PCI-E 2.0/3.0 bandwidth question.

[Luke]

So i'm really ignorant on this topic because every time i've heard something about PCI-E 2.0 vs 3.0 it's something along the lines of "graphics cards don't even saturate the full bandwidth on PCI-E 2.0 so 3.0 will make no difference"  but i've seen charts where there are slight differences in performance. What i'm wondering is: it only high end cards with high memory bandwidth that start seeing slight benefits from PCIE 3.0 vs 2.0? Or is the performance difference (as small as it is) the same across all PCI-E 3.0 cards?

 

 

The reason i'm asking this is because i'm selling a XFX HD 5850 for $65 firm and someone replied "Hello. Because is a PCI2 i will give you $50?"... 

 

I don't even know how to respond.

[antoainb]

Tell them that you won't. Difference in performance will only be big in PCI-E 2.0 8x or less, and there is no reason for you to make a discount because it's a PCI 2.0 card, if it's PCI 2.0 it's made to work there and ONLY there, since PCI-E isn't backwards or forwards compatible.

[flibberdipper]

If he wants to dock you $15 because it's PCIe 2.0, he needs to realize that 3.0 for that card didn't exist, and that it doesn't even need it.

[Glenwing]

It's just margin of error.  It's a few percentage points difference, and in context, when the games are running at 40-60fps a few percent is usually less than 1FPS difference.  If graphics cards were being bottlenecked by, say Gen3 x16 vs Gen3 x8, then if you cut it in half again and compare Gen3 x8 to Gen3 x4, you should see a massive drop in performance, since if Gen3 x8 is already limiting it, cutting that in half should make a big difference, but we don't see that.  It is just another few percentage points difference, again.  Or on the other hand, if it is in fact being limited by Gen3 x8, and dropping it to Gen3 x4 still hardly changes anything, then being limited by PCIe bandwidth apparently doesn't matter much anyway, so we shouldn't care about it in the first place.

 

The GPU's memory is soldered onto the graphics card and interacts directly with the GPU's memory controller.  The memory bandwidth is between the GPU and that memory, it doesn't go through the PCIe bus.

[ForsakenLive]

It's about a 1% difference in performance, that comes from the reduced latency of the different maximum bandwidths, not actual bottleneck on data transfer. It's like saying there is less delay on the bus that has more bandwidth to spare.

[Askew]

It can start to come into play the more GPUs are sharing the PCIE lanes in extreme situations on certain platforms, I'm not talking about Intel 2011 socket as there are more PCIE lanes to be distributed and you can run multiple GPUs with 16 lanes each, I'm talking about consumer sockets like 1155 for instance.

 

For example, if you run a single GPU in a PCIE x16 slot on an 1155 motherboard you get 16 PCIE lanes, but if you add a second card the slots become effective 8x slots, and each card has to use only 8 lanes.

 

PCIE 3.0 lanes are double the bandwidth of PCIE 2.0.

 

For single GPU there is virtually no performance difference between even PCIE 1.1 16x and 3.0 16x, but as you start to add multiple powerful GPUs we are tickling the limits of PCIE 2.0 when only using 8 or 4 lanes, but for the majority of most applications it still makes no difference.

 

There are some documented results of quad SLI cards being severely hampered by using PCIE 2.0 (I can dig it up if desired) there are a lot of conflicting tests on the issue with some tests showing zero difference between version 1.1 and 3.0 even in extreme scenarios.

 

However, all PCIE versions are intercompatible and your card being 2.0 is a preposterous reason to offer you less than its worth that's like pricing by the amount of VRAM a card has.

[Solkia]

So i'm really ignorant on this topic because every time i've heard something about PCI-E 2.0 vs 3.0 it's something along the lines of "graphics cards don't even saturate the full bandwidth on PCI-E 2.0 so 3.0 will make no difference"  but i've seen charts where there are slight differences in performance. What i'm wondering is: it only high end cards with high memory bandwidth that start seeing slight benefits from PCIE 3.0 vs 2.0? Or is the performance difference (as small as it is) the same across all PCI-E 3.0 cards?

 

 

The reason i'm asking this is because i'm selling a XFX HD 5850 for $65 firm and someone replied "Hello. Because is a PCI2 i will give you $50?"... 

 

I don't even know how to respond.

 

Tell him he's an idiot, he's not buying a PCI-e lane, he's buying the GPU that goes in it.

 

And $65 is a steal for this card. He's just being greedy.

[Luke]

Haha i love these comments. Thanks everyone.

[Askew]

Haha i love these comments. Thanks everyone.

 

I told you I don't always post idiotic comments but you didn't believe me. (just realized you are that guy lol)

[P e r s e p h o n e]

I told you I don't always post idiotic comments but you didn't believe me. (just realized you are that guy lol)

Ha, I remember that. 

[Luke]

I told you I don't always post idiotic comments but you didn't believe me. (just realized you are that guy lol)

 

No no you seen it was me so you posted that very helpful comment just to prove me wrong! Haha jk, Thanks =)