Graphics PCI-E Lanes X16, X8, X4 or even X1!

Hi,
I just wanted to share with you guys how silly some people are - when people are not able to provide X16 bandwidth for their "new" "powerful" graphics card!

Well, here's the truth: even if you run a powerful graphics card like the 5870 at PCI-E v2 4X, the difference in performance is within margin of error, i.e. just fluctuations with no actual difference!

Even when you plug a X16 card into an PCI-E v2 X1 slot by either cutting the end off the slot or the card, you still recieve around 75% of the performance. So the actual bottleneck occurs somewhere around 3X or 2X. Therefore, anyone pondering whether their 8X wired slot is giving full bandwidth to their card - trust me it is!

So, anyone for PCI-E 3.0?!

Thanks

Evidence http://www.techpowerup.com/reviews/AMD/HD_5870_PCI-Express_Scaling/

 

Ok. We've long known we don't saturate the pci-e, but what about with a 6990 or a 590? Or the dual versions of the next cards? Hell what about this situation? Then add a third card?

But yes, we are aware of the marketing gimmick nature of the PCI-E v3

 

I think LinusTechTips on youtube did a test where he tested a 6990 with 16X, 8X and 4X and I think the 8X was within margin of error and the 4X had about 85% of the origional performance so 4X should be fine but 8X would be better.

"Hell what about this situation? Then add a third card?"

comparing a 5870 with 2 580s ?! please explain...

 

Isn't there a maximum bandwidth all cards can use across all the available slots? I haven't read into it but have always assumed there would be... which could be my mistake, but why else would certain configurations of PCI-E lanes mean you can only do 16x, 16x, 8x?

But regardless, your second post seems to imply we are reaching a reasonable time to update to 3.0. In a couple of generations you would be saying 16x v2.1 is good, but 16x 3.0 is better

 

Yeah my original idea was to put 6 GTX580s( ) into 1 motherboard - now you may be thinking-no normal mobocracy has 6 PCI-E slots with dual width. So I came up with the idea of using the PCi-E X1 slot in my mobo to house the cards.

And I know in a few generations time the 16 lanes will be finally used but really amazes me is that a GTX580 can run at about 80% of the original computations/second on rendering apps like 3d mark vantage.

 

In the 8x, 8x, 8x scenario; it's simply because the Motherboard doesn't have enough PCI-E lanes to be able to run it all at 16x, 16x, 16x. Most boards run into this limitation, because otherwise you'd need a Huge northbridge, or an even more beefed up processor to be able to manage the extra inputs and outputs.