3070 Running at pci 3.0 x4?!

[jordzie]

Hi there
I have the graphics card in the top slot without using any NVME on my motherboard and still achieving x4.
I have confirmed the gpu is running at x4 speed. The speed of my graphics card should 16gb/s but only 4 ish atm.
Any help would be appreciated.
I have the latest bios for my card and motherboard.
I have a:
3950x ryzen
Asus Crosshair VI hero wifi ac
3070 evga xc3 ultra

 

[silentbogo]

There are few scenarios that may've happened (I'll list those sorted by severity):
1) GPU is lose in a slot, or sagging too much
2) PCIe slot is dirty
3) You've damaged some slot pins during installation (it happens occasionally on PCIe and RAM slots, and in 99% cases it's not even user's fault)
4) You knocked off some caps near PCIe of the card, which may limit it's capabilities to x4 only
5) Motherboard issues.
6) Your card is defective.

That 3070 is a big and heavy card. I'm more inclined towards PCIe slot damage or sagging.

BTW, a little offtopic question...
Have you tried downgrading BIOS on your MoBo? I think they still have ver.7201 listed in downloads, which at least initially had PCIe 4.0 enabled.

 

[jordzie]

There are few scenarios that may've happened (I'll list those sorted by severity):
1) GPU is lose in a slot, or sagging too much
2) PCIe slot is dirty
3) You've damaged some slot pins during installation (it happens occasionally on PCIe and RAM slots, and in 99% cases it's not even user's fault)
4) You knocked off some caps near PCIe of the card, which may limit it's capabilities to x4 only
5) Motherboard issues.
6) Your card is defective.

That 3070 is a big and heavy card. I'm more inclined towards PCIe slot damage or sagging.

BTW, a little offtopic question...
Have you tried downgrading BIOS on your MoBo? I think they still have ver.7201 listed in downloads, which at least initially had PCIe 4.0 enabled.

Blew on the card and slot like a n64 cartridge and reseated and everything miraculously started working! thanks!!
Was Actually weighing trying that version 7201 due to pcie 4.0 but i think for me stability is probably the best on latest bios.
BUT HOW DARE THEY REMOVE PCI 4.0 FOR X370!
You know your stuff bogo.

 

[Mussels]

The PCI-E 4.0 testing revealed a lot of errors according to AMD, leading to stuff like GPU driver crashes and M.2 drives vanishing/reporting errors

It was very beta for no real gain, so IMO not worth it (apart from a few select M.2 NVME drives, it doesnt matter. Even a 3090 can handle 16x 3.0 fine)

edit: and now i've got the phrase "dirty, dirty slot" stuck in my head when i think about this

 

[Caring1]

Blew on the card and slot like a n64 cartridge and reseated and everything miraculously started working! thanks!!

That brings back memories

 

[silentbogo]

You know your stuff bogo.

"I know my stuff" is the only line on my resumé )))

BUT HOW DARE THEY REMOVE PCI 4.0 FOR X370!

The PCI-E 4.0 testing revealed a lot of errors according to AMD, leading to stuff like GPU driver crashes and M.2 drives vanishing/reporting errors

That was same-old marketing bullshit to sell new boards. Given the amount of time that passed between rolling-out the update, and quickly removing it, I'd say it was a mistake on AMD's side, who didn't want to add PCIe4.0 support off the very start, but somehow forgot to disable it (or tell OEMs to disable it).
Kinda like what happened w/ non-functional mining blocker in recent Nvidia dev. driver release.

 

[8BitZ80]

Blew on the card and slot like a n64 cartridge and reseated and everything miraculously started working! thanks!!
Was Actually weighing trying that version 7201 due to pcie 4.0 but i think for me stability is probably the best on latest bios.
BUT HOW DARE THEY REMOVE PCI 4.0 FOR X370!
You know your stuff bogo.

I recommend a spray bottle with a 50/50 IPA and water mixture. Spray it onto a microfiber towel and wipe down the contacts with it. Often when I have used memory sticks causing boot loops, that technique helps to get the system running properly again.

 

[TheLostSwede]

That was same-old marketing bullshit to sell new boards. Given the amount of time that passed between rolling-out the update, and quickly removing it, I'd say it was a mistake on AMD's side, who didn't want to add PCIe4.0 support off the very start, but somehow forgot to disable it (or tell OEMs to disable it).
Kinda like what happened w/ non-functional mining blocker in recent Nvidia dev. driver release.

Sorry, but this is NOT marketing BS and it NOT about selling more boards.
First of all, 300-series boards were never designed for anything but PCIe 3.0, even the few 400-series boards that met the PCIe 4.0 certification standard would've caused problems for AMD, as some boards would've passed, some wouldn't. What do you think the end user backlash on that would've been?
The thing with certification is that you either pass, or you don't, there's no in-between, because then it's not a certification.
The OP has an X370 board and I hope you're not seriously implying that AMD should've made the board makers design in PCIe 4.0 support back when that board was launched?

 

[Mussels]

"I know my stuff" is the only line on my resumé )))



That was same-old marketing bullshit to sell new boards. Given the amount of time that passed between rolling-out the update, and quickly removing it, I'd say it was a mistake on AMD's side, who didn't want to add PCIe4.0 support off the very start, but somehow forgot to disable it (or tell OEMs to disable it).
Kinda like what happened w/ non-functional mining blocker in recent Nvidia dev. driver release.

I dont agree with that - it came out as an AGESA beta and the mobo manufs publicly released it, fast.
With the amount of variation between all the revisions of these boards, what would have happened if certain boards just black screened on POST with a 3090 or 6x00 card once PCI-E 4 was mainstream?

