Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.debian.user > #260554 > unrolled thread

AMD GPU hard lockups

Back to article view | Back to linux.debian.user

Contents

  AMD GPU hard lockups Celejar <celejar@gmail.com> - 2023-08-01 17:50 +0200
    Re: AMD GPU hard lockups piorunz <piorunz@gmx.com> - 2023-08-01 18:00 +0200
      Re: AMD GPU hard lockups Dan Ritter <dsr@randomstring.org> - 2023-08-01 18:40 +0200
        Re: AMD GPU hard lockups Celejar <celejar@gmail.com> - 2023-08-01 19:20 +0200
      Re: AMD GPU hard lockups Celejar <celejar@gmail.com> - 2023-08-01 19:20 +0200
        Re: AMD GPU hard lockups piorunz <piorunz@gmx.com> - 2023-08-01 19:30 +0200
          Re: AMD GPU hard lockups Celejar <celejar@gmail.com> - 2023-08-01 20:10 +0200
            Re: AMD GPU hard lockups piorunz <piorunz@gmx.com> - 2023-08-01 20:20 +0200
              Re: AMD GPU hard lockups Celejar <celejar@gmail.com> - 2023-08-02 03:30 +0200
                Re: AMD GPU hard lockups zithro <slack@rabbit.lu> - 2023-08-02 23:20 +0200
                  Re: AMD GPU hard lockups Celejar <celejar@gmail.com> - 2023-08-02 23:40 +0200
                    Re: AMD GPU hard lockups piorunz <piorunz@gmx.com> - 2023-08-02 23:40 +0200
                      Re: AMD GPU hard lockups Celejar <celejar@gmail.com> - 2023-08-03 01:30 +0200
                        Re: AMD GPU hard lockups zithro <slack@rabbit.lu> - 2023-08-03 02:10 +0200
                          Re: AMD GPU hard lockups Celejar <celejar@gmail.com> - 2023-08-03 16:40 +0200
                      Follow the specs :) (was Re: AMD GPU hard lockups) zithro <slack@rabbit.lu> - 2023-08-03 01:40 +0200
                        Re: Follow the specs :) (was Re: AMD GPU hard lockups) Celejar <celejar@gmail.com> - 2023-08-03 03:00 +0200
                        Re: Follow the specs :) (was Re: AMD GPU hard lockups) gene heskett <gheskett@shentel.net> - 2023-08-03 03:30 +0200
                      Re: AMD GPU hard lockups Cindy Sue Causey <butterflybytes@gmail.com> - 2023-08-03 05:00 +0200

#260554 — AMD GPU hard lockups

Hello,

I have a system running Debian unstable with an AMD RX-570. It has been
working fine for a while, but recently, anything that uses the more
advanced features of the GPU causes the system to hard lockup: black
screen, no response to keyboard, no network connectivity.

I'm not sure exactly which functionality of the GPU causes this:
ordinary web browsing, development work, etc. never cause problems, but
games, the Unigine Heaven benchmark, and even glmark2 invariably do,
sometimes immediately, sometimes after a few seconds or minutes.

I'm not sure exactly when this began: I hadn't been using the system
for any of the problematic tasks for a while.

I've tried looking in the logs. Running 'journalctl -b -1' after a
lockup generally shows nothing. I've tried to catch the error with
'tail -F /var/log/syslog', and most of the time I see nothing (just the
hang, with no warning in the log), but once I caught this:

