Live Performance Benchmarking Comparison for rFactor 2

Its easy, just download that replay, use the settings in the screenshots forrfactor 2. Run fraps, you'll have to look at what the default key is to start the benchmark but I think its F11.

 

OK, I've re-run the benchmark after some tweaks in Inspector to ensure GPU core clock stays consistent during the run.

Power and Temperature Target: 125%/91°C
Over Voltage: +87.5mV
Fan: 100% (just in case!)


CPU: i7-4770K @ 4.3GHz PCIe 3.0 x 16
rFactor2 Build: 860
Graphics driver version: 344.11 WHQL

GPU: EVGA GTX 980 SC default clocks (Core 1240MHz, Boost 1380MHz, Mem 3506MHz)

Run #1
Frames: 11972 - Time: 67627ms - Avg: 177.030 - Min: 144 - Max: 202

Run #2
Frames: 11965 - Time: 67564ms - Avg: 177.091 - Min: 144 - Max: 202

GPU: EVGA GTX 980 SC overclocked (Core+160Mhz, Mem+500MHz) (Core 1400MHz, Boost 1540MHz, Mem 4001MHz)

Run #1
Frames: 13126 - Time: 67501ms - Avg: 194.456 - Min: 157 - Max: 223

Run #2
Frames: 13132 - Time: 67533ms - Avg: 194.453 - Min: 157 - Max: 223

 

TYVM rouge22

 

AMD calls it "flip queue size". I remember not having great luck with it when I last tried it with my HD 7950 and 7970; it was as if the setting didn't actually work or something. I remember back in the day (ATI Radeon x850XT Platinum Edition) the setting worked perfectly.

Plus, most people don't have Radeon Pro, and having to download it just to make the benchmark even with NVidia users is just a bit of a hassle and I'm guessing most people won't even bother. It would make much more sense to just leave NVidia and AMD both on completely, 100% default control panel settings for this benchmark, with the exception of 4xMSAA of course.

 

All definitely set correctly

I just tried another run after installing the later driver (344.16), but it made absolutely no difference...

GPU: EVGA GTX 980 SC overclocked (Core+160Mhz, Mem+500MHz) (Core 1400MHz, Boost 1540MHz, Mem 4001MHz)
Frames: 13144 - Time: 67564ms - Avg: 194.541 - Min: 157 - Max: 223

Edit: I'm not sure how sustainable my 500MHz vRAM overclock is, all seems well with the fan @ 100% but with fan at default speeds I see some texture corruption and the GPU resets a few seconds into the lap.

Edit2: I tried a run with Texture Sharpening="0" (i.e. same as enabling the old Alt-M to disable any negative mipmap LOD bias)..

Frames: 13284 - Time: 67674ms - Avg: 196.294 - Min: 158 - Max: 227

There's a teeny improvement but not much.

 

Going to try it in a benchmark just out of curiosity.

I've always used radeon pro, I use it for help with crossfire.

I just never knew what flip queue size was so I left it at default which is a setting of 3. So I'll drop it to 1 and see what happens.

 

1470/3800 (or as close as I could get) gives...
Frames: 12672 - Time: 67580ms - Avg: 187.511 - Min: 152 - Max: 215

And here's the Inspector graphs during that run...

View attachment 14402

 

Spinelli, after setting flip queue to 1 to max nvidias, made no. I thought it gave me gains but I think it was due to me starting the benchmark too early or late.

 

Sweet, and you should find that repeats of the benchmark average score should something crazy consistent like 99.96% of each other. This way you don't need to post both scores anymore, just pick either, they'll be seperated by less than 0.1-0.2 of an fps, completely insignificant.

 

Well, it's there to ensure cpu utilisation is not clipping 100%. If someone with AMD wants the same certainty that the cpu is not bottlenecking, they need only check the cpu utilsation graphs from during the benchmark. If non of the cores clipped 100%, they don't need max pre-rendered frames = 1.

 

This is a live benchmark, there is no replay file.

 

Do you still see this extra CPU usage if you don't limit MPRF to 1?

If so please try something for me - disable shadows in rF2, then see if it still occurs.

Sent from my SM-G900F using Tapatalk

 

Can you perform the benchmark without queue to 1 and show me your cpu utilisation graph from during the benchmark please. Also check the performance went down again with queue 1 back to how it was before.

If you cpu utilisation graphs show cores clipping 100%, then it really was a cpu bottleneck issue without setting queue to 1.

 

Not as simple as that. Framerates still have a chance of being lower when not set to 1 because the CPU, I believe, is busy preparing frames (or something like that). Regardless of which way you put it, it's not the same, and it should either be instructed to everyone to download radeon pro and set flip queue size to 1, or just for everyone (NVidia and AMD) to use default. One or the other - having all NVidia users using a setting of 1 while all AMD users use default (most likely 2, maybe 3) is unfair for AMD users.

2D SLI players are also stuck with default (or I believe max of 2) pre-rendered frames per second; even if they set it to 1 with 2D SLI, the back-end ignores the setting of "1". I wouldn't use this as a reason to change the base settings for the rF2 benchmark settings though, if that's the way SLI is then that's the way SLI is regardless of what you decide to choose for the benchmark settings, but I just felt like throwing this piece of info in anyways

 

Whilst i understand it's not ideal, the solution is not to let the problem of incorrect scores occur with both AMD and Nvidia users.

The solution is to find a way for AMD users to also ensure they can avoid this problem for certain as well.

If you send me a post with pictures and/or written instructions on how to set the equivalent "Max Pre-rendered frames = 1" for AMD cards, i will gladly add it to the OP.

 

Yeah I realize what I told him, I haven't had my coffee yet, lol.

 

Ok, if this is true then perhaps it's best to make it a requirement that all nvidia users set "Max pre-rendered frames = 1" and the same equivalent for AMD users?