x16/x16 VS x8/x8 SLI Results...awkward

[heliax2112]

So I ordered an SLI bridge that is advertised to fit 1st and 4th slot. For SLI to work on my motherboard at x16/x16 it is advertised for 1st and 4th slot...well as you can see at the bottom of the first picture that I took the specs too literally and assumed they were compatible.(My motherboard has an extra gap between slots 1 and 2 that caused this issue.) I was going to return the bridge, then I said screw it, let me see if moving up the card up 1 slot and running it at x8/x8 speeds reduces my performance, the following I did not expect.

 

PC Settings for both tests

 

So if you are worried about buying a processor that doesn't support 40 pci lanes  (Example: 5820k) or a motherboard with just 1x16 slot, don't worry because it doesn't matter.

 

 

5930k @ 4.5

DDR4 @ 3200

980 ti's @1480

[dragoon20005]

that is really weird

 

but 3Dmark only tell one part of the story

 

can you test with games and see the fps difference?

[MegaDave91]

This is mostly common knowledge (on LTT)  that one is no worse off using x8 x8 over x16 x16.

[thekeemo]

that is really weird

 

but 3Dmark only tell one part of the story

 

can you test with games and see the fps difference?

this is actually not weird and normal

[Virtual01]

Running 8x instead of 16x will give you 1-2% less performance at absolute worst. Most of the time there is no difference.

[dragoon20005]

this is actually not weird and normal

well i do know x8 x8 configs will still be ok

 

but never seen the scores to be higher than running full x16 mode

 

also 3DMark only shows part of the story

 

games may perform differently with x8 vs x16 configs

[Najuno]

It isn´t surprising for me to see that. I told and will tell everybody PCIe3.0 x16 x16 is not necessary nowadays because there´s literally no GPU on the market that is fast enough to make use of it. PCIe 3.0 x8 x8 is completely sufficient.

 

The reason for a 40 native lane CPU is either a lot of cards to connect with higher lane demand or more than 2 GPUs, because this is where the 5930K and the 5960X start to shine. 3 or 4way SLI/CF systems. And of course if you are an OCer to achieve higher results in 3D Mark´s Physics score.

[Sam Z Man]

Who doesn't know this?

[Enderman]

that is really weird

 

but 3Dmark only tell one part of the story

 

can you test with games and see the fps difference?

How is that weird?

 

EVERYONE knows that you dont need 3.0 x16 or even 3.0 x8 with less than 3/4 GPUs

it makes zero difference, a graphics card does NOT use even PCIe 2.0 x8 speeds

[dragoon20005]

How is that weird?

 

EVERYONE knows that you dont need 3.0 x16 or even 3.0 x8 with less than 3/4 GPUs

it makes zero difference, a graphics card does NOT use even PCIe 2.0 x8 speeds

i expect a drop in performance with x8 maybe by a few hundred point

 

but an increase?

 

maybe i have yet to see more results

[Sam Z Man]

OP just so you know even if you used a 4X slot on a GPU it only gives you 5-10% performance hit with flagship GPU's (Fury X, Titan X) the only reason they upgrade the bandwidth of PCI interfaces is to keep up with server bandwidth requirements.

[thekeemo]

well i do know x8 x8 configs will still be ok

 

but never seen the scores to be higher than running full x16 mode

 

also 3DMark only shows part of the story

 

games may perform differently with x8 vs x16 configs

Its a margin of error

[dragoon20005]

Its a margin of error

yea which is why i need OP to test in games as well

[Sam Z Man]

i expect a drop in performance with x8 maybe by a few hundred point

 

but an increase?

 

maybe i have yet to see more results

∆2.5% is a margin of error because your PC is always doing small tasks in the background, this had been tested before by people who make a living wiring PC articles and running Benchmarks and it always shows that 3.0 8x vs 16x makes no difference with consumer applications.

[thekeemo]

yea which is why i need OP to test in games as well

That wouldnt help.

Im saying the extra few FPS are coming from say an update that was downloading when he was doing x16 x16 or maybe a background malware checked

[dragoon20005]

∆2.5% is a margin of error because your PC is always doing small tasks in the background, this had been tested before by people who make a living wiring PC articles and running Benchmarks and it always shows that 3.0 8x vs 16x makes no difference with consumer applications.

i do aware the margin of error

 

which is why in the first post i made

 

i would like OP to test with actual games

[MegaDave91]

i would like OP to test with actual games

It's already been tested to death.

[Enderman]

i expect a drop in performance with x8 maybe by a few hundred point

 

but an increase?

 

maybe i have yet to see more results

increase = inconsistency

 

hes obviously not in safe mode with minimal background processes

 

a few hundred points is within margin of error

[Sam Z Man]

i do aware the margin of error

which is why in the first post i made

i would like OP to test with actual games

The only reason they had to go from a 15.75GB/s to 31.5GB/s interface with PCI express 3.0 was servers wanted network/raid cards with more connectivity.

[dragoon20005]

The only reason they had to go from a 15.75GB/s to 31.5GB/s interface with PCI express 3.0 was servers wanted network/raid cards with more connectivity.

technologies for server will then trickle down to consumers or mainstream products

 

although its still overkill for the normal person to use PCI-E SSD with average 1GB/s Read/write speeds

[Cheddle]

I have experience that says PCIE 2.0 x16/x16 provides a bottleneck when running SLI 980ti - compared to PCIE 3.0 16x/8x. PCIE 2.0 could not go over 96fps (hard FPS wall) when displaying 4k (even subsampling render at 1080 provided the same result) tried different SLI bridges with no change.  approx 10% loss of firestrike graphics score between the two when displayin at 1080p.

 

there IS a bottleneck below PCIE 3.0 8x and it seems to be only at very high resolutions.

[Sam Z Man]

technologies for server will then trickle down to consumers or mainstream products

 

although its still overkill for the normal person to use PCI-E SSD with average 1GB/s Read/write speeds

Really it's mainly because the network cards that run the fiber optic cable that connect cities, countries and contents need more bandwidth, the dual 100Gb/s networking cards of today are impossible to run on a 16x PCIe 2.0 connection.

[dragoon20005]

Really it's mainly because the network cards that run the fiber optic cable that connect cities, countries and contents need more bandwidth, the dual 100Gb/s networking cards of today are impossible to run on a 16x PCIe 2.0 connection.

haha well i still complaining my 1GB fiber is not enough for my "ahem" stuffs i get

[ybriK]

This is pretty common knowledge in the GPU technology, surprised not many people actually know about it. Same thing will happen with PCIe 3.0 vs 2.0, barely any performance difference on which slot the GPU goes to. Probably at higher surround display systems it might have an impact.