Does PCIe 2.0 matter yet?

[spuddyt]

When PCIe2.0 came out initially, everyone said (and indeed correctly, according to tests) that the first cards to use it, would essentially not use it - the 3800's were unable to use this large bandwidth increase for any notable performance gains. However, the 3800's have been replaced these days by the 4800's, and indeed, the 4870x2's.
So my questions are: Will a single GPU 4800 (say, 4890) series be bottlenecked by a PCIe 16x 1.1 slot, and if not, will a double GPU (4870x2) do so.

 

[vj8usa]

http://alienbabeltech.com/main/?p=2249&all=1

The Lost Planet results are especially interesting.

 

[MrK6]

There's not much recent information on the web about this. When I upgraded from a 4870 to my GTX 295, I also upgraded my motherboard from a P35-DS3P (PCIe 1.1) to a EP45-UD3P (PCIe 2.0). The little information I found suggested there is a difference when using an X2 card since all data must be pumped through only one slot. Here's the article I read: http://www.tomshardware.com/re...press-2.0,1915-10.html . As you can see, games show a performance increase going from PCIe 2.0 8x (equivalent to PCIe 1.1 16x) to PCIe 2.0 16x. Whether or not this is a valid form of comparison, I don't know. However, my reasoning was that if a 9800GX2 showed improvement (throwing numbers around, 512MB on each card, 64 GB/s memory bandwidth per card, so consider the data it's capable of processing and how much the PCIe will have to supply), my GTX 295 would show even greater improvement (896MB per card, 112GB/s bandwidth per card at stock, etc.). For me, the ~$25 upgrade after selling my old board was worth it. Now I think when using a single card solution there might be less (even negligble) improvement, but who knows.

EDIT: Nice find vj8usa! Judging by the results, performance seems more driver-related rather than capped by bandwidth. Wish there was something like this on the GTX 295.

 

[spuddyt]

Originally posted by: vj8usa
http://alienbabeltech.com/main/?p=2249&all=1

The Lost Planet results are especially interesting.

Ah, extremely informative, thanks
Since it seems the 4870 is just hitting the edge on bandwidth starting to cause issues, I think something in the region of a 4770 might be in order....

 

[cbn]

Originally posted by: vj8usa
http://alienbabeltech.com/main/?p=2249&all=1

The Lost Planet results are especially interesting.

That is pretty strange that a PCI-E 1.x slot affects GTX 280 in that benchmark when at the same time the 4870x2 is unaffected.

 

[Nathelion]

Didn't Anandtech do an article on this awhile back? IIRC the double-GPU cards were the only ones to show a significant performance increase with PCI-E 2.0

 

[cbn]

Speaking of PCI-E 2.0 isn't PCI-E 3.0 supposed to be out this year?

 

[imported_Shaq]

I think 3.0 is to be released next year.

 

[dflynchimp]

It looks from the link that the GTX280 is most severely effected by the bandwidth difference. My guess is it has something to do with the large 512bit bus on the G200's compared to the 256bit bus used on the 4870 cards.

 

[error8]

I still think that for the current high end cards on the market, PCi express 1.1 gives enough bandwidth, even though there is a frame loss here and there, nothing too important. But it's very clear that the next generation will surely need pci express 2.0 to run as they should.