PCI 4.0 16x slot reported as a PCI 5.0 8x with the AMD 9070 XT Reaper GPU

[Teeinae]

As mentioned in the title both the AMD adrenalin software and the GPU-Z software report the Bus type and setting as PCI Express 5.0 x8 which should be impossible when my motherboard only supports PCIe 4.0. I know that PCI 5.0 x8 = PCI 4.0 16x so there might be some weird scuff with the computer detecting the bandwidth of a PCI 4.0 x16 and confusing it with a PCI 5.0 x8 slot. Just ask for more information if needed, I'm new to these forms so am a little out of place! (Picture 1: GPU-Z, Picture 2: AMD adrenalin software)

I should note that in the first picture of GPU-Z it says PSI 5.0 x16 but when I hover my mouse over it, it says PCI 5.0 8x

 

[Caring1]

It's reporting on the GPU not the motherboard bus speed.

 

[Marcus L]

Shall we guess what motherboard you are using? Who needs details anyway

 

[Teeinae]

Sorry, as said I'm not used to this kind of forums! I'm using a MAG Tomahawk b650 wifi!

It's reporting on the GPU not the motherboard bus speed.

Thank you for the information! Will trust you on that one and just keep going on with my day!

 

[Marcus L]

I mean pcie 5.0 x8 is the same as pcie 4.0 x16, so in theory it's not wrong, but yea you look to be running GPU in ~PCIE slot 4x-16

 

[LabRat 891]

As mentioned in the title both the AMD adrenalin software and the GPU-Z software report the Bus type and setting as PCI Express 5.0 x8 which should be impossible when my motherboard only supports PCIe 4.0. I know that PCI 5.0 x8 = PCI 4.0 16x so there might be some weird scuff with the computer detecting the bandwidth of a PCI 4.0 x16 and confusing it with a PCI 5.0 x8 slot. Just ask for more information if needed, I'm new to these forms so am a little out of place! (Picture 1: GPU-Z, Picture 2: AMD adrenalin software)

I should note that in the first picture of GPU-Z it says PSI 5.0 x16 but when I hover my mouse over it, it says PCI 5.0 8x

Assuming this isn't a simple mis-reading...
Navi 48's Integral PCIe Switch is probably handshaking a full 4.0x16 Upstream, and 5.0x8 to the GPU Downstream.

I mean pcie 5.0 x16 is the same as pcie 4.0 x8, so in theory it's not wrong, but yea you look to be running GPU in ~PCIE slot 4x-16

Yup.
x8 32GT/s vs x16 16GT/s, both work out to 256GT/s

PCIe Switches are capable of such 'generation-agnostic bandwidth re-allocation'.
Ex. A Gen3x16 'Squid' NVMe Quad Expander, in a legacy system: 1.1x16<->Gen3x4 x4. Ea. 3.0x4 endpoint (NVMe) gets a non-dedicated full 3.0x4 bandwidth.

Navi III had integral Upstream and Downstream PCIe Switching. -Makes sense that Navi IV does too.

 

[Marcus L]

Assuming this isn't a simple mis-reading...
Navi 48's Integral PCIe Switch is probably handshaking a full 4.0x16 Upstream, and 5.0x8 to the GPU Downstream.

Which, PCIe Switches are capable of: Ex. 1.1x16<->Gen3x4 x4, each 3.0x4 endpoint getting non-dedicated full bandwidth

View attachment 389826View attachment 389829

Navi III had integral Upstream and Downstream PCIe Switching. -Makes sense that Navi IV does too.

Motherboard only has PCIE 4x16, think SW is misreading it even though the specs and bandwidth of 5x8 and 4x16 are the same, maybe @W1zzard can chime in

Motherboard only has PCIE 4x16, think SW is misreading it even though the specs and bandwidth of 5x8 and 4x16 are the same, maybe @W1zzard can chime in

Card is capable of pcie-5, motherboard only pcie-4, sure I had a similar thing when I upgraded motherboard from b450 (pcie3) to b550 (pcie4), regardless, there is no difference in performance

 

[AsRock]

Try running a game and see if it changes.

