 | |||
|
|
|
|
|
 |
PCI Express Performance: Two Years Later October 11, 2006 Brandon Sandman Bell |
|
 |
|
| |
Summary: With faster hardware and more powerful games, we set out to test the performance improvement x16 PCIe brings over x8. You may be surprised by some of the performance gains we found, and where we saw them in this article!
 Introduction | Page:: ( 1 / 13 ) |
What we’ve set out to do with this article is to take an updated look at PCI Express performance with newer titles than what we initially used a few years ago, since PCI Express first debuted in 2004 we’ve seen more shader-intensive titles like F.E.A.R. be introduced, as well as more demanding benchmarks like 3DMark 06. We’ll also be using NVIDIA’s SLI technology exclusively for this article; with two GeForce 7900 GTX cards connected together running in SLI mode at both 4xAA and 8x/16x Super AA modes at resolutions as high as 2560x1600, we should hopefully push the PCI Express interface a little more.
Since NVIDIA hasn’t produced an AGP variant of the GeForce 7900 GTX to this point, we won’t be comparing the performance of PCI Express to AGP in this article, instead we’ll be examining the difference in performance between the 8x and 16x flavors of PCI Express.
To test this out we’ve gathered nForce motherboards from ASUS, the P5N32-SLI SE Deluxe and P5NSLI, which support 8-lane and 16-lane PCI Express operation for SLI. While these motherboards are based on different chipsets, the nForce4 SLI X16 and nForce 570 SLI, the architecture of both is fairly similar, only the nForce4 SLI X16 chipset used on the P5N32-SLI SE Deluxe has more PCI Express lanes: 48 versus the 20 lanes on the nForce 570 SLI. As we found in our ASUS Core 2-ready motherboard roundup, both motherboards perform similarly to one another under equal conditions, so we feel like we should be able to make valid performance comparison between the two.
We realize that there are a lot of you still out there with AGP systems and we’ve got an article in the works for you that should help in determining the best AGP upgrade path for you.
With PCI Express 2.0 set to debut next year offering twice the bandwidth of today’s PCI Express 1.0 devices, we’ll probably be running similar tests a year from now asking the same question all over again, but that’s for another day...
| Test Systems | Page:: ( 2 / 13 ) |
System Setup
Intel Core 2 Extreme X6800
ASUS P5N32-SLI SE Deluxe (PCIe 16x system)
ASUS P5NSLI (PCIe 8x system)
2GB Corsair TWIN2X2048-6400C4
NVIDIA GeForce 7900 GTX 512MB
Driver version ForceWare 91.47
250GB Maxtor Hard Drive Maxline III SATA Hard Drive w/16MB Cache
Windows XP Professional SP2
DirectX 9.0c
Benchmarks
Half-Life 2 Lost Coast
F.E.A.R. 1.07
Quake 4 1.2
Elder Scrolls IV: Oblivion
Call of Duty 2 1.3
3DMark 06
Notes
Again keep in mind that we’re testing the NVIDIA cards with the image quality setting at “High Quality” mode rather than the driver default setting of “Quality”. We’ve noted that the HQ setting significantly reduces the amount of texture shimmering in games such as Battlefield 2. This change does negatively impact NVIDIA’s performance, but it’s a tweak many NVIDIA users seem to be doing with their own cards so we’re doing it too.
We were unable to test Call of Duty 2 under NVIDIA’s 8x and 16x Super-AA modes with the P5N32-SLI SE Deluxe motherboard. Unfortunately at those settings the game would repeatedly lockup or crash during testing.
