Jump to content

Testing a x16 graphics card in x8 results in no performance difference?

Go to solution Solved by Eigenvektor,
35 minutes ago, CoolJosh3k said:

Is this to be expected, or did I make some naïve mistake in testing?

To be expected, see:

https://www.techpowerup.com/review/nvidia-geforce-rtx-3080-pci-express-scaling/

 

It makes much more of a difference on some low end GPUs that have (too) little VRAM and have a smaller number of PCIe lanes to begin with, e.g.

https://www.techpowerup.com/review/amd-radeon-rx-6500-xt-pci-express-scaling/

Because I may need to use my second x8 PCIe slot, I did some testing with Time Spy for my RX 6800. However I noticed no appreciable difference in score.

 

I did this testing using the bifurcation setting in my bios and verified the lanes with HwInfo.

 

I expected at least a loss of 10% performance or something, but I only recorded a loss of less than 0.5%. Is this to be expected, or did I make some naïve mistake in testing?

 

My scores were:

 

@x16:

16,596

16,640

16,645

 

@x8:

16,589

16,584

16,569

 

My system for those curious is:

AMD Ryzen 7 5800X3D
Gigabyte Radeon RX 6800 GAMING OC 16G
G.skill Trident Z Neo 4x16GB DDR4-3600 CL14

Gigabyte X570S AORUS MASTER

Link to comment
Share on other sites

Link to post
Share on other sites

Set to PCIe 4.0 or 3.0? 4.0 x8 is the same bandwidth as 3.0 x16, and you need a 3090 to see a 1-3% hit (depends on resolution) at that bandwidth. 

Intel HEDT and Server platform enthusiasts: Intel HEDT Xeon/i7 Megathread 

 

Main PC (jank until UPS sorts my claim for mobo damage in shipping) - i9 7980XE @per-core-OC 4.2-4.5GHz - EVGA CLC 280 w/2x Noctua Redux P14s - EVGA X299 Dark - 4x8GB 3200Mhz CL16 DDR4 - Nvidia 2060 Super FE w/GPU only 280mm loop - 1TB 970 Evo + 250GB 960 Evo + 1TB Crucial MX500 - EVGA 1600W T2 - Corsair 750D Airflow - Windows 11

 

Proxmox Rig - i7 6950X - Noctua NH-L12S - EVGA X99 Micro 2 - 4x4GB 2800MHz DDR4 - Radeon PRO WX2100 - 3x 2TB Seagate Barracuda + 1TB WD Blue + 1TB WD SE - EVGA 1000W G3 - Fractal Design Node 804 - Proxmox 7.2

 

Rest of the HEDT collection - Xeon X5675/EVGA X58 Classified 4-Way SLI - i7 4930K/EVGA X79 Dark - EVGA Super-Record 2 (not currently booting, need to sort that out eventually)

 

Daily Devices: 2015 15" MacBook Pro running macOS Monterey (Work Laptop) - iPhone 13 Pro Max - Apple Watch S3

 

Other Misc Devices: 2012 MacBook Pro 13" running Linux Mint - Mac Pro 1,1 (2x dual cores as I haven't swapped the quads in yet, 32GB DDR2 and a 512MB HD4870) running El Cap/Lion/sometimes other Mac OS X versions - iPod Video (5th gen)

Link to comment
Share on other sites

Link to post
Share on other sites

4 minutes ago, CoolJosh3k said:

Is this to be expected, or did I make some naïve mistake in testing?

That's expected, in games GPUs are very rarely bus bottlenecked. 3090 Tis only really drop 3% at the most when using x8 compared to x16 Gen 4. 

 

If you dropped down to gen 3 you might see a bit more substantial of a difference, but with a 6800 I wouldn't really expect any different. 

 

6 minutes ago, CoolJosh3k said:

G.skill Trident Z Neo 4x16GB DDR4-3600 CL14

Just curious, are you even able to get XMP to run on that CPU? The IMC usually gives up when trying to run speeds that aggressive with quad rank memory, 3466MT/s is usually the most you can hope for with 4 dual rank DIMMs on a Ryzen 3000/5000 CPU (16GB 3600 CL14 DIMMs are one of the few left that are guaranteed dual rank). 

