PCI Express x4 vs. x16

[Haven]

How big of a performance hit will I take with say.... a gtx260.

 

[eidairaman1]

ok the 16x slot has 4x the bandwidth that 4x does, but apparently the story behind 4x is the 4x slot has the samebandwidth as the AGP 8x Slot.

 

[PVTCaboose1337]

Overall, if you made a card fit in a 4x slot, it would not be that big of a performance decrease.

 

[JC316]

You might lose at most, 20% of your performance and that is at the most. It would be better in a 16X for sure.

 

[Sasqui]

Here are some real world benchmarks for you:

http://www.tomshardware.com/reviews/pci-express-2-0,1915-10.html

http://www.tomshardware.com/reviews/crossfire-meets-pci-express,1761-4.html

 

[Black Panther]

Overall, if you made a card fit in a 4x slot, it would not be that big of a performance decrease.

And if you can't fit it in.... you know what you can do.....

 

[Deleted member 3]

Here are some real world benchmarks for you:

http://www.tomshardware.com/reviews/pci-express-2-0,1915-10.html

http://www.tomshardware.com/reviews/crossfire-meets-pci-express,1761-4.html

Interesting, didn't know they redid the test with a modern card. Quite different results from the first they did. Though some intensive applications do get a serious hit, many don't to my surprise.

 

[PVTCaboose1337]

And if you can't fit it in.... you know what you can do.....

A while back, someone on TPU cut down a PCI-E16x card and fit it into a 1x slot. It worked. It was just slower.

EDIT: O BTW, that photo is shopped so bad.

 

[spearman914]

GTX260 is a pretty high end card, at 4x u'll be bottlenecked a little, like JC316 saids only 20% performance loss no more.

 

[wolf]

ive used a 9600GT in a 4x slot and it seemed quite a bit worse for it, maybe just my personal experiences...

 

[DarkMatter]

Guys you are answering too fast, IMO.

First of all we need to know which mobo? PCIe 1.0? 1.1? 2.0?
Which CPU?

IT MATTERS. Because, first and foremost a mobo/chipset that only has x4 PCIe (that's what I supose) will probably be a bottleneck by itself and usually the CPU that goes with a mobo like that is slowish too. (Just assuming, take no offense.) Both can be a bottleneck.

If it's not 2.0, PCIe x4 will cripple performance A LOT, I dare to say that you would lose 75% of the performance of a GTX 260. PCIe 2 is twice as fast as PCIe 1. PCIe 1 is actually a little bit slower than equivalent AGP. i.e PCIe 1.0 8x vs AGP 8x.

If it is PCIe 2.0 you will still see a good amount of bottleneck in a GTX 260 in modern games (for older ones a GTX260 is way too much). Look at Tom's Hardware review posted above, COD4 and Crysis (and FSX but it's not the best example) are bottlenecked a lot*. Things will get worse in future games. Probably way more than 20%, though even 20% is a lot actually. If you are going to run a GTX260 at 80% all the time, just get a HD4850 or 9800gtx or even 9800 gt/hd4830, you will get same performance for less. NOW if you are going to upgrade the mobo soon it won't matter if the card is bottlenecked, as it will play games smooth until then, no problem, but it's not worth to pay more if you won't get anything in return.

*And yet, IMO those benchmarks are not well done, specially on Crysis: it's obvious they ran the test on Medium settings and without AF or AA (in an attempt to avoid GPU bottleneck) and that's a tremendous error to test what they want, actually. Lesser quality means that less memory is used and thus less data needs to be moved from system to GPU and forth, completely faking the results in Crysis and some other games (i.e. 8x AF instead of x16 in COD4).

 

[LittleLizard]

if it 2.0 version there will be liitle performance hit but if the first version, then, 15 - 20% hit

 

[DarkMatter]

if it 2.0 version there will be liitle performance hit but if the first version, then, 15 - 20% hit

On a GTX260? More, much much more.

 

[KainXS]

I ran a 8800GS @854/2000/2010, spec wise at that overclock it was near 9800GTX performance on pci-e 4x and it got hit by about 30% on a old 945GZM board, so i can only imagine that a new GTX would get a massive hit, its gotta be at least 50%

 

[spearman914]

On a GTX260? More, much much more.

