PCI-E 2.0 Is enough for a 970 ?

[AGGELOS_970]

Hello!

 

A friend of mine is thinking of buying a GTX 970. His motherboard is a ASUS P8H67 and it only supports PCI-E 2.0.

Will he loose any performance ? What's the max bandwidth PCI-E 2.0 can handle ?

 

Thanks!!!

[AustinKZombie]

Probably not 

[NeatSquidYT]

Nope, PCI 2.0 won't bottleneck any graphics card on the market AFAIK

[Shroomster]

The performance loss is negliegable.

[Izaya Orihara]

Nope, PCI 2.0 won't bottleneck any graphics card on the market AFAIK

This is correct but it has got to be PCIe 2.0 16x. If it's PCIe 2.0 8x it wil not work.

[NeatSquidYT]

This is correct but it has got to be PCIe 2.0 16x. If it's PCIe 2.0 8x it wil not work.

^

[Sauron]

If it's PCIe 2.0 8x it wil not work.

 

Why? Gpus can run on any pcie slot afaik, even 1x, although under a certain bandwidth you incur in a bottleneck.

[AGGELOS_970]

Ok thanks guys!! I was pretty sure myself but I wanted to be extra sure!

[LukaP]

This is correct but it has got to be PCIe 2.0 16x. If it's PCIe 2.0 8x it wil not work.

it will work, the only time it doesnt work is trying to SLI in x4 mode. but while you cant sli in 3.0 x4 you can in 2.0 x8 which is the same bandwith

[Ryan Leech]

PCIe 2.0 x16 has enough bandwidth for any graphics card on the market.

[Jumper118]

This is correct but it has got to be PCIe 2.0 16x. If it's PCIe 2.0 8x it wil not work.

why wouldn't work with pcie 2.0 8x, if a 780ti can work at pci 2 x8 why wouldn't a 970 work.

[GloriousPain]

Pcie 2 won't bottleneck a Titan Black and is neglible on 2x295x2s (which may I remind you use PCIe based crossfire only doing extra saturation on the lanes) aslong as it's PCIe 2.0 x16 it can handle most things, it's bandwidth is more for servers with high end NIC cards...

[Lays]

Pcie 2 won't bottleneck a Titan Black and is neglible on 2x295x2s (which may I remind you use PCIe based crossfire only doing extra saturation on the lanes) aslong as it's PCIe 2.0 x16 it can handle most things, it's bandwidth is more for servers with high end NIC cards...

 

it makes almost zero difference at lower resolutions, but higher resolutions with 2+ GPU's it makes quite a difference, sometimes 10+ fps

[GloriousPain]

it makes almost zero difference at lower resolutions, but higher resolutions with 2+ GPU's it makes quite a difference, sometimes 10+ fps

ah I saw a test a while back TitanBlack @2160p PCIe 2->3 made 0 fps difference in 4 different games and a synthetic actually gave lowerscore (by like 3 points but with nearly identical fps)

And did Crossfire 295x2s @2160p Pcie 2-> made a 1-3 fps difference in the same 4 games as last time and synthetic gave an extra ~100 points with maybe 4 fps higher 

[Jumper118]

it makes almost zero difference at lower resolutions, but higher resolutions with 2+ GPU's it makes quite a difference, sometimes 10+ fps

that is interesting. so when i'm running dual 970's on pcie 2 in 16x/8x at 4k i will be loosing performance? 

[Lays]

ah I saw a test a while back TitanBlack @2160p PCIe 2->3 made 0 fps difference in 4 different games and a synthetic actually gave lowerscore (by like 3 points but with nearly identical fps)

And did Crossfire 295x2s @2160p Pcie 2-> made a 1-3 fps difference in the same 4 games as last time and synthetic gave an extra ~100 points with maybe 4 fps higher 

 

there's a boat load of things if you google image search "SLI 5760x1080 pci-e 2.0 vs pci-e 3.0" and things like that showing performance differences, at really high res with multiple GPU's it tends to make a big difference from what I've seen

 

  

[Lays]

that is interesting. so when i'm running dual 970's on pcie 2 in 16x/8x at 4k i will be loosing performance? 

 

Maybe, some places I've seen show 1-2 FPS, while other tests I've seen make a big deal, so I'm not entirely sure.