Link to comment
Share on other sites

Link to post
Share on other sites

35 minutes ago, CoolJosh3k said:

Is this to be expected, or did I make some naïve mistake in testing?

To be expected, see:

https://www.techpowerup.com/review/nvidia-geforce-rtx-3080-pci-express-scaling/

 

It makes much more of a difference on some low end GPUs that have (too) little VRAM and have a smaller number of PCIe lanes to begin with, e.g.

https://www.techpowerup.com/review/amd-radeon-rx-6500-xt-pci-express-scaling/

Remember to either quote or @mention others, so they are notified of your reply

Link to comment
Share on other sites

Link to post
Share on other sites

56 minutes ago, Zando_ said:

Set to PCIe 4.0 or 3.0? 4.0 x8 is the same bandwidth as 3.0 x16, and you need a 3090 to see a 1-3% hit (depends on resolution) at that bandwidth. 

I didn’t test 3.0 @ x16, since I was able to test 4.0 @ x8 using the bifurcation setting.

Link to comment
Share on other sites

Link to post
Share on other sites

53 minutes ago, RONOTHAN## said:

That's expected, in games GPUs are very rarely bus bottlenecked. 3090 Tis only really drop 3% at the most when using x8 compared to x16 Gen 4. 

 

If you dropped down to gen 3 you might see a bit more substantial of a difference, but with a 6800 I wouldn't really expect any different. 

 

Just curious, are you even able to get XMP to run on that CPU? The IMC usually gives up when trying to run speeds that aggressive with quad rank memory, 3466MT/s is usually the most you can hope for with 4 dual rank DIMMs on a Ryzen 3000/5000 CPU (16GB 3600 CL14 DIMMs are one of the few left that are guaranteed dual rank). 

Everything is very much stable and verified with a 1:1 infinity fabric clock. Nothing special going on either.

Link to comment
Share on other sites

Link to post
Share on other sites

If your gpu is 4.0 and your motherboard is pci 5.0.  Putting the gpu in top slot, would that auto change to pci 4.0 because that’s what the gpu is?

 

If so would that mean all of your pci 5.0 lanes are disabled?  So really you can’t use any?

CPU:                       Motherboard:                Graphics:                                 Ram:

i7-8700k@5.3ghz   Asrock z370 Taichi         1080 FTW hybrid@2164mhz    Trident Z 3200mhzc14

Link to comment
Share on other sites

Link to post
Share on other sites

20 minutes ago, Shzzit said:

If your gpu is 4.0 and your motherboard is pci 5.0.  Putting the gpu in top slot, would that auto change to pci 4.0 because that’s what the gpu is?

 

If so would that mean all of your pci 5.0 lanes are disabled?  So really you can’t use any?

No idea, but I’d expect that slot to run a the highest the card can do.

Link to comment
Share on other sites

Link to post
Share on other sites

30 minutes ago, Shzzit said:

If your gpu is 4.0 and your motherboard is pci 5.0.  Putting the gpu in top slot, would that auto change to pci 4.0 because that’s what the gpu is?

 

If so would that mean all of your pci 5.0 lanes are disabled?  So really you can’t use any?


 It works at 4.0.

 

What I’ve noticed with the new AM5 boards:

 

Those X670E boards, even the top models mention that the PCI Express slot will work at 5.0 x8 speed if you populate the M2 slots. That means that a 4090 which is 4.0 will run at 4.0 x8 on a mobo like that that uses an M2 SSD on the CPU lanes. 
 

Example X670e Aorus Extreme:

 

Expansion Slots
CPU:
  1. 1 x PCI Express x16 slot, supporting PCIe 5.0* and running at x16 (PCIEX16)
    * Actual support may vary by CPU.
    * The M2B_CPU and M2C_CPU slots share bandwidth with the PCIEX16 slot. When the M2B_CPU or M2C_CPU slot is populated, the PCIEX16 slot operates at up to x8 mode.

 

 

Link to comment
Share on other sites

Link to post
Share on other sites

52 minutes ago, PaulFCB said:

That means that a 4090 which is 4.0 will run at 4.0 x8 on a mobo like that that uses an M2 SSD on the CPU lanes. 

