Live Performance Benchmarking Comparison for rFactor 2

http://www.3dmark.com/fs/2983198 sli 17960

http://www.3dmark.com/fs/2983129 single 11572

 

before I changed my dual sli gtx670 to dual crossfire x with r290x I tested many many hours with different settings and bits. There was no chance (for me) to find the reason for the bad scaling. Finally I gave up the topic and changed to AMD.

But one thing I've learned during this time. It's very important to save the stock driver profile or you have to reset your driver settings with nvidia control panel after every test. You can see in nvi that over the time the unknown bit's grow.
I've no idea from where they come but I know it was hard to find a working procedure that ensure that I have a default profile as a base to see how the impact of some settings is (without doing every time a complete driver reinstall.)

Regardless that I'm now on AMD I have to say top thread with facts and a analytical method. Looking forward and hope really that the reason why most of the sli systems scale bad come out because I know how frustrating it is to have such a system.

 

Ok, at least the inconsistency extends to firestrike as well. It's not quite the 35% extra performance you have in rf2 vs other 970 owners with a single card but it's +11% in the firestrike gpu score.

Edit 1:

Simon, have you been using 3dmark results page to report your gpu core clock speed when running anything 3d? Or have you been reading the value from software like afterburner?

Reason i ask is when i manually set the gpu core clock to be 1300mhz in load through afterburner and then run 3d mark firestrike benchmark, the results page shows me 1072mhz core clock yet your's shows 1300mhz. Assuming this is a faulty reading and it's reading the non boost clock speed (i.e. not the idle and not the load speed but the rated speed) i decided to test how much i'd have to overclock my gpu core clock in load to get a reading of 1300mhz (or in my case 1299mhz which is close enough) in the 3d mark firestrike results page and i had to overclock my gpu core clock to 1552mhz in load. My gpu score was then 13240 which now matches your 13158 (99.4% the same).

With the gpu core clock 1072mhz reading in the result page: http://www.3dmark.com/3dm/4379961

&

With the gpu core clock 1299mhz reading in the result page: http://www.3dmark.com/3dm/4380309

So can you please confirm for me you true gpu core clock reading at load with a program like afterburner. I suspect it should be around 1550mhz range?

Edit 2:

Right and i just repeated the rf2 live benchmark and here are my results @ 1552mhz core, 1900mhz mem:

Frames, Time (ms), Min, Max, Avg
11705, 67595, 140, 198, 173.164 -

and these were your results for reference:

So that's two different benchmarks confirming the same performance. You must have read your gpu core clock from 3d mark and not from a program like afterburner which is the only reliable way to read these values.

So providing that's true (and it must be or else you must have a GTX 980 or something else under the hood), then your results here are correct.

Interestingly still though, we can confirm your single gpu is performing right so you have performed the live benchmark correctly and achieving correct results for a single 970 but your 970 sli is only getting +42% scaling. Whilst it's not the same +83% scaling as you observed before (bizarrely) you are still getting far better scaling than Spinelli and others -35% scaling.

 

Thanks for all your testing, I just put the clocks up through after burner going by what others had put in, so it come up as 1300, but if it jumps to 1550 when under loads, sounds like I might need more cooling?
I just added 186Mhz to the cores through Afterburner, going from what others had overclocked there's to?
Anyone else here running these speeds.? Don't want to damage my cards
standard core for the MSI gaming is 1114Mhz btw

Glad at least the test is now correct, the only way I can get sli working in rfactor though is to change the sli compatibility bit to 0x00500005

 

Something I've noticed is that enabling and disabling SLI can cause any overclock to be returned to default. Could the SLI results be @ default clocks?

Sent from my SM-G900F using Tapatalk

 

The best way is to manually and progressively test overclocks. It's very simple and so long as your not overzealous you won't damage anything.

First set core voltage to default (+0). You can increase power limit (%) to 110 (perfectly safe).