2023-08-01T10:41:23.531381-04:00 lucy kernel: [38532.241396] gmc_v8_0_process_interrupt: 15 callbacks suppressed
2023-08-01T10:41:23.531394-04:00 lucy kernel: [38532.241401] amdgpu 0000:02:00.0: amdgpu: GPU fault detected: 147 0x06508401 for process heaven_x64 pid 14771 thread heaven_x64:cs0 pid 14792
2023-08-01T10:41:23.531395-04:00 lucy kernel: [38532.241407] amdgpu 0000:02:00.0: amdgpu:   VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x060004CA
2023-08-01T10:41:23.531396-04:00 lucy kernel: [38532.241408] amdgpu 0000:02:00.0: amdgpu:   VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x02084001
2023-08-01T10:41:23.531396-04:00 lucy kernel: [38532.241409] amdgpu 0000:02:00.0: amdgpu: VM fault (0x01, vmid 1, pasid 32778) at page 100664522, read from 'TC7' (0x54433700) (132)
2023-08-01T10:41:23.531397-04:00 lucy kernel: [38532.241429] DMAR: DRHD: handling fault status reg 2
2023-08-01T10:41:23.531398-04:00 lucy kernel: [38532.241433] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xfbff5000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531399-04:00 lucy kernel: [38532.241438] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xfbfd1000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531400-04:00 lucy kernel: [38532.241442] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xfbffd000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531409-04:00 lucy kernel: [38532.241445] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xfffe0000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531409-04:00 lucy kernel: [38532.241449] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xfbffd000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531410-04:00 lucy kernel: [38532.241453] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xffff8000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531411-04:00 lucy kernel: [38532.241456] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xfffd8000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531412-04:00 lucy kernel: [38532.241460] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xfbbc1000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531412-04:00 lucy kernel: [38532.241460] pcieport 0000:00:02.0: AER: Uncorrected (Fatal) error received: 0000:00:02.0
2023-08-01T10:41:23.531413-04:00 lucy kernel: [38532.241464] DMAR: [DMA Write NO_PASID] Request device [02:00.0] fault addr 0xeffc0000 [fault reason 0x05] PTE Write access is not set
2023-08-01T10:41:23.531414-04:00 lucy kernel: [38532.241477] pcieport 0000:00:02.0: PCIe Bus Error: severity=Uncorrected (Fatal), type=Transaction Layer, (Receiver ID)
2023-08-01T10:41:23.531415-04:00 lucy kernel: [38532.241482] pcieport 0000:00:02.0:   device [8086:6f04] error status/mask=00040000/00000000
2023-08-01T10:41:23.531415-04:00 lucy kernel: [38532.241487] pcieport 0000:00:02.0:    [18] MalfTLP                (First)
2023-08-01T10:41:23.531416-04:00 lucy kernel: [38532.241492] pcieport 0000:00:02.0: AER:   TLP Header: 00001000 020024ff aaa800c0 00000000
2023-08-01T10:41:23.531417-04:00 lucy kernel: [38532.241500] [drm] PCI error: detected callback, state(2)!!

I've found similar reports online, e.g.:

https://unix.stackexchange.com/questions/327730/what-causes-this-pcieport-00000003-0-pcie-bus-error-aer-bad-tlp
https://forums.linuxmint.com/viewtopic.php?t=380748
https://gitlab.freedesktop.org/drm/amd/-/issues/2358

But I'm really not clear whether these represent the same problem, or
are just different variations of a more general driver / firmware
problem. (I'm assuming it's software / firmware, since everything
worked fine previously, although I suppose it's possible that something
physical has broken in the hardware.)

Any ideas?

-- 
Celejar

[toc] | [next] | [standalone]

#260555

On 01/08/2023 16:44, Celejar wrote:

> Any ideas?
> 
Revert to Debian Stable and check there. I have AMD Radeon card and it's 
100% stable on Stable, and it had issues last year back when bookworm 
was in Testing.

-- 
With kindest regards, Piotr.

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system
⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org/
⠈⠳⣄⠀⠀⠀⠀

[toc] | [prev] | [next] | [standalone]

#260556

piorunz wrote: 
> On 01/08/2023 16:44, Celejar wrote:
> 
> > Any ideas?
> > 
> Revert to Debian Stable and check there. I have AMD Radeon card and it's
> 100% stable on Stable, and it had issues last year back when bookworm was in
> Testing.

If that doesn't work, try a new power supply. Video cards and
CPUs are typically the #1 and #2 draw on power, so power
problems show up there first.

-dsr-

[toc] | [prev] | [next] | [standalone]

#260559

On Tue, 1 Aug 2023 12:12:01 -0400
Dan Ritter <dsr@randomstring.org> wrote:

> piorunz wrote: 
> > On 01/08/2023 16:44, Celejar wrote:
> > 
> > > Any ideas?
> > > 
> > Revert to Debian Stable and check there. I have AMD Radeon card and it's
> > 100% stable on Stable, and it had issues last year back when bookworm was in
> > Testing.
> 
> If that doesn't work, try a new power supply. Video cards and
> CPUs are typically the #1 and #2 draw on power, so power
> problems show up there first.

Thanks, but I can't easily / conveniently do that. I don't have one
handy, and the machine is an HP Z440, which I think may have trouble
using standard power supplies anyway.

