PCI Express 1.0 vs. 2.0 vs. 3.0 Performance With NVIDIA/Radeon Graphics On Linux

[phoronix]

Phoronix: PCI Express 1.0 vs. 2.0 vs. 3.0 Performance With NVIDIA/Radeon Graphics On Linux

Following last week's news about PCI Express 4.0 and 5.0, a Phoronix Premium member had requested some graphics card benchmarks when comparing PCI Express 1.0 vs. 2.0 vs. 3.0 performance under Linux.

http://www.phoronix.com/vr.php?view=24775

[fuzz]

but long story short the R9 Fury saw much better performance with PCI-E Gen1 than Gen1 although very minor gains with Gen3

Typo!

[chuckula]

Looks like AMD cards are more chatty over PCIe and like the extra bandwidth. Also, PCIe 1.0 is definitely long in the tooth at this point.

I'm half tempted to get a Skylake X but I'm also torn that PCIe 4.0 is going to be coming down the pike and you'll need a platform upgrade to use it.

[bug77]

Weird showing by the Fury. I'm pretty sure that doesn't happen on Windows (will see if I can unearth some tests later).

[torsionbar28]

Very interesting, awesome benchmarks Michael. It seems there's no practical difference between PCIe 2.0 vs. 3.0, at least with today's graphics cards and games. And with Nvidia cards, even PCIe 1.0 only reduces frame rates by a tiny amount.

Therefore, allocating 16 lanes of PCIe 3.0 to a graphics card for gaming is a waste. If your BIOS allows allocating only 8 lanes, and using the other 8 elsewhere, that's a better utilization strategy. Good to know!

[schmidtbag]

Looks like AMD cards are more chatty over PCIe and like the extra bandwidth.

I'm not so sure about that. Notice how almost all of the 1.0 results are nearly exactly 50% of the other two. This leads me to believe the drivers depend on the availability of all 16 lanes, or at least a certain amount of bandwidth. Meanwhile, perhaps the motherboard doesn't actually use 1.0 speeds, but rather does 2.0 speeds and just chops the slot to 8x lanes. Just a theory.

Michael - it might be interesting to test the Fury on a 2.0 slot with 8x lanes. I'm also curious if Windows for this same build will yield the same results, because if you look at this article, you'll find the PCIe 1.1 16x slots aren't a whole lot slower than the 2.0 and 3.0 16x slots. In many cases, even 8x slots don't have any major difference either.

If we can prove this is a driver situation, this could be very important for the performance of some GPUs for many people. I'm curious how many other AMD GPUs on the open-source drivers and Windows' closed drivers may share this issue.

Therefore, allocating 16 lanes of PCIe 3.0 to a graphics card for gaming is a waste. If your BIOS allows allocating only 8 lanes, and using the other 8 elsewhere, that's a better utilization strategy. Good to know!

And yet people were moaning about the AMD X370 chipset "only" having 16 lanes total for 2x GPUs.

[doublez13]

Thanks Michael. I was super curious about this after your post last week about 4.0.

[devius]

Looks like AMD cards are more chatty over PCIe and like the extra bandwidth.

Or maybe it's because the Nano only has 4GB of RAM, and needs to transfer data more often, although I doubt that applies to these tests.

[DIRT]

Dang. I guess that explain it why amd cards on my pc seem to run a little slower than it should, I'm on pcie v1. I had Nvidia cards before and they ran fine.....Time to upgrade!

[microcode]

Looks like AMD cards are more chatty over PCIe and like the extra bandwidth. Also, PCIe 1.0 is definitely long in the tooth at this point.

I'm half tempted to get a Skylake X but I'm also torn that PCIe 4.0 is going to be coming down the pike and you'll need a platform upgrade to use it.

I would advise against getting a Skylake X, they're prettymuch the same as the socket 2011 equivalents, but more expensive, and with mainboard-level DRM on chipset features.