Then leave the memory or core clock at default and only change one. I usually start with memory. Increase it by the amount reviewers of the 970 say they can do it safely by. In my case i went straight for their max overclocks with +500mhz and hit apply. Click the "K" button in the top left and run kombustor. You need to be observant of the tiniest of little sparkles in some pixels and thats artifacting. When/If that happens then lower the memory clock by 25mhz and hit apply and check again for artifacts. Rinse and repeat until you can't see any of them for at least 3-5 minutes. Then reduce your final memory clock by 50-100mhz in afterburner just to be safe. If you don't immediately see artifacting however, then proceed to add 25mhz to the memory clock, hit apply and check. Rinse and repeat. I manage to go all the way up to 600-625mhz on my MSI gtx 970 before i saw artifacts which was such a strike of the luck. Still, i lowered mine down to 500mhz for my final clock.

For the gpu clock speed, the test is much more simple because you just set a new clock and run kombuster and your card will crash (or rather you'll get a black screen followed by windows reporting that your drivers restarted) which is completely safe so long as you don't go for some ridiculous overclocks. Same as memory, i started with the clock speeds achievable by reviews (in my case +185mhz) and i managed to go all the way to 225mhz before i crashed. With the gpu i'd use +10mhz intervals for testing and you only need to test for 10-15 seconds. When you hit the ceiling it'll crash pretty quickly. When you suddenly get a crash, go back 5mhz until you find a speed that doesn't crash after 5 minutes. At that point, i would lower the final clock speed by 25mhz and add +10mV to the core voltage.

Save and apply all the settings and go play. If you observe any crashes that result in a driver restart then simply decrease the core clock by -10mhz.

Forgot to mentioned, i was also really luck with the gpu clock speeds. I didn't need to add any core voltage to get +225mhz on the core but it's different for each card and again a bit of a lottery.

 

That's a really good point.

 

I'm starting to get lost here, lol.

Simon, weren't you using a custom Inspector profile which gave you much more average fps, in single GPU mode, than the stock profile? I remember you saying that the average then dropped to around 175 fps once you re-loaded the original base Nvidia-rF2 profile...

Would you be able to upload this profile for us? I would love to test it...

 

This thread has turned more into a pci-e 3.0/2.0/8x/16x, SLI, and different settings benchmark, rather than just the simple benchmaek thread it was supposed to be. I think a new benchmark thread has to be created with only benchmarks, and only at the correct settings (anything else should get deleted).

Oh, and due to the consistently super-high framerates that people are achieving, the benchmark should be bumped from 4x MSAA to 8x, 9AM start time to 12AM, and sunny to rain. I tested all of these on their own, and combined, and they provide just as consistent framerate results as the original settings while pushing our hardware a bit more while still not being excessive.

 

How are those 290x treating you. Is crossfire working for you in rfactor 2?

 

I was doing this process with Power Limit set to 105% (max that Afterburner would allow) . Got Mem Clock up to +500Mhz for several minutes. Never saw sparkles but +525 made Kombuster crash. Came back down to +500Mhz and moved on to Core Clock. Core Clock ran at +150Mhz for a few minutes so upped it to +165 which made Kombuster crash. Ever since that crash I cannot get the Core Clock to run above 550Mhz (absolute, not +550). Even after clicking reset in Afterburner. Power now runs at a steady 50% so I think my card is in some kind of safe mode. Do you have any idea how free it up?

MSI 780ti Gamer, old default was core 1125Mhz, mem 3500Mhz

Edit: temp leveled out at 75C

 

Sound like msi afterburner causing the problem. Try uninstalling it, restart and use something like gpu-z to read the real-time clocks in a game/benchmark to check if all is well again.

I doubt your card is damaged.

 

Well... I did something even simpler as a first step. Rebooted the PC. That fixed it! It has to be some kind of safety mode. I will continue to play with it. I'll try to establish that the +150Mhz core boost is solid and leave it there.

I did notice what appears to be some kind of throttling going on. The core freq graph in Aftreburner was squiggly on top, with a periodic rhythm... ie, squiggles for a few seconds the solid, then squiggles then solid. More safety stuff I bet. I also noticed that my CPU was not steady at it's turbo speed. I need to explore the BIOS to see if I can get that taken care of.

 

Get rid of that throttling crap, it's a joke and causes stutters. Not to mention, sometimes the card thinks you don't need anymore power and then decides to lower your clocks.