Only if the b and c M.2 slots are populated, so if you have a m.2 only in the top(A) slot the GPU slot is still x16

If you want me to answer, please use the quote function or tag me. I dont get notified unless you do

Link to comment
Share on other sites

Link to post
Share on other sites

 Which is bad because why would I buy such a mobo and limit myself to 1 M2 SSD /use more but have my GPU lanes cut to half.

 

 Rn I’m using 2 on Z690 Master and planning for a while to get a 3rd. But 1x is on the CPU lanes and the rest are chipset lanes, Extreme wants to be the flagship for Giga and this solution sucks bad.

 

 

Link to comment
Share on other sites

Link to post
Share on other sites

Because there are not infinity lanes.

 

When designing a board, you need to make choices as to where you allocate those lanes. That particular board vendor has elected to bifurcate the main 16x CPU link to provide extra m.2's. Other manufacturers (or other boards from the same manufacturer) will take alternative approaches and distribute things differently.

 

Not every X670E board will be configured in the same manner.

 

As a random example, the X670E ACE from MSI has 4 m.2 slots, one from the CPU and three from the chipset. And they have instead bifurcated the GPU connection to one of the other PCIe slots.

Link to comment
Share on other sites

Link to post
Share on other sites

21 minutes ago, Aragorn- said:

Because there are not infinity lanes.

 

When designing a board, you need to make choices as to where you allocate those lanes. That particular board vendor has elected to bifurcate the main 16x CPU link to provide extra m.2's. Other manufacturers (or other boards from the same manufacturer) will take alternative approaches and distribute things differently.

 

Not every X670E board will be configured in the same manner.

 

As a random example, the X670E ACE from MSI has 4 m.2 slots, one from the CPU and three from the chipset. And they have instead bifurcated the GPU connection to one of the other PCIe slots.


Reading m.2 instruction on motherboards gives me brain cancer. 

CPU:                       Motherboard:                Graphics:                                 Ram:

i7-8700k@5.3ghz   Asrock z370 Taichi         1080 FTW hybrid@2164mhz    Trident Z 3200mhzc14

Link to comment
Share on other sites

Link to post
Share on other sites

5 hours ago, CoolJosh3k said:

I didn’t test 3.0 @ x16, since I was able to test 4.0 @ x8 using the bifurcation setting.

Yeah, point is that's the same bandwidth. 4.0 x8 = 3.0 x16, and you need a 3090 or higher to have a barely measurable performance hit in games (time spy runs closer to a game than a workstation load or something).  

Intel HEDT and Server platform enthusiasts: Intel HEDT Xeon/i7 Megathread 

 

Main PC (jank until UPS sorts my claim for mobo damage in shipping) - i9 7980XE @per-core-OC 4.2-4.5GHz - EVGA CLC 280 w/2x Noctua Redux P14s - EVGA X299 Dark - 4x8GB 3200Mhz CL16 DDR4 - Nvidia 2060 Super FE w/GPU only 280mm loop - 1TB 970 Evo + 250GB 960 Evo + 1TB Crucial MX500 - EVGA 1600W T2 - Corsair 750D Airflow - Windows 11

 

Proxmox Rig - i7 6950X - Noctua NH-L12S - EVGA X99 Micro 2 - 4x4GB 2800MHz DDR4 - Radeon PRO WX2100 - 3x 2TB Seagate Barracuda + 1TB WD Blue + 1TB WD SE - EVGA 1000W G3 - Fractal Design Node 804 - Proxmox 7.2

 

Rest of the HEDT collection - Xeon X5675/EVGA X58 Classified 4-Way SLI - i7 4930K/EVGA X79 Dark - EVGA Super-Record 2 (not currently booting, need to sort that out eventually)

 

Daily Devices: 2015 15" MacBook Pro running macOS Monterey (Work Laptop) - iPhone 13 Pro Max - Apple Watch S3

 

Other Misc Devices: 2012 MacBook Pro 13" running Linux Mint - Mac Pro 1,1 (2x dual cores as I haven't swapped the quads in yet, 32GB DDR2 and a 512MB HD4870) running El Cap/Lion/sometimes other Mac OS X versions - iPod Video (5th gen)

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share


×