Huge FPS gains on PCIe 3.0 vs 2.0 @ 1080p in rFactor 2

[jimithebear]

Source;

http://isiforums.net/f/showthread.php/22009-Massive-FPS-gains-in-rf2-using-PCI-e-3-0-x16-with-higher-end-cards!

 

Is this story an Intel plant hoping this will go viral in an effort to force users off Sandy Bridge? Will AMD be forced to 'rush' a chipset supporting PCIe 3.0 to market?!

 

The general consensus up until now is that there are almost negliable performance gains running GPUs over PCIe 3.0 x16 versus x8 / PCIe 2.0 x16, especially at lower resolutions.

 

TechAde on the rFactor 2 forum however has pointed the finger at the PCIe bus as to the reason a performance disparity exists benchmarking the game on identical GPUs.

 

And he appears to be correct?! I've posted MrPix's benchmarking results below;

 

PCIe 2 vs 3..... with desktop and game at 1080p and scaling on Display not GPU...

GPU: EVGA GTX 980 SC OC #1 (1550MHz, Mem 1900MHz, TDP < 88%)

PCIe 2.0 @ x16
Time: 67875ms - Avg: 145.655 - Min: 113 - Max: 176

PCIe 3.0 @ x16
Time: 67627ms - Avg: 187.975 - Min: 153 - Max: 216

so a 29% average increase, 22% max, and the most important ... 35% min increase on a single card!

 

Pasted from;

http://isiforums.net/f/showthread.php/21983-Live-Performance-Benchmarking-Comparison-for-rFactor-2?p=311380&viewfull=1#post311380

 

DrR1pper reports his GTX 970 suffers a 12% performance drop due to PCIe 2.0 x16 - thanks to his Sandy Bridge processor.

 

But don't take their word for it, run the benchmark for yourselves! Download the free demo, all the settings you need are here;

http://isiforums.net/f/showthread.php/21983-Live-Performance-Benchmarking-Comparison-for-rFactor-2

 

Credit to TechAde, DrR1pper and MrPix on the ISI Forum!

[Spongy141]

For some reason I can't help but think this is fake.

[Sauron]

-

 

Welcome to the forum! Please remember to follow your topics so you're notified when someone answers in my sig you can find links to the CoC and the forum's F.A.Q. check them out!

 

To be honest when we're talking about values over 140 fps it doesn't really matter... probably the pcie 2.0 bus limits the number of output frames per second but the value of this limit must be over 120. Basically, the amout of frames the card can push out is reduced because it has to wait for the information to come through in order to calculate each frame, but under 120fps the time it has to wait is very likely lower than the time it takes to calculate a new frame. So, don't worry about it, it doesn't really matter. Even pcie 2.0 x4 is still enough to drive pretty much any card, the performance loss really isn't that high.

[Trik'Stari]

I noticed a complete and total difference when I moved from my fm2 pci-e 2.0 AMD board and APU to a lga1150 pci-e 3.0 board. Everyone tells me I'm crazy and that it doesn't make a difference, but I did notice one. Literally the only things that changed were the board and cpu, same graphics card, same monitor, same everything else.

[AMICLG]

I'm pretty sure it's his methodology of testing or board that's flawed because NO game @ 1080p/1440p (as of now, star citizen might change that) needs more bandwidth than 2.0 @ x8 and there are plenty of benchmarks to back this up so this is either bad testing or fake.

[AMICLG]

I noticed a complete and total difference when I move from my fm2 pci-e 2.0 AMD board and APU to a lga1150 pci-e 3.0 board. Everyone tells me I'm crazy and that it doesn't make a difference, but I did notice one.

That's probably the IPC increase your noticing not the bandwidth.

[Trik'Stari]

That's probably the IPC increase your noticing not the bandwidth.

Hate to sound noobish, but what's IPC? I'm just saying I did notice an extreme difference, I can now play things on high/ultra without severe choppiness, even with the same gpu.

 

I'm pretty sure it's his methodology of testing or board that's flawed because NO game @ 1080p/1440p (as of now, star citizen might change that) needs more bandwidth than 2.0 @ x8 and there are plenty of benchmarks to back this up so this is either bad testing or fake.

Source?

[MEC-777]

Hate to sound noobish, but what's IPC? I'm just saying I did notice an extreme difference, I can now play things on high/ultra without severe choppiness, even with the same gpu.

 

IPC = Instructions per clock. In general, Intel's CPUs are capable of approximately twice the IPC per core vs. AMD CPUs. 

 

This is partly why higher-end GPUs are held back (bottlenecked) by lower-end CPUs. The CPU IPC just can't keep up with how fast the GPU can run at. 

[Enderman]

I'd like to believe that, but ONLY if there are more than one benchmarks, and when it is something more demanding that wont get hundreds of fps.

I can image the PSIe bus throttling over 200 fps, but what about when its 60fps or less? Then there should not be any difference at all.

[SkilledRebuilds]

IPC = Instructions per clock. In general, Intel's CPUs are capable of approximately twice the IPC per core vs. AMD CPUs. 

 

This is partly why higher-end GPUs are held back (bottlenecked) by lower-end CPUs. The CPU IPC just can't keep up with how fast the GPU can run at. 

