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

Oh bugger. Are you 100% sure?

 

Yes, DrR. There are only a certain amount of highway lanes (16 or 40) regardless of the maximum amount of cars that can drive through those lanes every minute (3.0, 2.0, 1.0).

 

balls. Then if that's true, we need 2 people, someone like SimonV6 who also have 2x 970's and get him to run his at pci-e 2.0 x16/x16 with same clock and mem speeds and compare results.

 

We can still apply the test in single GPU mode like I explained in post # 199.

We already know the huge benefits rF2 (mysteriously) shows in PCI-E 3.0 @ 16x compared to 2.0 @ 16x. Also, we already know that this same PCI-E mode advantage holds true in SLI mode. Therefore the single GPU test results of 3.0@8x VS 2.0@16x should almost certain carry over to SLI.

 

SimonV6 and Prodigy, we need your help please to confirmed PCI-e 3.0 x8 = PCI-e 2.0 x16 performance in rfactor 2.

If you have the time can you please run the live benchmark with the benchmark settings and set your gpu and memory clock speeds to 1300mhz and 1750mhz exactly when in load please. Afterburner is great for this as it has a live graph and built in program (kombuster) to make your card run at load.

Here are the differences though for each of you.

Prodigy, keep your cards in PCI-e 3.0 mode (which should be running in x8/x8 on your system automatically) and test with SLI enabled and disabled.

SimonV6, you'll need to enter your bios and change your graphics card pci-e mode from "auto" (or "gen 3") to "gen 2". This will force your cards into PCI-e 2.0 x16/x16 mode. Then run the benchmark in both SLI enabled and disabled mode.

And report back your scores please. Thank you in advance if you have the time to do this.

 

Ah, think i see what you mean now. So even with SLI disabled, so long as he had 2 cards installed on the motherboard it will still force the single card into 3.0 x8 mode?

 

Yup because that single GPU is still "stealing" 8 lanes by being plugged-in. He might not even have to remove the second GPU to test the first one in 2.0 16x mode; he may be able to just disable the second GPU's actual PCI-E slot from the BIOS.

 

I swapped out my 2500k for a 3570k. Verified with GPU-Z that my single 780ti PCI-e runs at 3.0 x16. Ran a wide-screen "standard" test similar to Spinelli (5260x1050, multiview) except I used a BT-20 cause I don't have Hondas installed.

CPU: i5-3570K @ 3.6GHz
rFactor2 Build: 860
Graphics driver version: 331.82 (need to update)
780ti core 1125Mhz, mem 3500Mhz

pci 2.0
Avg: 66 - Min: 54 - Max: 78

pci 3.0
Avg: 76 - Min: 62 - Max: 90

improvement:
+15% -- +15% -- +15%

Then I drove one lap manually, following 20AI with 25deg FOV to see how it would do in a more race-like condition. This included the standing start which is the toughest FPS test:

pci 2.0
Avg: 46 - Min: 29 - Max: 71

pci 3.0
Avg: 59 - Min: 46 - Max: 82

improvement:
+28% -- +50% -- +15%

I'm pretty pleased with that and thank TechAde, Spinelli, DrR, and all the others who found and explored this technical glitch. For me, already having the proper mobo and vid card, it was a cheap exploitation just swapping my 2500 for a 3570.

 

Ricknau, nice job

 

Nice results there Ricknau.

 

Should be sticky in Hardware this thread.

Just as a headups for people upgrading, etc.

I found same as Ricknau my 4690 running only 3.9GHz blows the doors off my 2500K@4.5Ghz in rFactor2.

......that's all I care about.

 

What sort of tests? Online and many non-A.I. opponents? Offline and many opponents? No opponents?

 

You mean me ?

Online 8 cars was at least 20-25% improvement at 600MHz less.

Offline and many opponents? No opponents? I have no interest in.

 

Haha.

Yes, online with real-life opponents is a much better test for CPU than offline.

 

Not sure I understand how you're relating DD's online test to his results. Are you saying online offers more resistance to good FPS and that my 3570 should be better able to overcome, compared to my 2500? Or let me ask a different way... How should the offline gains I got with the 3570 pci3 vs 2500 pci2 compare to an online race with equal opponent count?
Can you predict that?

 

I am not talking at all about different PCI-E versions, or even the GPU at all. Durge just said something about his new CPU, and I asked if it was offline or online because I noticed when I set my CPU to 3.3 GHZ and then set it to 4.5 GHz I hardly had an fps improvements when offline, however, when I was online, especially with more real-life opponents, then there was a difference going from 3.3 GHz to 4.5 GHz. That's why I was curious if Durge's fps improvements were offline or online.

EDIT: Ahhhh, I just realized that Durge's improvements were mostly because of his GPU now using PCI-E 3.0 @ 16x. The way he worded his post made me think that he was referring to the new CPU itself giving him more fps, lol. I can see why I confused you Ricknau, lol. My mistake.

 

Hmm, i'm a little dubious of their being an fps gain with the clock speed increase when playing online Spinelli. I could be wrong but how did you compare before and after overclocks online and how similar were the test conditions, etc. You'd need exacting conditions to make it a fair test, very difficult/impossible to do online.

 

It wasn't exact science I just noticed that when offline and alt-tabbing out of game my fps would hardly increase going from 3.3 GHz to 4.5 GHz. While online, I remember getting more gains from the increased clockspeed. I think the CPU get's used quite a bit more online. I know it's the same with quite a few other games as well, for example, Battlefield is a very good test for CPUs and CPU overclocks in multiplayer mode (especially with many opponents), offline though, it's all GPU.

Maybe I'll do some tests later to see if my apparent results are valid.

 

Yeah mate I attribute most of the performance to the Pcie3.0 most def.

However, the 4690 and chipset is just faster in everything like I said at 600MHz less in most things it runs identical or a little snappier.

Like some things would be slower but I have not noticed anything yet.


Even windows install picked up a minute my trueimages backups , restore and backup a tiny faster , desktop boot is faster ( and I dont use fast boot ie: I do not like efi or GPT partitions or mixing them with my other drives while hot swapping ) So I give fast boot a miss I am not that anal about speed.

 

I would like to see the bench-test format extended to include 1, 12 and 24 cars.

I think we could get some interesting data , when it comes to better CPU's and faster clockspeeds vs same GPU.