-- 
Celejar

[toc] | [prev] | [next] | [standalone]

#260560

On Tue, 1 Aug 2023 16:56:37 +0100
piorunz <piorunz@gmx.com> wrote:

> On 01/08/2023 16:44, Celejar wrote:
> 
> > Any ideas?
> > 
> Revert to Debian Stable and check there. I have AMD Radeon card and it's 
> 100% stable on Stable, and it had issues last year back when bookworm 
> was in Testing.

I'm confused - current stable *is* bookworm, so if you had issues with
bookworm, why would reverting to (current) stable help? Do you mean
oldstable / bullseye? I actually did test using kernel 6.1.0-10-amd64
from stable / bookworm, but the problem remained.

-- 
Celejar

[toc] | [prev] | [next] | [standalone]

#260561

On 01/08/2023 18:16, Celejar wrote:
> On Tue, 1 Aug 2023 16:56:37 +0100
> piorunz <piorunz@gmx.com> wrote:
>
>> On 01/08/2023 16:44, Celejar wrote:
>>
>>> Any ideas?
>>>
>> Revert to Debian Stable and check there. I have AMD Radeon card and it's
>> 100% stable on Stable, and it had issues last year back when bookworm
>> was in Testing.
>
> I'm confused - current stable *is* bookworm, so if you had issues with
> bookworm, why would reverting to (current) stable help? Do you mean
> oldstable / bullseye? I actually did test using kernel 6.1.0-10-amd64
> from stable / bookworm, but the problem remained.
>

I had problems with Testing when it was maturing to become Stable /
Bookworm. Meaning, pre-release Bookworm, it has problems. So there is that.
You are using Sid, so problems are to be expected.

Revert to Stable / Bookworm, which is very stable for me, all problems
which it has last year (while Bookworm was Testing) have been fixed. I
did not have single crash or freeze for months.

--
With kindest regards, Piotr.

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system
⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org/
⠈⠳⣄⠀⠀⠀⠀

[toc] | [prev] | [next] | [standalone]

#260563

On Tue, 1 Aug 2023 18:25:39 +0100
piorunz <piorunz@gmx.com> wrote:

> On 01/08/2023 18:16, Celejar wrote:
> > On Tue, 1 Aug 2023 16:56:37 +0100
> > piorunz <piorunz@gmx.com> wrote:
> >
> >> On 01/08/2023 16:44, Celejar wrote:
> >>
> >>> Any ideas?
> >>>
> >> Revert to Debian Stable and check there. I have AMD Radeon card and it's
> >> 100% stable on Stable, and it had issues last year back when bookworm
> >> was in Testing.
> >
> > I'm confused - current stable *is* bookworm, so if you had issues with
> > bookworm, why would reverting to (current) stable help? Do you mean
> > oldstable / bullseye? I actually did test using kernel 6.1.0-10-amd64
> > from stable / bookworm, but the problem remained.
> >
> 
> I had problems with Testing when it was maturing to become Stable /
> Bookworm. Meaning, pre-release Bookworm, it has problems. So there is that.
> You are using Sid, so problems are to be expected.
> 
> Revert to Stable / Bookworm, which is very stable for me, all problems
> which it has last year (while Bookworm was Testing) have been fixed. I
> did not have single crash or freeze for months.

Okay, thanks for the clarification. It seems, however, that I'm dealing
with a hardware issue (as Dan Ritter suggested): I did some more
testing, and after the latest crash, the system won't boot at all, and
the power LED blinks red six times, which according to the
manufacturer "indicates a pre-video graphic card error."

"To resolve the issue, complete the following steps:

1) Reseat the graphics card.

2) Replace the graphic card.

3) Replace the system board."

https://support.hp.com/gb-en/document/c03599666

I suppose I'll begin by trying the first and see if it helps ...

-- 
Celejar

[toc] | [prev] | [next] | [standalone]

#260564

On 01/08/2023 19:09, Celejar wrote:
> Okay, thanks for the clarification. It seems, however, that I'm dealing
> with a hardware issue (as Dan Ritter suggested): I did some more
> testing, and after the latest crash, the system won't boot at all, and
> the power LED blinks red six times, which according to the
> manufacturer "indicates a pre-video graphic card error."
>
> "To resolve the issue, complete the following steps:
>
> 1) Reseat the graphics card.
>
> 2) Replace the graphic card.
>
> 3) Replace the system board."
>
> https://support.hp.com/gb-en/document/c03599666
>
> I suppose I'll begin by trying the first and see if it helps ...

