290X Crossfire on an x16 and a x4 PCI-E 2.0. Is that bad?

[crystal6tak]

Motherboard is the GA-970A-UD3.

 

Just realized the 2 slot is an x16 and an x4. Will there be an issue of it slowing down the cards? Hawaii GPU doesn't have an crossfire bridge and relies entirely on PCI-E and plus it's x4... Should I be worried?

[AlwaysFSX]

Yes, you just took a hatchet to the second card's huevos.

[Jobbe03]

1 in an 8x slot would be acceptable. But 4x?.......

[Altecice]

I think Crossfire is able to run down at x4, SLI requires x8 minimum... give it a shot but you may be lacking in performance on that card.

[Darkman]

If it was a x8 slot, I wouldn't worry, but that's a x4...

[SuperPug]

Why oh why would you possibly spend so much money on GPUs and then put it in a budget board? Also to answer your question - Nvidia cards just straightup refuse to work on 4x bandwith so that should give you a pretty good idea.

[WinNut]

A PCI-E v2 x4 slot is equivalent to a PCI-E v3 x2 slot. That's a big bottleneck right there.

[Diederik_Gr]

From what i've heard, it'll work, but there will be a performance hit that keeps the 2nd card from its full potential.

[TheUnknown]

Let me get his straight forward.You are wasting the second gpu.

[hoggydoggy]

you will probably see a small difference in performance compared to if you were to use 16x 8x, but can you not do 8x 8x or does your motherboard only have a 4x pci slot free?

[Glenwing]

A PCI-E v2 x4 slot is equivalent to a PCI-E v3 x2 slot. That's a big bottleneck right there.

 

It'll be a bottleneck, but it won't be that big.

[spwath]

It'll be a bottleneck, but it won't be that big.

This.

I did not notice a diferance when I moved my card from x16 to x4 (2.0)

[crystal6tak]

It'll be a bottleneck, but it won't be that big.

 

This.

I did not notice a diferance when I moved my card from x16 to x4 (2.0)

Do you know any rough numbers? Like is it 10% or 20% worse than no bottleneck?

[Paul_VB]

It will function, just not nearly at 100%

I know from personal experience... im running two xfired 7850's with one on a PCI-E 16x 2.0 (which is totally acceptable) and the second one on a PCI-E 4x 1.1 (lol). Gotta love dat Z68 chipset though

[Sakiz105]

i was running 2 r9 280x on an h97 chipset (the motherboard was "crossfire ready").The difference from x8 x8 and mine x16 x4 was like 5%.But you are talking about 2 r9 290x so the loss of power will be marginally higher (i'm estimating around 12 to 15% or even higher in some games)

[Briggsy]

you may get lower results in benchmarks, but for most games I can't see there being much of a difference. You're really just at the mercy of data being transferred from system memory to GPU memory, which for pci-e 2.0 4x is about 2 Gigabytes/second.

 

This comparison should put your mind at ease: (it seems that in this case the AMD card relies less on pci-e bandwidth than Nvidia for single monitor 1200p gaming, which makes sense given that the 680 in this example has less video memory)

 

[Glenwing]

Do you know any rough numbers? Like is it 10% or 20% worse than no bottleneck?

 

I lost about 10% on my own tests on PCIe 2.0 x4.  With crossfire added in it may get up to 15-20% but I didn't test that.

[Cheddle]

Motherboard is the GA-970A-UD3.

 

Just realized the 2 slot is an x16 and an x4. Will there be an issue of it slowing down the cards? Hawaii GPU doesn't have an crossfire bridge and relies entirely on PCI-E and plus it's x4... Should I be worried?

 

AMD suggest minimum PCIE 3.0 8x/8x for r9 290 crossfire. you will have better performance than with a single card however this will be limited.  for two reasons.

