My PCIe 3.0 GTX 780 in a PCIe 2.0 slot, how much performance am I losing?

spschillerstrom

Distinguished
Apr 5, 2013
35
0
18,530
I have a motherboard problem and some questions.

My motherboard is a Biostar Z77A, the reason I bought the board was because of an extreme price cut and a bundled 8gb stick of Corsair Vengeance RAM ($70 for all of it). Since then I have been slowly improving my PC whenever I find really good deals. Recently I replaced my Asus AMD HD 7850 with a Zotac OC GTX 780. Apparently when I removed my old GPU I damaged the PCIe 3.0 slot and was forced to put my new GPU in a 2.0 slot.

My questions are:
How much performance am I losing putting a GTX 780 in a PCIe 2.0 slot? I have read wildly different claims on the internet and I was hoping to find a little more clarity, perhaps a link to an actual benchmark or something testing the difference.

What would some of you recommend as a replacement board for $100 or less? Clearly I bought this original board on a tight budget and I want something a little more up to the task. Currently I'm loving the 780, it's still a major improvement and I'm not in a rush to re build.

Specs:
i5 3570K
Biostar Z77A mobo
Zotac GTX 780
8gb (1600) Corsair Vengeance RAM
Corsair TX650W PSU
 
Solution
I found another interesting article:
http://www.techpowerup.com/reviews/AMD/HD_5870_PCI-Express_Scaling/11.html

This suggests a HD 5870 had the following relative performance on a PCI-E 2.0 interface:
x16 100%
x8 97.76%
x4 91.91%
x1 53.48%

A GTX 780 is twice the speed of a HD 5870.
Each lane of PCI-E 3.0 has twice the bandwidth of one PCI-E 2.0 lane.
If the higher rendering performance of the card correlates to bandwidth, this would suggest the following with a GTX 780:
PCI-E 3.0 x16 100%
PCI-E 3.0 x8 or PCI-E 2.0 x16 97-98%
PCI-E 3.0 x4 or PCI-E 2.0 x8 91-92%

This lines up with the GTX 680 results remarkably well.
This suggests you get relative performance of 97-98% using a PCI-E 2.0 x16 slot compared to a PCI-E x16 slot.
Even...

spschillerstrom

Distinguished
Apr 5, 2013
35
0
18,530
I don't know if it's a 16 lane slot but the board is advertised as being crossfire compatible. Visually, besides being a different color and lower down on the board, the slots are identical.

I think I will watch this board for price drops but I think my 2.0 slot is fine for now.
http://www.newegg.com/Product/Product.aspx?Item=N82E16813157330

Do you guys think this is a good board? It has two PCIe 3.0 if I want to SLI in the future, I also like a good on board DAC, it has Realtek ALC892.

Also just a heads up if it changes the performance in regards to being 2.0 or 3.0 I run a 2560X1440 PLS display.
 
There are a few reviews around showing only a very small advantage with a GTX 680 or HD 7970 on a PCI 3.0 x16 slot compared to a PCI 2.0 x16 slot.
This is the only article I could find where they found a difference:
http://forums.evga.com/tm.aspx?m=1537816

Effectively what this article is testing is a GTX 680 on a PCI-E 3.0 x8 slot compared to a PCI-E 2.0 x8 slot.
That is effectively saying that a GTX 680 would be severely limited by a PCI 1.0 x16 slot, but not by a PCI 2.0 x16 slot.

The GTX 780 is a little faster than the GTX 680, but I wouldn't expect any significant bottleneck with the PCI-E 2.0 x16 slot.
 
I found another interesting article:
http://www.techpowerup.com/reviews/AMD/HD_5870_PCI-Express_Scaling/11.html

This suggests a HD 5870 had the following relative performance on a PCI-E 2.0 interface:
x16 100%
x8 97.76%
x4 91.91%
x1 53.48%

A GTX 780 is twice the speed of a HD 5870.
Each lane of PCI-E 3.0 has twice the bandwidth of one PCI-E 2.0 lane.
If the higher rendering performance of the card correlates to bandwidth, this would suggest the following with a GTX 780:
PCI-E 3.0 x16 100%
PCI-E 3.0 x8 or PCI-E 2.0 x16 97-98%
PCI-E 3.0 x4 or PCI-E 2.0 x8 91-92%

This lines up with the GTX 680 results remarkably well.
This suggests you get relative performance of 97-98% using a PCI-E 2.0 x16 slot compared to a PCI-E x16 slot.
Even halving this bandwidth with a PCI-E 2.0 slot operating at x8 speed (often the case with SLI), you still get relative performance of 91-92%.
 
Solution