[Jumper118]

Maybe, some places I've seen show 1-2 FPS, while other tests I've seen make a big deal, so I'm not entirely sure.

in that case when i get them i'll test them, in my mpower first on pcie 2, then get an i7 for my soc force and test them with pcie 3. that can be my project for 2015 . i'll be interesting to see the difference. i will fined that 1-2 fps very important in benchmarks.

[Izaya Orihara]

why wouldn't work with pcie 2.0 8x, if a 780ti can work at pci 2 x8 why wouldn't a 970 work.

It will work however it's a pretty sevre bottleneck. 

[i_build_nanosuits]

pcie 2.0 @ 16X is fine for even a GTX 980...a titanZ...an R9 295X2...no noticeable performance loss.

[TriBeard]

I think amd said they tested crossfire on R9 290x's down to 8x/8x and they saw no real performance losses, and that's on AMD cards where all the crossfire talk is done over the PCI-e bus rather than a crossfire/SLI Bridge. Sure, it's not ideal, but it wouldn't stop me from buying a second card if that's what was holding me back. 

[i_build_nanosuits]

I think amd said they tested crossfire on R9 290x's down to 8x/8x and they saw no real performance losses, and that's on AMD cards where all the crossfire talk is done over the PCI-e bus rather than a crossfire/SLI Bridge. Sure, it's not ideal, but it wouldn't stop me from buying a second card if that's what was holding me back. 

correct, pcie 2.0 at 4x speed is where the real bandwidth bottleneck starts to occur on most mid to high end GPU's

[pavachan]

Information about this seemed sketchy, different sources have said different things, so I ended up testing this myself.

 

On an i7 2600K @ 4.5, Z68 motherboard (pci-e 2.0 only) and a GTX 980 OC'd to 1500 and a dedicated 750Ti PhysX card, here's what I found with Batman : Arkham Origins. (fps is listed as min, avg, max)

*Everything is on Max settings, including Physics on High and TXAA High unless stated otherwise.

 

PCI-E x8/x8 - 1080p + Dedicated PhysX : 69, 92, 123

PCI-E x8/x8 - 1080p + GTX 980 PhysX   : 57, 75, 99

PCI-E x8/x8 - 1080p + 980 Normal PhysX : 82, 126, 247

----------------------------------------------------------------------------------

PCI-E x8/x8 - 4K + Dedicated PhysX : 43, 60, 96 (No AA)

PCI-E x8/x8 - 4K + GTX 980 PhysX : 35, 44, 56 (No AA)

PCI-E x8/x8 - 4K + 980 Normal PhysX : 43, 65, 177 (No AA)

----------------------------------------------------------------------------------

PCI-E x16 - 1080p : 67, 87, 117

PCI-E x16 - 1080p+Normal PhysX : 93, 130, 185

PCI-E x16 - 4K : 43, 55, 70 (No AA)

PCI-E x16 - 4K Normal PhysX : 47, 71, 189 (No AA)

 

 

It would appear that with PhysX @ 1080p, the extra bandwidth comes to very good use, as the single card numbers are almost the same as the x8 with dedicated PhysX.

Reducing the PhysX load @ 1080p still yields better numbers at x16 as opposed to x8, and while not as large a gap, still meaningful, especially the min. fps.

Going up to 4K,  with High PhysX, the same large improvements are seen at x16, and with normal PhysX, the gap remains the same.

 

According to these results, which are quite limited (only 1 game, only 1 system), I am still inclined to get rid of the 750Ti, mostly because with x16 PCI-E 2.0 I will get almost the same numbers as with a dedicated card @ x8, and because newer, much more demanding games like Dragon Age Inquisition (which like many, doesn't use dedicated PhysX), will produce an overall noticeable higher performance.

+10 fps, +15 fps is meaningful, either when hanging around the ~50 fps range or when trying to really benefit from a 120Hz+ monitor.

 

As for 4K, I am either hitting a GPU bottleneck or a PCI-E bottleneck. I cant really know which since I don't have a PCI-E 3.0 motherboard and CPU handy, though there is still improvement.

[Samsterstorm]

You will only lose like 5-10 FPS max it really will not effect it. What CPU is he running?