PCIe 3.0 vs. PCIe 2.0

As part of our testing on the X79 Extreme11, we decided to test both PCIe 2.0 and PCIe 3.0 scenarios.  Due to the PLX chips onboard giving us a full x16/x16/x16/x16 minus any PLX latency, it should give a rough idea of how these two technologies perform.  Our testing incorporated each benchmark at 2560x1440 using full eye candy settings.  Here are our results, indicated by percentage difference of PCIe 3.0 over PCIe 2.0:

PCIe 3.0 vs. PCIe 2.0
2560x1440, Full AA/AF
ASRock X79 Extreme11
x16/x16/x16/x16
  Metro 2033 Dirt3
1x 7970 -0.3% +3.8%
2x 7970 +2.6% +4.3%
3x 7970 +1.2% +4.2%
4x 7970 +1.9% +0.5%

As we can see, there is an improvement for Dirt3 and Metro2033, though the difference is barely noticable. The effect of PCIe 3.0 depends on the different engines using DirectX and OpenGL – each system, and thus each gaming engine, uses the PCIe bus differently.  In the games where the PCIe bus is used extensively, then PCIe 3.0 will win out.  Otherwise we are at the whim of statistical variation between runs.

Dirt 3

Dirt 3 is a rallying video game and the third in the Dirt series of the Colin McRae Rally series, developed and published by Codemasters.  Using the in game benchmark, Dirt 3 is run at 2560x1440 with full graphical settings.  Results are reported as the average frame rate across four runs.

Dirt 3 - One 7972

Dirt 3 - Two 7972

Dirt 3 - Three 7972

Dirt 3 - Four 7972

Due to the PLX chips, we would expect the X79 Extreme11 to fall behind slightly in single and dual GPU performance, which is confirmed in the benchmark results.  In four-way GPU however, the X79 board falls behind some Z77 boards.

Dirt 3 - One 580

Dirt 3 - Two 580

Using NVIDIA GPUs, Dirt3 is still agnostic to any CPU or PCIe performance.

Metro2033

Metro2033 is a DX11 benchmark that challenges every system that tries to run it at any high-end settings.  Developed by 4A Games and released in March 2010, we use the inbuilt DirectX 11 Frontline benchmark to test the hardware at 2560x1440 with full graphical settings.  Results are given as the average frame rate from 4 runs.

Metro2033 - One 7972

Metro2033 - Two 7972

Metro2033 - Three 7972

Metro2033 - Four 7972

Metro 2033 mirrors similar findings from Dirt3 - the ASRock cannot keep pace with the other boards.  This must suggest that having dual PLX chips offers a much bigger hit to frame rates than previously thought.

Metro2033 - One 580

Metro2033 - Two 580

Computation Benchmarks Testing the LSI SAS 2308 PCIe Controller
Comments Locked

62 Comments

View All Comments

  • AssBall - Tuesday, September 4, 2012 - link

    It is like buying an F350 Superduty Harly Davidson Ed. when a Honda Fit would haul all your stuff. But hey, more power to ya!
  • Taristin - Monday, September 3, 2012 - link

    Almost 2013 and the board still says ATi Crossfire X? Hasn't AMD officially retired the ATi brand yet?
  • Grebuloner - Monday, September 3, 2012 - link

    This is bothering me as I started reading: You keep mentioning 70 lanes of PCIe with the two PLX chips+CPU leftover...divided into 64 for the x16's and "the other 8" for the LSI chip. 64+8=70? Why not just write 72 and end the confusion?
  • IanCutress - Monday, September 3, 2012 - link

    Yes, simple math fail. A brain fart. Call it what you will, I made a mistake, and it should now be corrected. Though a simple email would be a lot more polite... :)

    Ian
  • Performance Fanboi - Monday, September 3, 2012 - link

    Nope, you made a simple arithmetic or grammar error on the internet, LET THE LASHINGS COMMENCE!
  • Grebuloner - Monday, September 3, 2012 - link

    Fair enough, my apologies, next time I shall email. Being a math teacher I get all riled up at math errors.
  • errorman1 - Monday, September 3, 2012 - link

    Mr math teacher, not to be rude but just thought I'd pass this bit of scientific wisdom along.

    When people make errors they do so largely unconsciously, human beings DONT live in reality sadly, including you. What you understand or makes it to your conscious awareness is only a fraction of what is going on in your brain which by and large you don't control.

    Just remember the world doesn't operate on our expectations, morality and 'free will' it operates on cause and effect - the laws of nature. Everyone forgets this fact every day and it leads to great comedy!

    The human mind has limited resources with which to attempt to model the world and it's environment so we should expect errors and blunders as just a law of nature that in many instances people not as blesses in terms of well functioning complex of biological processes.

    http://bit.ly/dYaWUc
  • Death666Angel - Tuesday, September 4, 2012 - link

    I don't see posting the mistake in the comments as impolite. I have done so on a few occasions (typos, wrong calculations...) and did so because it was the fastest way for me to do that. Expecting me to fire up the email program or go to my online email, log in, copy your email adress, type up a formal email, all to appear polite to you while trying to help you is asking for a lot. :-)
  • mfenn - Monday, September 3, 2012 - link

    The comment in the article, "the LSI controller allows for RAID 0, 1 and 10 only, which is a little odd," struck me as a little odd.

    The LSI SAS 2308 is a lower-end chip based on their Fusion-MPT architecture, which has never had an onboard cache or parity (RAID5 and 6) support. Fusion-MPT chips, suitably rebranded of course, are typically used as the base option for SAS connectivity in enterprise-grade servers, with an option to upgrade to a more featureful RAID controller with onboard cache and parity support.
  • FunBunny2 - Monday, September 3, 2012 - link

    Real Men don't use RAID 5/6.

Log in

Don't have an account? Sign up now