| 3DMark 06 | Page:: ( 3 / 13 ) |
3DMark 06 – Direct3D
| HDR: HL2 Lost Coast | Page:: ( 4 / 13 ) |
Half-Life 2: Lost Coast – Direct3D
| Quake 4 | Page:: ( 5 / 13 ) |
Quake 4 – OpenGL
| F.E.A.R. Performance | Page:: ( 6 / 13 ) |
F.E.A.R. – Direct3D
| F.E.A.R. Performance 1920x1200 | ||
| Card | Min FPS | Max FPS |
| 7900 GTX SLI PCIe 16x | 42 | 264 |
| 7900 GTX SLI PCIe 8x | 42 | 263 |
| ||
| Oblivion Mountains HDR | Page:: ( 7 / 13 ) |
Oblivion – Direct3D
| Oblivion Performance 1600x1200x32 | ||
| Card | Min FPS | Max FPS |
| GeForce 7900 GTX SLI PCIe16x | 44 | 83 |
| GeForce 7900 GTX SLI PCIe8x | 44 | 82 |
| ||
| Oblivion Foliage HDR | Page:: ( 8 / 13 ) |
Oblivion – Direct3D
| Oblivion Performance 1600x1200x32 | ||
| Card | Min FPS | Max FPS |
| GeForce 7900 GTX SLI PCIe16x | 29 | 35 |
| GeForce 7900 GTX SLI PCIe8x | 28 | 38 |
| ||
| Call of Duty 2 | Page:: ( 9 / 13 ) |
Call of Duty 2 – Direct3D
| SLIAA: Quake 4 | Page:: ( 10 / 13 ) |
| SLIAA: FEAR | Page:: ( 11 / 13 ) |
| F.E.A.R. Performance 1920x1200 | ||
| Card | Min FPS | Max FPS |
| 7900 GTX SLI8x PCIe 16x | 23 | 144 |
| 7900 GTX SLI8x PCIe 8x | 22 | 144 |
| 7900 GTX SLI16x PCIe 16x | 8 | 115 |
| 7900 GTX SLI16x PCIe 8x | 11 | 76 |
| ||
| SLIAA: Half-Life 2 Lost Coast | Page:: ( 12 / 13 ) |
| Conclusion | Page:: ( 13 / 13 ) |
Under 4xAA testing, Quake 4 performance improved by up to 6% at 1600x1200 and 1920x1200, at the mega resolution of 2560x1600, we still saw a tangible improvement of 3%. At that point it looks like the GPU begins to become more of a bottleneck than the PCI Express interface itself. F.E.A.R., a much more shader-heavy game than Quake 4 however, shows basically the same result between 8x PCIe and 16x PCIe. The same applies to Oblivion, and even in Half-Life 2 Lost Coast performance was pretty similar; only at 1920x1200 did we see a performance boost of 3%.
All the other apps we tested with though showed nice gains for x16 PCI Express. Call of Duty 2’s performance improved by as much as 10%, again at the lower resolutions of 1600x1200 and 1920x1200. For this batch of tests we set CoD 2’s texture quality at the game’s “Extra” setting, normally we leave it at “High” because the performance hit from running at extra is just too great. Under lesser texture quality modes the performance improvement may not be as great, as you’re obviously dealing with smaller textures under those modes.
Meanwhile the x16 PCIe system registered a 7% improvement in 3DMark 06. Game Test 4 showed a nice 15% performance improvement with the 16-lane PCI Express system.
Under the greater demands of NVIDIA’s SLI-AA, the added bandwidth provided by the 16-lane PCI Express setup isn’t taken advantage of, at least from our tests with Quake 4, Lost Coast, and FEAR. It’s a shame we couldn’t get the x16 PCIe system to run more stable in CoD 2 under the SLI-AA modes. Considering our 4xAA results with CoD 2, we were really eager to see what would happen under 8x SLI-AA. Oh well.
It’s beginning to look like added bandwidth provided by PCI Express is finally beginning to be used by some applications, although clearly there's still lots of room for improvement. With next-generation DX10 hardware and games right on the horizon, the PCI Express interface will likely be taxed even more heavily in the near future.
Based on this, those of you shopping for new motherboards may want to make sure it sports dual 16-lane PCIe operation, at least if you crave squeezing the most performance out of your PC and its underlying components.
However, this may not necessarily mean that those of you with AGP systems should upgrade to PCI Express based solely on these numbers (the margins would likely be slightly different had we used PCI Express ATI cards as well). Under a single GPU configuration, it’s quite possible that you may not see the benefits we’re seeing here, or the margins between AGP and PCI Express may not be as great, particularly if you’re dealing with slower hardware than the GeForce 7900 GTX. We’ll have to test that theory out when we get a chance…