PCI-E 1.0/1.1/2.0 have VERY close performance hits. 2% at it's maximum and about 500 more marks in 3dmark 06. Something like GTX295 will have a little bigger hit like >10%. In crossfire and sli configuartions pci-e 1.0/1.1/2.0 has a pretty large impact on performance however.

http://www.tomshardware.com/reviews/crossfire-meets-pci-express,1761-5.html

 

[DarkMatter]

PCI-E 1.0/1.1/2.0 have VERY close performance hits. 2% at it's maximum and about 500 more marks in 3dmark 06. Something like GTX295 will have a little bigger hit like >10%. In crossfire and sli configuartions pci-e 1.0/1.1/2.0 has a pretty large impact on performance however.

http://www.tomshardware.com/reviews/crossfire-meets-pci-express,1761-5.html

Not at all and that review doesn't help at all, the faster the cards more bottlenecked you will be. The GTX is twice as fast and will be bottlenecked at around the same performance as those in the review and that review talks about 8x and 16x, not 4x, a very big difference.

 

[Joe Public]

ive used a 9600GT in a 4x slot and it seemed quite a bit worse for it, maybe just my personal experiences...

I have as well. I tried it in both a x4 (AsRock mobo) and a mobo an x16 slot and the performance difference was quite noticeable. I can't imagine the performance hit that a GTX260 would take. I wouldn't really bother.

 

[KainXS]

http://www.tomshardware.com/reviews/pci-express-scaling-analysis,1572-8.html

http://www.tomshardware.co.uk/PCI-Express-2-0-Crossfire,review-29948-8.html

4x on even an 8800 is murder depending on what game

 

[eidairaman1]

cards only utilize bandwidth needed, not the full bandwidth, that way it has expansion so to speak, just like HDs.

 

[Jarrstin]

So if I added another 4870 512 MB to my rig to make it at 3x 4870, I should see an increase in performance?

 

[Hayder_Master]

http://www.tomshardware.com/reviews/pci-express-scaling-analysis,1572-8.html

http://www.tomshardware.co.uk/PCI-Express-2-0-Crossfire,review-29948-8.html

4x on even an 8800 is murder depending on what game

agreed that's what i think to put it too

 

[wolf]

Guys you are answering too fast, IMO.

your right, usually i dont get in depth enough and really explain my point of view, ill try harder bud.

the 9600GT in a pci-e 4x slot @ 1.1 vs 16x @ 1.1 seemed quite a bit worse off, later on i have had it in a pci-e 8x @ 2.0 slot and it seemed perfectly fine, which seems reasonable given it should equate to pci-e 16x @ 1.1

so now a question i use my GTX 295 in a pci-e 16x @ 1.1 slot and it seems fine, i know that those 2 GPU's together could probably benefit from more bandwidth.... but my vantage score is still P22k with a Q6600 @ 3.2, which seems right in line with benchmarks... (given the card is overclocked too)

thoughts?

 

[DarkMatter]

your right, usually i dont get in depth enough and really explain my point of view, ill try harder bud.

the 9600GT in a pci-e 4x slot @ 1.1 vs 16x @ 1.1 seemed quite a bit worse off, later on i have had it in a pci-e 8x @ 2.0 slot and it seemed perfectly fine, which seems reasonable given it should equate to pci-e 16x @ 1.1

so now a question i use my GTX 295 in a pci-e 16x @ 1.1 slot and it seems fine, i know that those 2 GPU's together could probably benefit from more bandwidth.... but my vantage score is still P22k with a Q6600 @ 3.2, which seems right in line with benchmarks... (given the card is overclocked too)

thoughts?

Probably you could see a benefit, but not as high as a comparable performing single GPU card IMO. My own theory about that is that dual-GPU cards don't need as much transfers because they can share a lot of the required data through the SLI bridge, when a tranfer occurs it does in a twice long burst, reducing the requests and overhead compared to a single GPU card doing the same fps. Maybe the micro-stuttering happppened because of this same thing.