ASRock.com Homepage
Forum Home Forum Home > Technical Support > AMD Motherboards
  New Posts New Posts RSS Feed - Dual GPU not working on x399 Taichi
  FAQ FAQ  Forum Search Search  Events   Register Register  Login Login

Dual GPU not working on x399 Taichi

 Post Reply Post Reply Page  <123>
Author
Message
MisterJ View Drop Down
Senior Member
Senior Member


Joined: 19 Apr 2017
Status: Offline
Points: 1097
Post Options Post Options   Thanks (0) Thanks(0)   Quote MisterJ Quote  Post ReplyReply Direct Link To This Post Posted: 20 Nov 2017 at 4:59am
happyferret, is there any chance you would be willing to try W10?  Please flesh out your specifications with drives, memory, SSDs and anything else.  I suspect you you are well aware of how to run W10, but will be glad to give you some pointers if needed.  Thanks and enjoy, John.

EDIT:  Fun User ID!


Edited by MisterJ - 20 Nov 2017 at 5:00am
Fat1 X399 Pro Gaming, TR 1950X, RAID0 3xSamsung SSD 960 EVO, G.SKILL FlareX F4-3200C14Q-32GFX, Win 10 x64 Pro, Enermx Platimax 850, Enermx Liqtech TR4 CPU Cooler, Radeon RX580, BIOS 2.00, 2xHDDs WD
Back to Top
noktek View Drop Down
Newbie
Newbie


Joined: 17 Nov 2017
Status: Offline
Points: 10
Post Options Post Options   Thanks (0) Thanks(0)   Quote noktek Quote  Post ReplyReply Direct Link To This Post Posted: 20 Nov 2017 at 6:41am
Originally posted by parsec parsec wrote:

An HX 1200i should be enough for a TR system and two video cards. This PSU's +12V rail is capable of 100A output. If the PC is drawing over 100A from the +12V rail just booting and at idle, so much for power efficiency of TR:

http://www.corsair.com/en-us/hxi-series-hx1200i-high-performance-atx-power-supply-1200-watt-80-plus-platinum-certified-psu

Switching to single rail will skip the need to balance the power on the +12V rails, but just starting and running the PC should not cause that much of a power draw. 40A on a rail might be a limitation for the CPU, which could interfere with the PCIe lanes, just a thought. But the CPU here has multiple power sources/connections.

I would start over by clearing the CMOS/UEFI, connect all the PSU cables to the board as required, and use one video card. Possibly starting with the 780 alone in the board. Once one card is working normally, add the other. If you've never tried the 780 by itself, that's what I would try now.

I'm wondering if the VBIOS in the 780 SC is UEFI/GOP compatible. It should be if it is an EVGA video card, EVGA went to UEFI compatible VBIOS starting with the 700 series. The 1080Ti will have a UEFI/GOP compatible VBIOS. Since you use a 960 Pro NVMe SSD as the OS drive, the Windows 10 installation should be UEFI booting by default. Windows 10 installed via a USB flash drive?

 "You lost me on "If i remove uninstall the 780 it then shows up as VGA adapter displaying no errors." That's the Device Manager entry, correct? I assume that is before you restarted the PC? Or does that ghost entry remain after a restart?"

I wonder if this issue is related to the early Window 10 RS1 versions and the IOMMU option being enabled, not being compatible. Obviously the Windows 10 installation boots, so you are either using a Windows 10 RS2 version, or have IOMMU disabled.

So the driver version is identical for both video cards? During the installation, are you prompted to select which card is the target of the installation? Or do both cards just get the driver installed? No errors from the installer during or after the driver installation process?

What happens if you run the NVIDIA Control Panel software with both cards in the PC? Can you check how many PCIe lanes are being used by the 780 with that or other software? If you've used two different video cards like this in the past, is there anything you've had to do to get them working?

Usually two video cards don't have a problem with the amount of memory Windows needs to deal with both cards. The UEFI option to change that is usually called Top Of Lower Usable Dram, or TOLUD, setting it to Dynamic. If and where that option is in your board's UEFI, I'm not sure, usually in the Chipset Configuration screen, but could be elsewhere with X399.

Try to keep an open mind about the mother board being the problem. We see that being blamed for issues all the time, and when a new board fixes nothing, which we also see all too often, it's more than just a waste of time.

hi and thanks for your throrough answer!
PSU wise i am not worried, 1200W should really be enough for 3 GPUS actually!
The GPU is also not the issue in my opinion, as i have tried it on another machine and it works just perfectly fine. Have also updated the VBIOS today to a later version to no avail.
I have indeed flashed multiple times the UEFI and reset the CMOS. also recabled everything.
there is no way to make the 780 work on the x399 even if its the first and only card installed: upon nvidia drivers, black screen and no output, or in case its booting, image stops at the loading wheel (kinda last step of the "pre-windows" area) instead of proceeding to the login screen.