Agreed, but Not just the GPU ability to run fast,.... but the Game window & for those that didn't think of it... Windows random background data, Driver linkages, The Game Client Window behind the scenes stuff you don't see while it's rendering and playing your game.

More IPC the Better.

[Trik'Stari]

IPC = Instructions per clock. In general, Intel's CPUs are capable of approximately twice the IPC per core vs. AMD CPUs. 

 

This is partly why higher-end GPUs are held back (bottlenecked) by lower-end CPUs. The CPU IPC just can't keep up with how fast the GPU can run at. 

oooooooooh, would that cause choppy framerates?

[AlexGoesHigh]

seems fishy, needs more benches by different people, this may be good for a linus video considering he has all the 980 and 970 out there, well almost

[Laputacake]

My eyes the blue hurts my eyes 

[Trik'Stari]

seems fishy, needs more benches by different people, this may be good for a linus video considering he has all the 980 and 970 out there, well almost

@LinusTech @Slick

 

I for one would like to see this, not to mention see the difference with the same system, except for an fm2 apu and a 4690k. Note I was running a gtx660 on both boards. (yes I know an apu is pointless with an Nvidia GPU, but the only dual graphics gpu for my apu wasn't as good as a 660)

[That Norwegian Guy]

I'd expect the consistent FPS to be very similar, but the PCIe 2.0 scenario hitting a lower minimum FPS during heavy loading/ purging of textures.

 

IE, after loading/ asset streaming, once the frame rate stabilizes, I don't think there'll be much difference.

[cesrai]

oooooooooh, would that cause choppy framerates?

85% I think it would and probably stuttering too, but maybe I'm wrong.

[jimithebear]

My eyes the blue hurts my eyes 

 

Apologies! Blue removed.

[jimithebear]

Welcome to the forum! ...To be honest when we're talking about values over 140 fps it doesn't really matter...

 

Many thanks and agreed, those are some pretty high numbers. The minium FPS DrR1pper reported for his affected GTX 970 is a little more relevant I guess? Although 12% isn't as much of a drop in performance?

 

GPU: MSI GTX 970 Gaming 4G (manual overclock) - 1489 core/1847 mem (+311/+95)

Time: 67564ms - Min: 108 - Max: 165 - Avg: 138.121

 

Note, I don't know whose other GTX 970 he's comparing his results with?

 

If you could hammer the card further (like enabling ShadowPlay for instance?) plus utilise one or two extra PCIe cards, could it spell more danger I wonder?

[rcarlos243]

I haven't looked at the threads or anything but is he using the exact same test platform (same cpu, mobo, etc...) if so how did he switch from PCIe 3.0 to PCIe 2.0?

 

It sounds to me like the test was done on a different CPU thus explains the difference in performance.

[flibberdipper]

I'm doubting this.

[SkilledRebuilds]

I haven't looked at the threads or anything but is he using the exact same test platform (same cpu, mobo, etc...) if so how did he switch from PCIe 3.0 to PCIe 2.0?

 

It sounds to me like the test was done on a different CPU thus explains the difference in performance.

You can switch PCIE speeds within some bios's, I have it manually on PCIE3, otherwise it's defaults to AUTO,I have to choose PCIE 1.0/2.0/3.0 from a drop down list.

[ltguy]

I dont believe this for a moment. When i went from pci-e 2.0 x16 to pci-e 3.0 x16 with my SLI Titans I saw no difference. Furthermore switching from pci-e 3.0 x16 to pci-e 3.0 x8 showed no difference.

[jimithebear]

I haven't looked at the threads or anything but is he using the exact same test platform (same cpu, mobo, etc...) if so how did he switch from PCIe 3.0 to PCIe 2.0?

 

It sounds to me like the test was done on a different CPU thus explains the difference in performance.

 

Same system/components for MrPix;

 

I'm on I7-3820... I have to force PCIe3.0

 

100% sure on 3.0 @ x16 (after applying with force gen3 executable as admin, rebooting and checking in AIDA64 Engineers edition!

 

I believe he's referring to this patch?

http://nvidia.custhelp.com/app/answers/detail/a_id/3135/~/geforce-gen3-support-on-x79-platform

[MEC-777]

Agreed, but Not just the GPU ability to run fast,.... but the Game window & for those that didn't think of it... Windows random background data, Driver linkages, The Game Client Window behind the scenes stuff you don't see while it's rendering and playing your game.

More IPC the Better.

True, but when I look at my CPU usage at idle (CPU at ~800-1000mhz) it's only maybe 3-5% usage, which isn't much. Mind you, I'm running an i5-4570 so perhaps system idle CPU usage on a CPU with half the IPC would be more like 6-10%? That could make a bit of a difference while gaming.

 

oooooooooh, would that cause choppy framerates?

It could but only if your CPU is at or near 100% usage. I see you're running a GTX 660 which should not have been bottlenecked by the APU you had before. The i5's IPC is just that much stronger and that is probably the difference you're experiencing. A GTX 660 is nowhere near close to saturating a PCIe 2.0. 

 

 

As for the topic; I take it with a very tiny grain of salt until more information and testing can show otherwise. 

 

Linus and Slick... This would make for an interesting myth-buster type video.

[ltguy]

This isnt a new topic. Almost every tech site has benchmarks for this very issue.

http://www.pugetsystems.com/labs/articles/Impact-of-PCI-E-Speed-on-Gaming-Performance-518/