I would have preferred a hidden "force PCI-E 4.0" option for power users, but i can tell people would just screw it up
The PCMR group on FB is full of people with issues from using PCI-E 3.0 risers with 4.0 cards and boards, and i cant see how this would be any different, tech support nightmare

 

[silentbogo]

Sorry, but this is NOT marketing BS and it NOT about selling more boards.

I dont agree with that - it came out as an AGESA beta and the mobo manufs publicly released it, fast.
With the amount of variation between all the revisions of these boards, what would have happened if certain boards just black screened on POST with a 3090 or 6x00 card once PCI-E 4 was mainstream?

I preemptively apologize for off-topic, but PCIe 4.0 is near-identical to 3.1 from electrical perspective.
The only differences are:
1) Tighter tolerances for signal loss (mostly handled on RX/TX side)
2) Higher susceptibility to crosstalk (needs minor adjustments to ground pins and via placements)
On short distances neither is going to cause major issues (assuming board partners didn't skimp on parts and followed PCIe 3.1 spec). Worst case - it'll fall back to gen3.
Think of it as running 10GbE over Cat5e in an apartment or overclocking your RAM. It's not "compliant", but works just the same.
There are many exceptions (boards with older redrivers or PCIe MUXes etc), but the principle still stands.
Nothing drastic to cause mass-outrage towards AMD, or bash Gigabyte and Asus for deploying a new feature. And no one said "all boards". AMD only mentioned 400-series boards, and board partners narrowed it down to only certain models, and later added a couple X370 (like OPs Crosshair 6, or GB X370 K7). ASUS even went as far as doing proper compliance testing, hence only a dozen or so motherboards got on their list. Main reason: x570 too expensive, B550 delayed again, and neither AMD or board partners weren't sure if AsMedia will ever go through with it (remember, the miracle update came out in early 2019, while B550 didn't come to fruition 'till summer of 2020).

Call me a conspiracy theorist, but please tell me that my rambling makes less sense than $80 B550 board which has even worse layout than its predecessor and has decoupling caps nearly twice as far away as older boards, and no recommended adjustments for crosstalk reduction, is more fit to drive PCIe 3.0 bandwidth than last-gen flagship solution. Heck, some "lucky" B550 owners have already encountered PCIe 4.0 issues on their boards (long POST, fallback to 3.0), and that's supposedly a fully-compliant design....

The OP has an X370 board and I hope you're not seriously implying that AMD should've made the board makers design in PCIe 4.0 support back when that board was launched?

It's not about designing the board with PCIe4.0, it's about the existing design being capable of 16GT/s.
OPs board with 7201 update in action:

 

[Mussels]

"Worst case - it'll fall back to gen3."

But it doesnt, all the people with PCI-E 4.0 systems on 3.0 risers prove that - so many issues with crashes, black screens and so on. Loooooots of people complaining about their new AMD and Nv cards with problems have turned out to be resolved when you change to 3.0 or ditch the riser - this isn't something i see so much on TPU but i see it multiple times a week in the FB groups i'm in where uneducated people slap together the prettiest parts they can find and wonder why shit crashes and black screens and scream abusive rants on FB only to find a simple ass fix

The tighter tolerances matter

 

[silentbogo]

But it doesnt, all the people with PCI-E 4.0 systems on 3.0 risers prove that

It doesn't prove anything. People with PCIe3.0 systems and 3.0 risers have problems as well, that just comes with the territory. For now, let's just talk about more conventional "I buy a card and plug it into my motherboard" situation, since that's what 99.9% of people do.

And regarding PCIe 4.0 and risers - that's a huge and complex topic for another day. To make it short: connectors are the weakest point in the equation, and having multiple connectors along the way only makes it worse. That applies not only to chinese crap slapped together from defective HDMI cables and sticky tape, but also more expensive PCB risers in servers.
On conventional boards most sources list a PCB route length of 7-7.5" (that's mobo+expansion card total), at which your PCIe lanes should stay within 4.0 tolerance.

Loooooots of people complaining about their new AMD and Nv cards with problems have turned out to be resolved when you change to 3.0 or ditch the riser

Also, lots of people complain about XMP/AMP not working or being unstable, causing black screens and BSODs etc. ...but no one is bashing Intel/AMD and board partners for implementing this feature.

If it was a simple case of "4.0 doesn't work on some boards so we won't do it for all boards", then AMD, Gigabyte, and ASUS wouldn't try so hard to erase any trace or mention of PCIe 4.0. Just look at GB F40 update, or ASUS 7201 update and find a single mention of PCIe 4.0. Then, move your eyes a bit higher to the next update and try to find any mention of removing PCIe 4.0 support.... AsRock doesn't even have 5.60 update. Even changelogs are gone from official sites. Back in 2019 it was all there.

 

[Mussels]

Yes it DOES prove something - that issues with PCI-E quality control (risers, non tested BIOS, etc) *causes problems*
People slap in a new GPU, get black screen crashes and blame the GPU - not the riser, not an experimental feature added in via BIOS they may not have heard of

 

[R-T-B]

But it doesnt, all the people with PCI-E 4.0 systems on 3.0 risers prove tha

That's because mining risers are shit quality and barely meet the 3.0 spec. I'd be surprised if some of them pass the 2.0 spec. They just aren't pumping a lot of data so they USUALLY work.

 

[Mussels]

i;m not talking mining risers dude i'm talking about the vertical mounting in cases that most people seem to want in new builds

 

[jordzie]

Personally i still dont understand why Asus didnt update for pci 4.0 and resizable bar. Its all a money grab.

 

[R-T-B]

i;m not talking mining risers dude i'm talking about the vertical mounting in cases that most people seem to want in new builds

Pretty much all risers sadly are of the same signal quality, mining or not.