Oh, that's more serious then. Could be power supply as well. Good luck
troubleshooting it.

--
With kindest regards, Piotr.

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system
⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org/
⠈⠳⣄⠀⠀⠀⠀

[toc] | [prev] | [next] | [standalone]

#260575

On Tue, 1 Aug 2023 19:13:42 +0100
piorunz <piorunz@gmx.com> wrote:

> On 01/08/2023 19:09, Celejar wrote:
> > Okay, thanks for the clarification. It seems, however, that I'm dealing
> > with a hardware issue (as Dan Ritter suggested): I did some more
> > testing, and after the latest crash, the system won't boot at all, and
> > the power LED blinks red six times, which according to the
> > manufacturer "indicates a pre-video graphic card error."
> >
> > "To resolve the issue, complete the following steps:
> >
> > 1) Reseat the graphics card.
> >
> > 2) Replace the graphic card.
> >
> > 3) Replace the system board."
> >
> > https://support.hp.com/gb-en/document/c03599666
> >
> > I suppose I'll begin by trying the first and see if it helps ...
> 
> Oh, that's more serious then. Could be power supply as well. Good luck
> troubleshooting it.

So, while troubleshooting today I thought I smelled an ominously acrid
odor of burning plastic. I wasn't sure whether it was real or a figment
of my stressed imagination, but when I opened the case, sure enough, the
cable feeding the GPU had burned and broken.

Fortunately, I don't see damage to the system's power cable or to the
GPU itself, just to the 6 pin to 8 pin PCIE adapter cable (the HP PSU
has only 6 pin cables, and the GPU needs an 8 pin connection). I
remember debating at the time (3 years ago) whether to buy the cheapest
one I could find, or to invest a few dollars more to buy a brand name
unit from a reputable company. I rolled the dice on the former, and it
did last three years - and hopefully didn't take out anything else with
it.

I'm going to order another one, from a reputable company this time, and
hope that it was just the cheapo implementation that was flawed, and
not the whole concept of 6 pin to 8 pin adapter cables ...

-- 
Celejar

[toc] | [prev] | [next] | [standalone]

#260604

On 02 Aug 2023 03:21, Celejar wrote:
> when I opened the case, sure enough, the
> cable feeding the GPU had burned and broken.
> 
> Fortunately, I don't see damage to the system's power cable or to the
> GPU itself, just to the 6 pin to 8 pin PCIE adapter cable (the HP PSU
> has only 6 pin cables, and the GPU needs an 8 pin connection). I
[...]
> 
> I'm going to order another one, from a reputable company this time, and
> hope that it was just the cheapo implementation that was flawed, and
> not the whole concept of 6 pin to 8 pin adapter cables ...
> 

Cables can burn because too much current (Amps) is flowing through them.
As problems only happen when gaming (or heavily using the GPU), this may 
be the case.

I think your GPU is drawing too much current from the 6 pins PSU rail.
(Are any other peripherals connected on the GPU rail ?).

I'd also compare the power requirements of the GPU to the total Amps the 
GPU PSU rail can provide (each rail and voltage has specific specs).

The difference between 6 and 8 pins can be seen like the difference of 
cable section between a small lamp and a computer (or a microwave).
Power your computer or microwave with a lamp cable, and you'll see a 
nice fire.

If the designers used 8 pins, my wild guess is that it has a reason ;)

[toc] | [prev] | [next] | [standalone]

#260605

On Wed, 2 Aug 2023 23:12:02 +0200
zithro <slack@rabbit.lu> wrote:

> On 02 Aug 2023 03:21, Celejar wrote:
> > when I opened the case, sure enough, the
> > cable feeding the GPU had burned and broken.
> > 
> > Fortunately, I don't see damage to the system's power cable or to the
> > GPU itself, just to the 6 pin to 8 pin PCIE adapter cable (the HP PSU
> > has only 6 pin cables, and the GPU needs an 8 pin connection). I
> [...]
> > 
> > I'm going to order another one, from a reputable company this time, and
> > hope that it was just the cheapo implementation that was flawed, and
> > not the whole concept of 6 pin to 8 pin adapter cables ...
> > 
> 
> Cables can burn because too much current (Amps) is flowing through them.
> As problems only happen when gaming (or heavily using the GPU), this may 
> be the case.
> 
> I think your GPU is drawing too much current from the 6 pins PSU rail.
> (Are any other peripherals connected on the GPU rail ?).

I don't think so.