I remember when I first got my first two 780 Tis (I've owned 3 different ones). I was doing tests in rFactor 1 with VSync on. The card would be at the max VSynced framerate and then all of a sudden when a much more graphics intensive part came then the framerates would plummet for a couple seconds and then go back up. I used monitoring software to discover that the card would downclock from around 1150 MHz, or whatever, to - I don't know - 900 MHz or so? I guess the card realized it hardly needed to try in order to push the desired framerates to achieve VSync. Then, in rFactor 1, when the camera would change (replay view TV cam) and show a different view that is much more graphically demanding, or while driving if there is all of a sudden a bunch of dust or smoke particles to render, then the GPU isn't ready for it and then you get framerate slowdowns and stutters because it's only running at like 800 or 900 MHz and takes a second or two to ramp back up. Not to mention, a lot of even the small clock changes brought upon microstutters.

That whole system is a joke.

Go download a custom BIOS and get rid of that auto-throttle crap, not to mention raise the power draw limits. I was going past my 107% power draw limits on stock voltage overclocks. It's a joke. Takes 2 minutes to install the BIOS. By the way raising the power limits doesn't do anything, you won't be automatically using more power or anything like that, it just allows you to use more if you need it, rather than throttle you back when you have so much freaking room to spare.

Kepler Bios Tweaker V1.26 (V1.27 added) TDP unlock GTX 780TI 780 Titan GTX770 ...

NVIDIA GTX 980 Owners' Club
NVIDIA GTX 970 Owners' Club
NVIDIA GTX TITAN Owners' club
NVIDIA GTX 780 Ti Owners' Club
NVIDIA GTX 780 Owners' Club
NVIDIA GTX 770 Owners' Club
NVIDIA GTX 760 Owners' Club
NVIDIA GTX 690 Owners' Club
NVIDIA GTX 680 Owners' Club

 

Yeah I really want to fix the throttling. Thanks for the link(s). They look very useful. I may not load a new BIOS right away but I will do it. Got some other stuff on my agenda this weekend. I was wanting to install a different BIOS anyway because the one I'm using (just upgraded to utilize the Ivy Bridge) doesn't allow CPU OC'ing beyond 38X. Was waffling on that though since CPU boost doesn't do that much for FPS. (Edit: plus I need a bootable DOS thumb drive! Ugh!)

Don't the C1E and C3(?) and C4(?) (if memory serves) settings have something to do with throttling? Sheesh so much to relearn. I only delve into this stuff every few years. I guess I'll look it all up... again.

OMG!!! The 780Ti owners club forum has 1380 PAGES! 13800 POSTS! In less than one year! I won't be digging into that!

 

I think Spinelli was talking about the graphics card bios, not the motherboard bios.

On another note, i've never had the problem of down-throttling when under load both on the gpu and the cpu. My cpu goes from 4400mhz down to 800mhz when not in use and dynamically jumps up as and when needed. Same for the graphics card, core goes down from 1500mhz to 135mhz and memory from 8000mhz to 600mhz; very power efficient this way. Once in a game, they both go to load speeds and stay there without issue.

 

Yeah checking my GPU-Z senors say the same, stays solid in game, as does my cpu

 

Ok, i need to amend what i said earlier about my graphics card not underclocking itself whilst in game, that seems to only be the case in games like rf2. In some others it's constantly down-clocking. In "adpative" mode it jumps around 800-1113mhz and in "performance mode" it locks at 1113mhz. It should be at 1519mhz but for some reason it's not.

edit: Scratch that. I started up my pc from sleep mode and restarting it somehow fixed the issue. Strange.

edit 2: False alarm. It was the vsync in those games which only used 60% of the gpu so the card scales down the gpu frequency accordingly to save power. Removing vsync or setting power mode to max performance removes the issue.

 

Yeah my mind was zoned in on the CPU fluctuations. I have figured out that that was caused by my dragging the Kombuster window around. It fills my entire monitor (3 screens) and I had to drag it over a little to see the cpuz readout behind it. That action drops the CPU freq way down with big fluctuations until I quit dragging it. Then it comes back up to a steady (or fairly steady) turbo speed. The multiplier does waver between 36 and 37. But I'm going to ignore that for now until I learn how to write a new BIOS to the GPU. That'll be a new process for me so I'm going to take my time until I have a real good handle on it.

What I need is a simple step by step tutorial. Those website/forums are for heavy duty nerds. If there's a tutorial buried in here it might take a while to dig it out. Sometimes the effort is a little overwhelming in order to squeeze out that last bit of performance on a card that really runs pretty good as is.