 

[LabRat 891]

Motherboard only has PCIE 4x16, think SW is misreading it even though the specs and bandwidth of 5x8 and 4x16 are the same, maybe @W1zzard can chime in

@Teenej could download and run HWinfo64, and see if my theory holds any water

Free Download HWiNFO Sofware | Installer & Portable for Windows, DOS

Start to analyze your hardware right now! HWiNFO has available as an Installer and Portable version for Windows (32/64-bit) and Portable version for DOS.

www.hwinfo.com

If I'm anywhere near correct, their "ATI/AMD Navi - Switch Upstream in PCIe" will show something like this

and their "ATI/AMD Navi - Switch Downstream in NBIF" would look something like this

 

[W1zzard]

@Teenej could download and run HWinfo64, and see if my theory holds any water

Free Download HWiNFO Sofware | Installer & Portable for Windows, DOS

Start to analyze your hardware right now! HWiNFO has available as an Installer and Portable version for Windows (32/64-bit) and Portable version for DOS.

www.hwinfo.com

If I'm anywhere near correct, their "ATI/AMD Navi - Switch Upstream in PCIe" will show something like this
View attachment 389830

and their "ATI/AMD Navi - Switch Downstream in NBIF" would look something like this
View attachment 389831

Yeah, please try that, I think @LabRat 891's theory is correct, but more data is needed

 

[Event Horizon]

Here's what I get on my B650 motherboard.

 

[Teeinae]

@Teenej could download and run HWinfo64, and see if my theory holds any water

Free Download HWiNFO Sofware | Installer & Portable for Windows, DOS

Start to analyze your hardware right now! HWiNFO has available as an Installer and Portable version for Windows (32/64-bit) and Portable version for DOS.

www.hwinfo.com

If I'm anywhere near correct, their "ATI/AMD Navi - Switch Upstream in PCIe" will show something like this
View attachment 389830

and their "ATI/AMD Navi - Switch Downstream in NBIF" would look something like this
View attachment 389831

Hello! Sorry for taking so long, I didn't get a notification on my gmail. I am very new to these advanced levels of hardware and the most difficult thing I have done on these levels was overclocking my CPU per core, so I am sorry if I am not giving adequate amount of information. Here is a picture of what I hope you're asking for if not I could really use a more tutorial-y step by step to the information you want!

 

[eidairaman1]

As mentioned in the title both the AMD adrenalin software and the GPU-Z software report the Bus type and setting as PCI Express 5.0 x8 which should be impossible when my motherboard only supports PCIe 4.0. I know that PCI 5.0 x8 = PCI 4.0 16x so there might be some weird scuff with the computer detecting the bandwidth of a PCI 4.0 x16 and confusing it with a PCI 5.0 x8 slot. Just ask for more information if needed, I'm new to these forms so am a little out of place! (Picture 1: GPU-Z, Picture 2: AMD adrenalin software)

I should note that in the first picture of GPU-Z it says PSI 5.0 x16 but when I hover my mouse over it, it says PCI 5.0 8x

Can you give a detailed list of your pc specs please

 

[W1zzard]

Hello! Sorry for taking so long, I didn't get a notification on my gmail. I am very new to these advanced levels of hardware and the most difficult thing I have done on these levels was overclocking my CPU per core, so I am sorry if I am not giving adequate amount of information. Here is a picture of what I hope you're asking for if not I could really use a more tutorial-y step by step to the information you want!

No worries, we understand

That's the right list, but we need the PCI Express section details view for all the items in the "path". So select each node and screenshot, showing both left and right section

 

[Teeinae]

No worries, we understand

That's the right list, but we need the PCI Express section details view for all the items in the "path". So select each node and screenshot, showing both left and right section

Got it!

 

[LabRat 891]

Got it!

Intriguing. Not what I was expecting.

I'm unsure precisely how HWinfo derives and generates its data, but GPU-z seems to be reporting the same.
-You do have a Gen5 capable SoC (not board), and AFAIK this (and RTX5k) are the first consumer-available Gen5 devices that use more than 4 lanes.

At this point I'm thinking:
A. Mis-reporting; maths mathing where it shouldn't math.
B. The new AMD card is commanding the SoC NBIO to negotiate gen5 (why x8 though? dunno.)

 

[Marcus L]

why x8 though? dunno

Because the pcie slot in question is pcie 4@16x that's the max bandwidth it could physically allocate which translates to pcie 5@8x

 

[LabRat 891]

Because the pcie slot in question is pcie 4@16x that's the max bandwidth it could physically allocate which translates to pcie 5@8x

That would hold true, in the case of mis-reporting. 16.0GT/s x16 = 32.0GT/s x8.

In the case of the NBIO actually running 32.0GT/s... Not so much.
Lanes are lanes, there are 16x physically attached to the SoC via the IoD's NBIO (IIRC, feel free to correct)
If the SoC's NBIO is running 32.0GT/s to the PCIe slot, it 'should' be a full x16 lanes.

But, if you've ever had a lame x16 riser, dirty golden fingers, or mis-seated a card, you can get 'lanes falling off'.
If, only the x16 slot's first 8-lanes are w/in spec for 5.0, maybe that's all it could 'handshake'?