> I'd also compare the power requirements of the GPU to the total Amps the 
> GPU PSU rail can provide (each rail and voltage has specific specs).

The Z440 officially supports up to an NVIDIA Quadro K6000 12GB, which
draws 234 watts, so it ought to be able to handle my Red Devil RX-570.
The Red Devil specifies a minimum system power of 450 watts, and my
Z440's PSU is 700 watts:

https://support.hp.com/us-en/document/c04506309#AbT3
https://www.powercolor.com/product?id=1493267392#spe

> The difference between 6 and 8 pins can be seen like the difference of 
> cable section between a small lamp and a computer (or a microwave).
> Power your computer or microwave with a lamp cable, and you'll see a 
> nice fire.
> 
> If the designers used 8 pins, my wild guess is that it has a reason ;)

But reputable companies do produce 6-8 pin adapters, e.g.:

https://www.startech.com/en-us/where-to-buy/PCIEX68ADAP
https://www.microcenter.com/product/615426/micro-connectors-6-pin-to-8-pin-pcie-adapter-power-cable-black-yellow
https://www.walmart.com/ip/Cable-Matters-2-Pack-6-Pin-PCIe-to-8-Pin-PCIe-Adapter-Power-Cable-4-Inches/944325506

(I just ordered the Cable Matters one.)

-- 
Celejar

[toc] | [prev] | [next] | [standalone]

#260606

On 02/08/2023 22:29, Celejar wrote:
> The Z440 officially supports up to an NVIDIA Quadro K6000 12GB, which
> draws 234 watts, so it ought to be able to handle my Red Devil RX-570.
> The Red Devil specifies a minimum system power of 450 watts, and my
> Z440's PSU is 700 watts:

More detailed info:
https://www.techpowerup.com/gpu-specs/powercolor-red-devil-rx-570-oc.b4455

Your GPU max TDP is 150W, meaning it will draw 75W from PCI-E slot and
75W from 8-pin cable.
Your 8-pin adapter must have been very poor quality. I don't know what
kind of adapter it was, but adapters which make 8-pin from 6-pin, are
dangerous. Better to use 2x 6-pin -> 1x 8-pin adapter to correctly
assign wires to each corresponding pin.

--
With kindest regards, Piotr.

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system
⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org/
⠈⠳⣄⠀⠀⠀⠀

[toc] | [prev] | [next] | [standalone]

#260608

On Wed, 2 Aug 2023 22:37:50 +0100
piorunz <piorunz@gmx.com> wrote:

> On 02/08/2023 22:29, Celejar wrote:
> > The Z440 officially supports up to an NVIDIA Quadro K6000 12GB, which
> > draws 234 watts, so it ought to be able to handle my Red Devil RX-570.
> > The Red Devil specifies a minimum system power of 450 watts, and my
> > Z440's PSU is 700 watts:
> 
> More detailed info:
> https://www.techpowerup.com/gpu-specs/powercolor-red-devil-rx-570-oc.b4455
> 
> Your GPU max TDP is 150W, meaning it will draw 75W from PCI-E slot and
> 75W from 8-pin cable.
> Your 8-pin adapter must have been very poor quality. I don't know what
> kind of adapter it was, but adapters which make 8-pin from 6-pin, are
> dangerous. Better to use 2x 6-pin -> 1x 8-pin adapter to correctly
> assign wires to each corresponding pin.

I'm not sure I understand your point: if we assume that the fact that
my adapter burned indicates that my particular adapter must have been of
very poor quality, than this implies that such adapters in general are
not dangerous (which, as I've noted, is supported by the fact that
reputable companies sell them, with no warnings that they're dangerous).

-- 
Celejar

[toc] | [prev] | [next] | [standalone]

#260614

On 03 Aug 2023 01:25, Celejar wrote:
> I'm not sure I understand your point: if we assume that the fact that
> my adapter burned indicates that my particular adapter must have been of
> very poor quality, than this implies that such adapters in general are
> not dangerous (which, as I've noted, is supported by the fact that
> reputable companies sell them, with no warnings that they're dangerous).

You will never find in a car owner's manual that driving on pedestrians 
may be dangerous ;)

Joke aside those companies just wanna sell products.
Do the products fit you ? Read my previous email: YOU must check.

To remove the confusion : your graphic card is rated at 150W, but it's 
the MAX power it can use, not the power it uses all the time.
So, until you don't stress the GPU, those adapters will be perfectly fine.
Example from a Win domU, GPU-Z reports ~40W in idle (browsing, videos, 
...) for a Polaris20 GPU (AMD RX580), with a TDP of 185W.
That's why you only had problems when REALLY using the GPU.

