204.325 pci-e 2.0 4x Frames, Time (ms), Min, Max, Avg 14070, 120000, 96, 138, 117.250 Alex showed that he got ~50% performance difference when going 2.0 8x vs 2.0 16x. People earlier reported ~20% or less difference going from 2.0 16x to 3.0 16x. My assumption is that there is unnecessary texture loading constantly going on to actually use PCI-e bandwidth in comparison to other games. I cannot think of other reasons. edit: I ran some memory bandwidth tests 1cores vs 4 cores - results were ~17GB/s vs ~24GB/s. Even 1 core is able to fill more GB/s than pci-e 2.0 x16 is capable of. edit2: 1 car driven by AI in silverstone for a lap 1 core pci-e 2.0 x16 Frames, Time (ms), Min, Max, Avg 20599, 120000, 125, 200, 171.658 4 core pci-e 2.0 x16 Frames, Time (ms), Min, Max, Avg 24423, 120000, 187, 223, 203.525 4 core pci-e 2.0 x4 Frames, Time (ms), Min, Max, Avg 14070, 120000, 96, 138, 117.250 Switching from 4 cores to 1 core has less FPS impact than switching from pci-e 2.0 16x to 4x .
Build 930 did have some core updates and to many with older CPU did experience CPU problems. To my PC (i7 4790k) Build 930 was huge performance and fps boost. I did also run back then PCI-E 3.0 16x vs 8x to find out if PCI-e bandwidth is related with CPU cores in rFactor2 and did notice that there is hardly any difference PCI-E 3.0 16x vs 8x in my PC. This is why I claim that rFactor2 Gmotor2 is updated a lot since 09-30-14, 09:01 PM , cpu core updates etc. I don't think test made before Build 930 are valid any more. To ask from ISI quad core support is huge task and it is not needed if rFactor2 is not going to be updated to DX12, 13 etc. Quad core support is huge fps boost in all games also in rFactor2. Yes, but the question is, is rFactor2 Gmotor2 capable to do that.
You know what....perhaps your right and it has changed with newer updates to rf2. I mean, my little quick test yesturday seemed to show the difference has increased from 12% (tested last year) to 26% with no other change to my system. My issue is the flippant statement that whiteshadow made: "I have just tested PCI-E 3.0 x16 vs 2.0 x16. There is no +20% difference with build 998 , claim is placebo. " because I made the claim based on a very clear observation of a 20% fps difference. Ok, he didn't observe this for whatever reason in his own test but it seems rather arrogant for something to just therefore conclude that me observing a 20% difference could be a placebo. I wouldn't make the claim that his results are a placebo because they disagree with mine because for starters i was not there to see his testing in person. At worst it could be erroneous result but a placebo? i don't think someone would mistake a 72 fps reading for a 91 fps reading especially after taking fps screenshots and also checking they definitely reverted back to 3.0 x16 with gpuz and tested also after different resolutions (1080p results were 86 fps and 110 fps). Secondly, i never ended up bothering to make a graph out of all the data from everyone here but at the time i remember that people with the same graphics cards experienced almost identical levels of improvements between the pci-e modes. Maybe that's changed...maybe it hasn't. I still think there is more chance that there is something wrong with WhiteShadow test results than either all others being wrong and even more likely than the possibility that both data sets could be right. Wouldn't it be more interesting to just try and figure out if this is the case or not for sure instead of just assume they are both correct?
Look at the facts which support that Gmotor2 is highly limited by pci-e bandwidth. More than any other game that I have seen tested. If there is a bug or inefficient texture loading or something similar(regardless of the available bandwidth) it wouldn't be strange if pcie3 vs pcie2 has FPS impact(depending on the scenarios).
Why are you showing me this guys result with heaven? No one has made a claim that there is a difference in heaven with x8 and x16, or any other game for that matter in fact. Only rf2. In fact i can confirm i have the same result with either x8 and x16 in every other game/benchmark i've tested. Even x4 mode in 3dmark gave the same exact results when i tested them last year.
It would be very interesting, more tests need to be done, thought not everyone has pci-e 3.0 For me the FPS difference of lower pci-e modes is alarming (considering that rF2 is exceptional here) and it seems not many want to find out the reason. It could be the case that something is done inefficiently causing lower FPS for majority than expected.
I think other forum members are less interested because they thought (and i still think more likely) that the issue of pci-e was thoroughly answered and put to rest last year. There is a wealth of data from these two threads where we thoroughly and satisfactorily concluded there is a difference after some time of not realising it was the pci-e mode causing the result discrepancy between people with the same gpu's. http://isiforums.net/f/showthread.p...f2-using-PCI-e-3-0-x16-with-higher-end-cards! http://isiforums.net/f/showthread.php/21983-Live-Performance-Benchmarking-Comparison-for-rFactor-2 I don't disagree though that things could have changed. I'm going to do the test exactly as prescribed again and see what my results are with the live benchmark settings.
It seem to be important to you who is right and who is wrong. I am simply sharing my findings right or wrong. PCI-E bandwidth differences can be you pc (Z87) vs my pc (Assus Maximus). When I am using words like "placebo" don`t mean that I am arrogant my location is not UK.
You said and i quote..."There is no +20% difference with build 998 , claim is placebo." That statement at face values is basically saying anyone who says they have observed a +20% difference is under an illusion and their results are not real. Ergo...arrogant because it seems like you think that only your results could possibly be real. Perhaps your results are true for your system (although i hold my own personal doubts and think it more likely that your test was erroneous but happy to be proven wrong ofc) but you can't conclude it therefore not possible to see +20% difference with build 998 for anyone else. I only care about getting to the truth. It does not mean i'm just going to roll over and agree that all the well established data is wrong because one person came out with a different result. I would say, let's look at the evidence and try to figure out what is going on which is exactly what i've been trying to do with my questions, etc. Also, no one has said everyone would see a +20% difference. Based on prior findings, it consistently varied from graphics card to graphics card. As of yesterdays mini test, i saw a +26% difference in average fps from 3.0 x8 to x16. However that was not with the live benchmark method so i'm going to go do that now.
Here are my latest test results with the rf2 live benchmarking test methodology as outlined here: http://isiforums.net/f/showthread.php/21983-Live-Performance-Benchmarking-Comparison-for-rFactor-2 CPU: i5-4690K @ 4.4GHz rFactor 2 Build: 998 Graphics driver version: 355.82 1080p: 2.0 x8 to 2.0x16/3.0x8 = 46% increase 2.0 x16/3.0 x8 to 3.0 x16 = 22% increase 1440p: 2.0 x8 to 2.0x16/3.0x8 = 37% increase 2.0 x16/3.0 x8 to 3.0 x16 = 16% increase
Have you compared with other DirectX 9 applications? It may be that DX9 is more inefficient in implementing certain things and thus requires more bandwidth. I know for a fact that DX9 requires duplication of video memory into process memory, which is why rF2 nearly always consumes 2 GB+ RAM when Assetto Corsa often uses less than 1 GB.
I have 2 other dx9 games but I haven't performed consistent testing. Regarding rf2 in my case with textures set to high(not full) it uses ~1,5GB vram. Ofc it depends on the car models and the track.
