UL Releases PCI Express Feature Test For 3DMark Ahead of PCIe 4.0 Hardware

[crazyeyesreaper]

With PCI-Express 4.0 graphics cards and motherboards soon to arrive, UL has released their PCI Express feature test for 3DMark. This latest addition has been designed to verify the bandwidth available to the GPU over a computer's PCI Express interface. To accomplish this, the test will make bandwidth the limiting factor for performance and does so by uploading a large amount of vertex and texture data to the GPU for each frame. The end goal is to transfer enough data over the PCIe 4.0 interface to thoroughly saturate it. Once the test is complete, the end result will be a look at the average bandwidth achieved during the test.





While real-life gaming performance is unlikely to be limited by PCIe bandwidth, the test does provide a repeatable and reliable way to measure PCIe bandwidth across generations and thereby the performance of different hardware configurations. The PCI Express feature test is available now in 3DMark Advanced Edition and 3DMark Professional Edition and is compatible with any system that has a Direct X 12 compatible discrete graphics card. The test will not run on systems using integrated graphics, and on multi-GPU systems, only the primary GPU will be tested. External GPU enclosures are not supported either.

View at TechPowerUp Main Site

 

[Wavetrex]

Not sure how much PCI-e bandwidth this test needs, but it could certainly use more Youtube video bandwidth !
Horrible encoding is horrible.

 

[Basard]

What is it supposed to be? Couldn't they have made it look like grass or something?

 

[Steevo]

Fairly certain W1zz has proven that by the time bandwidth is an issue the card will be obsolete. Multiple times.

Perhaps it might reveal weakness in other components that could cause stuttering

 

[rutra80]

It's scaled to do 1 FPS per 1GB/s of bandwidth.
It starts with fewer big polys and then goes up, on my gen 3 PCIe it looks quite nice at the end, on gen 4 it should be quite convincing field of grain

 

[Basard]

It's scaled to do 1 FPS per 1GB/s of bandwidth.
It starts with fewer big polys and then goes up, on my gen 3 PCIe it looks quite nice at the end, on gen 4 it should be quite convincing field of grain

A field of grain floating on a turbulent ocean?

 

[Jism]

We should have some more PCI-E related, benchmark tools. Because there's enough gains to be scored when oc'ing the PCI-E bus. GPU-Z for example reports the Bus width / link width but not the actual bandwidth. If we OC the bus to 110Mhz that should increase the bandwidth alot more then running on stock?

I know oc'ing the PCI-E bus helps the videocard but can cause issues for the NIC, S-ata controller and what more. But we've done it in the past to sqeeze a few more FPS out of a OC contest.

 

[Crackong]

I/O , Storage , Connectivity , those are the PCI-E 4.0 capabilities we want to see.

Modern graphics cards are proven to be effective enough to run on 3.0x8 with single digit performance loss.
The only "4.0 beats 3.0" scenario is external display card, but it is a very niche market .
Please stop doing these stupid PCI-E 4.0 graphics benchmarks.

 

[TheLostSwede]

I/O , Storage , Connectivity , those are the PCI-E 4.0 capabilities we want to see.

Modern graphics cards are proven to be effective enough to run on 3.0x8 with single digit performance loss.
The only "4.0 beats 3.0" scenario is external display card, but it is a very niche market .
Please stop doing these stupid PCI-E 4.0 graphics benchmarks.

Please stop voicing your stupid opinions? It's a test for crying out loud, it's been designed to show that there's a difference for the sake of having a test to show there's a difference. People love showing off the fact that they have fast hardware. What's so wrong with that? Yes, it might be a mostly pointless test, but stupid..?

 

[EarthDog]

Pcie 4.0 vs 3.0 today... like hooking up a fire hose to your garden spicket.

 

[Crackong]

Please stop voicing your stupid opinions? It's a test for crying out loud, it's been designed to show that there's a difference for the sake of having a test to show there's a difference. People love showing off the fact that they have fast hardware. What's so wrong with that? Yes, it might be a mostly pointless test, but stupid..?

There is no point of creating a benchmark with fancy visuals to test bandwidth.
Something looked like a AIDA64 memory test would be fancy enough for a bandwidth test.
And I have to remind you, this test isn't a PCI-E test because it needs a GPU to operate.
For example you cannot test your fancy PCI-E 4.0 x4 m.2 slot with it.
You can call it a "GPU data rate test" but PCI-E ? Hell NO.

It is a pointless test at the beginning and calling it a PCI-E bandwidth test is utterly stupid.

 

[Hotobu]

Fairly certain W1zz has proven that by the time bandwidth is an issue the card will be obsolete. Multiple times.

Can you provide links to this please? I've been trying to find this out.

 

[TheLostSwede]

There is no point of creating a benchmark with fancy visuals to test bandwidth.
Something looked like a AIDA64 memory test would be fancy enough for a bandwidth test.
And I have to remind you, this test isn't a PCI-E test because it needs a GPU to operate.
For example you cannot test your fancy PCI-E 4.0 x4 m.2 slot with it.
You can call it a "GPU data rate test" but PCI-E ? Hell NO.

It is a pointless test at the beginning and calling it a PCI-E bandwidth test is utterly stupid.

Of course there is, maybe not as an actual test, but you know as well as I do, that companies like to show things off in a visual way, as does overclockers and people that like benching. It might not server an actual purpose for the test itself, but it serves a purpose, even though you might not find it useful.

Can you provide links to this please? I've been trying to find this out.

Most recent one here https://www.techpowerup.com/review/nvidia-geforce-rtx-2080-ti-pci-express-scaling/

 

[TheoneandonlyMrK]

Anyone tried this on an actual pciex slot yet at home, surely someone has a Intel server pciex4 test to show?.

 

[INSTG8R]

Well I gave it a go. It was weird and I didn’t let it finish. But it’s was like 5FPS and my card was just idle during the test 5.9GBs was the highest I saw before I quit.

 

[TheLostSwede]

Well I gave it a go. It was weird and I didn’t let it finish. But it’s was like 5FPS and my card was just idle during the test 5.9GBs was the highest I saw before I quit.

You running your card in x8 mode?

 

[INSTG8R]

You running your card in x8 mode?

Actually funny that, since moving from Fury to Vega my slot runs full speed at all times...

 

[Jism]

X8 vs X16 for graphics does'nt offer that big of a change. It's more within the enterprise market that needs as much as bandwidth as possible. Think of this: https://www.anandtech.com/show/1449...d-epycbased-system-with-108-intels-ruler-ssds

108 intel ruler SSD's in there, offering a total capacity of 864GB and speeds of 30GB a second read/write. That's ten times the speeds of an high-end SSD for consumers right now.

 

[TheoneandonlyMrK]

My PC sustained 7GB/s on pciex 3

 

[ValenOne]

+12GB/s on X299

 

[srsbsns]

14.5 on Crosshair VII with a Radeon Vii

 

[arm1]

I7 8700 + rx 580 + Asrock z390 extreme4 - 13.98 GB/s

 

[candle_86]

hmm im gonna toss my GTX 550 ti on an nforce 4 board to see how bad PCIe 1.0 does