For instance, let's say you built a server, but only have a RX6600 as 
video card, a 6 pins connector and a 6-to-8 adapter.
THIS will be perfectly fine : your GPU will never exceed limits, as it 
will at most display a framebuffer.

(To go even further : AFAIK, most graphic cards won't boot without the 
external PCI-E power connector plugged, but if there wasn't such 
"protection", as the x16 PCIe slot provides 75W, in my example above it 
would be perfectly fine to use the card w/o the external plug).

[toc] | [prev] | [next] | [standalone]

#260627

On Thu, 3 Aug 2023 02:08:27 +0200
zithro <slack@rabbit.lu> wrote:

> On 03 Aug 2023 01:25, Celejar wrote:
> > I'm not sure I understand your point: if we assume that the fact that
> > my adapter burned indicates that my particular adapter must have been of
> > very poor quality, than this implies that such adapters in general are
> > not dangerous (which, as I've noted, is supported by the fact that
> > reputable companies sell them, with no warnings that they're dangerous).
> 
> You will never find in a car owner's manual that driving on pedestrians 
> may be dangerous ;)

But you certainly will find warnings regarding safe vs. unsafe use of
the product.

> Joke aside those companies just wanna sell products.
> Do the products fit you ? Read my previous email: YOU must check.
> 
> To remove the confusion : your graphic card is rated at 150W, but it's 
> the MAX power it can use, not the power it uses all the time.
> So, until you don't stress the GPU, those adapters will be perfectly fine.
> Example from a Win domU, GPU-Z reports ~40W in idle (browsing, videos, 
> ...) for a Polaris20 GPU (AMD RX580), with a TDP of 185W.
> That's why you only had problems when REALLY using the GPU.
> 
> For instance, let's say you built a server, but only have a RX6600 as 
> video card, a 6 pins connector and a 6-to-8 adapter.
> THIS will be perfectly fine : your GPU will never exceed limits, as it 
> will at most display a framebuffer.
> 
> (To go even further : AFAIK, most graphic cards won't boot without the 
> external PCI-E power connector plugged, but if there wasn't such 
> "protection", as the x16 PCIe slot provides 75W, in my example above it 
> would be perfectly fine to use the card w/o the external plug).

Thanks for the explanation.

-- 
Celejar

[toc] | [prev] | [next] | [standalone]

#260611 — Follow the specs :) (was Re: AMD GPU hard lockups)

On 02 Aug 2023 23:29, Celejar wrote:
 > But reputable companies do produce 6-8 pin adapters, e.g.:
[...]
 > (I just ordered the Cable Matters one.)

Out of curiosity, I checked the links.
Funny that "reputable companies" (I'm not attacking you, but them) don't 
even specify the max power rating as product specs, and when specified, 
it's only in users Q&A ...
You also have to dig deep to get the wires size in AWG (ie: zoom on 
pictures ...) !
Compare that to the information you get for the PSU and the GPU !

Anyways :
1st link: "The maximum power rating for the PCIEX68ADAP is 75W"
2nd link: "Max Power Rating 150W"
3rd link: (nowhere to be found, at least quickly)

1 and 2 use AWG18 wires, and on 3 they look even thinner (can't really 
tell as the pic is s***).
I'd eliminate the 3 from the start. No specs, no money.
And either 1 or 2 is lying : same AWG, different power rating (ok, the 
wire style is precised nowhere : single core, threaded, etc). But we can 
safely assume that 150W @12V is not possible : 12.5A is out of specs.
(Try "https://www.engineeringtoolbox.com/wire-gauges-d_419.html" for 
more info on AWG, and use the formula Watts = Volts x Amperes).

On 02 Aug 2023 23:37, piorunz wrote:
> Your GPU max TDP is 150W, meaning it will draw 75W from PCI-E slot and
> 75W from 8-pin cable.

So the wires will work around their limits.
One little power spike and kaboom.
Fire, exclamation mark ; fire, exclamation mark.
Joke aside, the fire hazard is real. Especially exceeding limits with 
low-end products, when you don't know if the materials are fire-proofed.

> Your 8-pin adapter must have been very poor quality. I don't know what
> kind of adapter it was, but adapters which make 8-pin from 6-pin, are
> dangerous. Better to use 2x 6-pin -> 1x 8-pin adapter to correctly
> assign wires to each corresponding pin.