 

1) not enough bandwidth for the cards to share info

2) high latency between the cards (assuming the 970a runs the 4x slot via the chipset)

 

I experienced lower performance and audio stutters when running my r9 290 crossfire via a 16x/4x pcie 3 system. see below results 

 

 - this one was a z97 Gigabyte board with 16x/4x PCIe 3.0

 

 - this onewas a z87 ASUS board with 8x/8x PCIe 3.0

 

 

I also went to about 16k gpu points when I ran pcie 16x/4x on the Z87 board

 

I went from about 95 fps to 99 fps in Valley

 

I noticed no difference in Tomb raider, but farcry 3 was a huge difference

you may get lower results in benchmarks, but for most games I can't see there being much of a difference. You're really just at the mercy of data being transferred from system memory to GPU memory, which for pci-e 2.0 4x is about 2 Gigabytes/second.

 

This comparison should put your mind at ease: (it seems that in this case the AMD card relies less on pci-e bandwidth than Nvidia for single monitor 1200p gaming, which makes sense given that the 680 in this example has less video memory)

 

 

 

r9 290's are a little different... they use xDMA to share data entireley over the PCI bus. Depending on mobo this can be via the chipset - introducing huge latency in accessing memory.  The above test does not consider brdigeless nor PCIe over chipset.

[Dabombinable]

So I take it that the second PCIe x16 (actually x4 gen 2) slot on my dead h87m pro would have been effectively useless then for crossfire (which was part of its marketing).

 

 

you may get lower results in benchmarks, but for most games I can't see there being much of a difference. You're really just at the mercy of data being transferred from system memory to GPU memory, which for pci-e 2.0 4x is about 2 Gigabytes/second.

 

This comparison should put your mind at ease: (it seems that in this case the AMD card relies less on pci-e bandwidth than Nvidia for single monitor 1200p gaming, which makes sense given that the 680 in this example has less video memory)

 

Thanks for the chart, by the looks of things my p5k-vm will be just fine with my gtx 970 (pcie v1-full x16), and will be relevant for quite some time.

[Cheddle]

So I take it that the second PCIe x16 (actually x4 gen 2) slot on my dead h87m pro would have been effectively useless then for crossfire (which was part of its marketing).

fine for crossfire cards with a bridge. The r9 290 is the first card to use xDMA and the first to require (basically) an Nvidia SLI certified board.

AMD still let you use 16x/4x for r9 290s but it's not optimal.

nvidia are just like "nope" and when you hack it to use hyperSLI you lose similar levels of performance

[rentaspoon]

I think this is the video

[Dabombinable]

so xDMA is were cards in crossfire actually use the link to share data, instead of relying heavily on PCIe slots? Might be worth looking into AMD for graphics if Nvidia are going to be slack. (I know about some of the current 'issues' people have with AMD cards, but just like all of my previous computers, if it works and does what I want when I want, nothing else matters ie. I'm not a fanboy)

 

Edit: looked it up, it probably will still affect performance on motherboards with a cheaply made pcie setup, because there is no real excuse for not having a x16 sized slot capable of delivering full x16, instead of 2x or 4x with all other pcie slots disabled (aka, there goes my wifi) other than a slightly higher cost. Perhaps if the bridge was dedicated for certain tasks to take some of the load off the pci slots?

[SkilledRebuilds]

What of the EXTRA latency that it brings that destroys performance smoothness? Meaning non-native 2.0 4x provided from the chipset not the CPU.

 

I get the native PCIE 2.0 4x is mostly fine, but what about the 2ndary chipset latency?

[Cheddle]

What of the EXTRA latency that it brings that destroys performance smoothness? Meaning non-native 2.0 4x provided from the chipset not the CPU.

These is a measured increase in frame time varience in metro ll benchmark

I get the native PCIE 2.0 4x is mostly fine, but what about the 2ndary chipset latency?

[SkilledRebuilds]

 

What of the EXTRA latency that it brings that destroys performance smoothness? Meaning non-native 2.0 4x provided from the chipset not the CPU.

These is a measured increase in frame time varience in metro ll benchmark

I get the native PCIE 2.0 4x is mostly fine, but what about the 2ndary chipset latency?

 

I've been reading around on it a bit more, there are a few cases of dual-gpu being worse off than single gpu because of thie variance in frame time delivery

One of the reasons I now am grabbing a new 8x8 @ 3.0/3.0 mobo first before another 290 for my existing 16x/4x <-3.0 and non-native 2.0