TBQH though,
'miscalculation' of the Generation and Lane count negotiated, seems most likely. (after all, these are amongst the first >x4 Gen5 devices)

BTW, how did the 990 Evo 'negotiate' between 5.0x2 and 4.0x4?
mis-reading, or not; I suspect that may be related. (PCI-SIG standards, and such)

 

[timsdf]

Hey, I wanted to register finally here from a long time reader.

MSI B650 tomahawk seems to run at full PCIE 5.0 16x speed on 9070.

55.24GB/s on 3dmark PCIE speed test and cpu-z cc below.

 

[LabRat 891]

Hey, I wanted to register finally here from a long time reader.

MSI B650 tomahawk seems to run at full PCIE 5.0 16x speed on 9070.

55.24GB/s on 3dmark PCIE speed test and cpu-z cc below.

*If* that's true, then OP probably needs to re-seat his card. -It's only handshaking 8-lanes.
(I don't think ASPM/PCIe power savings de-negotiates lanes, just clockspeed/generation?)

 

[timsdf]

My system reads 9070 always PCIE 5.0 x16 32GT/s in drivers, gpu-z, cpu-z, hwinfo and 3dmark stress tests.

All win11 power savings disabled. Guessing MSI just "forgot" to disable 5.0 like Asrock on some boards.

 

[Teeinae]

*If* that's true, then OP probably needs to re-seat his card. -It's only handshaking 8-lanes.
(I don't think ASPM/PCIe power savings de-negotiates lanes, just clockspeed/generation?)

Re-seat as in just plug it out and then in again in the same PCIE Slot? Also how is it even possible that a PCIE 4.0 16x slot can run at PCIE 5.0 16x speeds? Or maybe I'm understanding the conversation wrong?

 

[LabRat 891]

Re-seat as in just plug it out and then in again in the same PCIE Slot?

Correct.
More times than I ever care to recount, I've had a no POST (PCI/AGP) or partial PCIe handshake, resolved by 'taking it out and putting it back in again'

Also how is it even possible that a PCIE 4.0 16x slot can run at PCIE 5.0 16x speeds? Or maybe I'm understanding the conversation wrong?

Assuming timsdf's info is true and correct, it would have to do w/ the fact that I/O (largely) originates on the SoC (from the IoD).

The First M.2 slot and First x16 slot on AM5 motherboards are almost universally connected directly to the CPU(SoC).
Ryzen 7000 and Ryzen 9000 AM5 non-APUs have full PCIe 5.0 support In-SoC.

Apparently, Gen5 cards are 'overriding' the Mobo Manufacturer's Firmware settings and directly-commanding the SoC's NBIO into Gen5 (32.0GT/s) mode.
Which, as long as the trace layout/length is w/in spec, may well run reliably @ gen5 speeds.

[Note: back in the AM4 days, *many* Gen3 A320, B350, X370, B450, and X470 motherboards had short-lived PCIe 4.0 exposure. IIRC, The B550A was a B450 board w/ firmware allowing Gen4 PCIe.
For most non 500series AM5 boards, this capability was actively removed w/in 1-2 AGESA updates from Gen4's introduction. Not every board was 'built' to be reliable in 16GT/s (4.0) operation, and it also harmed the value of B550 and X570.]


Edit:
The MSI Mag B650 Tomahawk boards use an SMT PCIe x16 CPU-connected slot, not a through-hole mount design as Gen1-4 used.

This is an unnecessary implementation for strictly Gen3 and Gen4 connectivity.

Very clearly, at some point in time *after* the PCB was designed and laid out, MSI/AMD decided that B650 boards did not need Gen5 capability.
The board is 'built' to be capable of it. Also likely: multiple 'tier' boards, share the same base PCB layout.
[Tangential: ASrock's 'new' B650 X3D board, also has Gen5 x16, IIRC]

 

[Marcus L]

I think this is a SW reporting error rather than the CPU/SOC forcing a PCIE-4 MB into PCIE5 mode as you would also be able to add in a fully fledged PCIE-5 NVME drive and obtain full speed which I doubt you can, maybe should reach out to MSI if there is a rep registered on TPU to investigate further, though wouldn't be getting my hopes up thinking a PCIE-4 MB can just upgrade to PCIE-5, it's all speculation at best at this point, CPU-Z is still showing 32GT/s and I would trust CPU-Z reporting more than 3DMark

 

[Teeinae]

A little late as of now to do anything hardware related but tomorrow, I will try to just re-plug my GPU and see if that works and if that doesn't work, I guess I will contact MSI about it like Marcus said.