I agree, so that each wire (or group of wires) does not exceed the max 
current it can draw.

I found two pages cleanly explaining this, both were worth a read.
But don't quickly jump to conclusions, follow the flow !

https://www.cgdirector.com/gpu-power-cable-guide/
https://www.pcworld.com/article/395059/one-cable-or-two-for-powering-a-graphics-card-heres-the-answer.html


-- 
zithro / Cyril

[toc] | [prev] | [next] | [standalone]

#260615 — Re: Follow the specs :) (was Re: AMD GPU hard lockups)

On Thu, 3 Aug 2023 01:29:42 +0200
zithro <slack@rabbit.lu> wrote:

> On 02 Aug 2023 23:29, Celejar wrote:
>  > But reputable companies do produce 6-8 pin adapters, e.g.:
> [...]
>  > (I just ordered the Cable Matters one.)
> 
> Out of curiosity, I checked the links.
> Funny that "reputable companies" (I'm not attacking you, but them) don't 
> even specify the max power rating as product specs, and when specified, 
> it's only in users Q&A ...
> You also have to dig deep to get the wires size in AWG (ie: zoom on 
> pictures ...) !
> Compare that to the information you get for the PSU and the GPU !
> 
> Anyways :
> 1st link: "The maximum power rating for the PCIEX68ADAP is 75W"
> 2nd link: "Max Power Rating 150W"
> 3rd link: (nowhere to be found, at least quickly)
> 
> 1 and 2 use AWG18 wires, and on 3 they look even thinner (can't really 
> tell as the pic is s***).
> I'd eliminate the 3 from the start. No specs, no money.
> And either 1 or 2 is lying : same AWG, different power rating (ok, the 
> wire style is precised nowhere : single core, threaded, etc). But we can 
> safely assume that 150W @12V is not possible : 12.5A is out of specs.
> (Try "https://www.engineeringtoolbox.com/wire-gauges-d_419.html" for 
> more info on AWG, and use the formula Watts = Volts x Amperes).
> 
> On 02 Aug 2023 23:37, piorunz wrote:
> > Your GPU max TDP is 150W, meaning it will draw 75W from PCI-E slot and
> > 75W from 8-pin cable.
> 
> So the wires will work around their limits.
> One little power spike and kaboom.
> Fire, exclamation mark ; fire, exclamation mark.
> Joke aside, the fire hazard is real. Especially exceeding limits with 
> low-end products, when you don't know if the materials are fire-proofed.
> 
> > Your 8-pin adapter must have been very poor quality. I don't know what
> > kind of adapter it was, but adapters which make 8-pin from 6-pin, are
> > dangerous. Better to use 2x 6-pin -> 1x 8-pin adapter to correctly
> > assign wires to each corresponding pin.
> 
> I agree, so that each wire (or group of wires) does not exceed the max 
> current it can draw.
> 
> I found two pages cleanly explaining this, both were worth a read.
> But don't quickly jump to conclusions, follow the flow !
> 
> https://www.cgdirector.com/gpu-power-cable-guide/
> https://www.pcworld.com/article/395059/one-cable-or-two-for-powering-a-graphics-card-heres-the-answer.html

Thanks for the information and explanations!

> zithro / Cyril

Celejar

[toc] | [prev] | [next] | [standalone]

#260617 — Re: Follow the specs :) (was Re: AMD GPU hard lockups)

