Poor PCI-E Bandwidth Performance

[bigdot]

I’ve recently got a 3090 and I’ve been running some benchmarks. The results are pretty good, but I’m concerned that some performance is being left on the table.


I’ve been running some 3D Mark bench marks and the PCI-E Bandwidth benchmark seems a bit off. It’s measuring in at about 6.4gbps where I’d expect it to be roughly double that.


It’s in the top PCI-E Gen 3 16x slot.


I’ve also got an Intel Optant pci-e hardrive further down in a 4x slot which I thought might be causing the issue. It’s the system drive, so I can’t just take it out to test.

So I have a few questions:


1) Is 6.4gbps a limiting speed for my set up

2) Do you think that the Optane drive is causing the issue and if not, what could it be?


Any advice on this would be appreciated.

More spec:
GPU: 3090
Motherboard: MSI MEG Z390 ACE (MS-7B12)
CPU: i9-9900K
System drive: 280 GB INTEL SSDPED1D280GA
Memmory: 64gb (4x16) Corsair DDR4 @ 2,144 MHz

 

[VincentHanna]

Run the GPU render test in GPU-Z.

It'll tell you if it's running at 3.0 x16 or not

 

[Shiari]

You've never said which 3090 you have. For me, with a KFA2, I'm power limited. I can squeeze out a little more performance vs the out of the box performance by essentially undervolting it (allowing the clocks to go a little higher at the same power budget) but I can't exceed the 350W power limit without hardware or firmware modification.

 

[tyler_jrb]

Z390 only has 16 lanes for pcie devices so most likely the pcie drive is taking lanes away.

will it hinder performance. Most likely as the 3090 is a pcie 4.0 gen card. I don’t think there is much difference between 3.0 16x and 4.0 16x but 3.0 8x vs 16x could be a different story as it is a powerful card.

You’d be better off with a decent m.2 nvme.

 

[Armageus]

As per the rest of your motherboard manual:



With a 2nd PCI-E Card installd your GPU will only ever run at PCI-E 3.0 8x (~8GB/s) so your 6.4GB/s is reasonable after overheads

 

[Kaapstad]

With some old software PCI-E 4.0 can make a very big difference.

One I really enjoyed doing was smashing 8 Packs old 4 card LN2 cooled CPU score on Unigine Valley using a single 3090.

It was only possible using PCI-E 4.0 as 3.0 was knocking nearly 1000 points off the score.



The funny part about it is this is not my fastest card, that's in an intel PCI-E 3.0 PC that is nowhere quick enough for the Valley bench.

 

[bigdot]

Thank you for your responses. The card is an MSI Suprim X btw.


The slot was indeed running at 8x and in fact I found there is a board inspector feature in the bios that lets you see what is plugged in and what it’s running at.


So, I decided to take out the optane drive and install Windows on an M.2 that was previously an empty scratch drive.


Running the benchmarks again, the GPU bandwidth bench did indeed (roughly) double.


However,


There was no significant gain for Timespy: 17412 > 17469


And a modest gain in port royal: 13234 > 13403


Do the scores seem a little under par for this set up at stock?

 

[Kaapstad]

Scores will increase most where the FPS are very high.