On 8/2/23 19:30, zithro wrote:
> On 02 Aug 2023 23:29, Celejar wrote:
>  > But reputable companies do produce 6-8 pin adapters, e.g.:
> [...]
>  > (I just ordered the Cable Matters one.)
> 
> Out of curiosity, I checked the links.
> Funny that "reputable companies" (I'm not attacking you, but them) don't 
> even specify the max power rating as product specs, and when specified, 
> it's only in users Q&A ...
> You also have to dig deep to get the wires size in AWG (ie: zoom on 
> pictures ...) !
> Compare that to the information you get for the PSU and the GPU !
> 
> Anyways :
> 1st link: "The maximum power rating for the PCIEX68ADAP is 75W"
> 2nd link: "Max Power Rating 150W"
> 3rd link: (nowhere to be found, at least quickly)
> 
> 1 and 2 use AWG18 wires, and on 3 they look even thinner (can't really 
> tell as the pic is s***).
> I'd eliminate the 3 from the start. No specs, no money.
> And either 1 or 2 is lying : same AWG, different power rating (ok, the 
> wire style is precised nowhere : single core, threaded, etc). But we can 
> safely assume that 150W @12V is not possible : 12.5A is out of specs.
> (Try "https://www.engineeringtoolbox.com/wire-gauges-d_419.html" for 
> more info on AWG, and use the formula Watts = Volts x Amperes).
> 
> On 02 Aug 2023 23:37, piorunz wrote:
>> Your GPU max TDP is 150W, meaning it will draw 75W from PCI-E slot and
>> 75W from 8-pin cable.
> 
> So the wires will work around their limits.
> One little power spike and kaboom.
> Fire, exclamation mark ; fire, exclamation mark.
> Joke aside, the fire hazard is real. Especially exceeding limits with 
> low-end products, when you don't know if the materials are fire-proofed.
> 
>> Your 8-pin adapter must have been very poor quality. I don't know what
>> kind of adapter it was, but adapters which make 8-pin from 6-pin, are
>> dangerous. Better to use 2x 6-pin -> 1x 8-pin adapter to correctly
>> assign wires to each corresponding pin.
> 
Separate the men from the boys, and ask the vender what the ampacity of 
the wire used in their product is rated at. If they cannot answer that, 
put our card away and go someplace that knows.

That is a term used in the NEC, and represents the maximum current the 
wire can handle. AWG 10 gauge copper has an ampacity rating of 20 amps, 
so its sick bird to feed it with a breaker rated for more than 20 amps. 
12 gauge is 15 amps,  By the time you get to 18 gauge it close to an amp 
or two.

> I agree, so that each wire (or group of wires) does not exceed the max 
> current it can draw.
> 
> I found two pages cleanly explaining this, both were worth a read.
> But don't quickly jump to conclusions, follow the flow !
> 
> https://www.cgdirector.com/gpu-power-cable-guide/
> https://www.pcworld.com/article/395059/one-cable-or-two-for-powering-a-graphics-card-heres-the-answer.html
> 
> 

Cheers, Gene Heskett.
-- 
"There are four boxes to be used in defense of liberty:
  soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.
  - Louis D. Brandeis
Genes Web page <http://geneslinuxbox.net:6309/>

[toc] | [prev] | [next] | [standalone]

#260621

debian-user@lists.debian.org
On 8/2/23, piorunz <piorunz@gmx.com> wrote:
> On 02/08/2023 22:29, Celejar wrote:
>> The Z440 officially supports up to an NVIDIA Quadro K6000 12GB, which
>> draws 234 watts, so it ought to be able to handle my Red Devil RX-570.
>> The Red Devil specifies a minimum system power of 450 watts, and my
>> Z440's PSU is 700 watts:
>
> More detailed info:
> https://www.techpowerup.com/gpu-specs/powercolor-red-devil-rx-570-oc.b4455
>
> Your GPU max TDP is 150W, meaning it will draw 75W from PCI-E slot and
> 75W from 8-pin cable.
> Your 8-pin adapter must have been very poor quality. I don't know what
> kind of adapter it was, but adapters which make 8-pin from 6-pin, are
> dangerous. Better to use 2x 6-pin -> 1x 8-pin adapter to correctly
> assign wires to each corresponding pin.


Was coming in on this to say something similar. Just read this in last
week or so while having problems with booting my setup again.

Whatever I read had nothing to do with rebooting so I've forgotten
where I saw it. I just looked at the Cable Matters product, and that
was the very piece of hardware being chatted up.

What I'm remembering is that if the power of possibly just one of
those pins is not matched up properly, you can kill an entire
motherboard, not just burn through a wire. Seems like it was
semi-proprietary to a single product line. What I don't understand is
the company didn't care enough about consumers to make a proprietary
*shaped" pin setup so that consumers NEVER fry their systems. Make
that single, volatile hole/port star-shaped or something, anything.

This isn't what I read, but it seems to paint a similar picture of
what might happen:

https://forums.tomshardware.com/threads/is-it-possible-i-destroyed-my-psu-by-using-a-wrong-cpu-cable.3721595/

Wishing you the best of luck on this.. :)

Cindy :)
-- 
Talking Rock, Pickens County, Georgia, USA
* runs with.. a system that boots when it gets a mind to.. *

[toc] | [prev] | [standalone]

Back to top | Article view | linux.debian.user

csiph-web