regarding the question if the 780 is uefi compatible i honestly dont really understand what you mean, in the sense that im a newbie :) anyhow as i said i did update the vbios of the card on my other maching and then moving it over but didnt change much

yes, i did install windows 10 through download from ms website and put it on a usb stick


cit: You lost me on "If i remove uninstall the 780 it then shows up as VGA adapter displaying no errors." That's the Device Manager entry, correct? I assume that is before you restarted the PC? Or does that ghost entry remain after a restart?
 
with that i meant that basically when both the cards are in and i have the screen connected through the 1080 and i remove drivers, the 780 sometime sits there as generic driver adapter without the triangle. later on though it had happened that even as a generic driver adapter it maintained the triangle. go tell!

i am using the latest history and from what i can recall iommu is disabled by default and i have not enabled it not being sure what it is anyway

the driver version is supposedly the latest for both cards, i think 388.13 at this point in time, and gives no error whatsoever for the 1080. the 780 is simply ignored and does not figure in the nvidia control panel, no question is asked. no errors either

i have never installed myself 2 cards like this in the past but i have been using and updating them at work, it was never difficult, also because one was a quadro and the other one a gtx so it was a bit easier.

How would i go about checking what PCIE lanes are used and all that? I have tried looking for the settings you mentioned in the UEFI but theres nothing that looks similar to what you mentioned or that can be set to "Dynamic"

this picture https://i.imgur.com/NH6nHqL.jpg shows that gpuz can detect the presence of the hardware partially, but not nvidia control panel




Edited by noktek - 20 Nov 2017 at 6:59am
Threadripper 1950, Asrock Taichi x399, Noctua 14s, Vengeance LPX 64gb 3066, Asus Strix 1080Ti, EVGA 780 SC, Corsair Hx1200, Samsung 960pro 512, evo 1gb, Barracuda 3gb, win10 64, bios 1.8
Back to Top
noktek View Drop Down
Newbie
Newbie


Joined: 17 Nov 2017
Status: Offline
Points: 10
Post Options Post Options   Thanks (0) Thanks(0)   Quote noktek Quote  Post ReplyReply Direct Link To This Post Posted: 20 Nov 2017 at 7:03am
Originally posted by happyferret happyferret wrote:

Hi noktek,


I was having the exact same issue (except with 2x 1080Ti) last Friday, and spent more than 8 hours fighting with it until uncovering your post: you are not alone.

I am running linux, but have the exact same problem of the second GPU being detected, but in my case not being usable by CUDA. I have a "beefier" power supply (1200W gold Seasonic), and have had a very strange reactions to switching power cables around in the computer:

- I usually cannot use both cards (more than 99% of the time).
- However, sometimes, after I stop the machine, and switch cables around inside, I can get ONE boot where both cards are detected and can be used by CUDA. This does not last until the next reboot, and has worked oince when both cables were plugged with two cables each (and therefore no PSU power to the motherboard), and ONCE when the motherboard was plugged to the PSU (and therefore one of the GPU was connected to only one power cable (in this case, the detected card was under-powered in a way, but was detected and usable). I am sorry if this is confusing, I can elaborate: TL;DR: I can SOMETIMES get both cards to work, but it seems completely random, is very rare, and does not last after a reboot.

I am extremely confused. I have updated the BIOS to the latest patch (November 7), but this did not help. I suspect that it is not a power problem, but cannot find what seems to have any effect on this problem.

Thanks to anyone with new ideas, I would love to have the two GPUs actually work together :-)


well you're already one step ahead it seems :) what models are your gpus ?
Threadripper 1950, Asrock Taichi x399, Noctua 14s, Vengeance LPX 64gb 3066, Asus Strix 1080Ti, EVGA 780 SC, Corsair Hx1200, Samsung 960pro 512, evo 1gb, Barracuda 3gb, win10 64, bios 1.8
Back to Top
MisterJ View Drop Down
Senior Member
Senior Member


Joined: 19 Apr 2017
Status: Offline
Points: 1097
Post Options Post Options   Thanks (0) Thanks(0)   Quote MisterJ Quote  Post ReplyReply Direct Link To This Post Posted: 20 Nov 2017 at 7:19am
noktek, seen this?
https://forums.evga.com/EVGA-780-SC-UEFI-post-error-m2503436.aspx
May be out of date, but please check it out.
Driver release notes: https://us.download.nvidia.com/Windows/388.31/388.31-win10-win8-win7-desktop-release-notes.pdf
VGA BIOS UEFI: https://www.techpowerup.com/vgabios/139391/evga-gtx780-3072-130514
It says "UEFI Supported: Yes".

When all else fails, open a support ticket with the Vendor.  I suspect you have a vBIOS and Driver compatibility problem.  Have you modified the CSM or IOMMU settings?  Your user manual will tell you how many lanes each socket supports.  Also try GPU-Z - here: https://www.techpowerup.com/gpuz/
For Asus Strix 1080Ti, ASUS shows a different driver: https://www.asus.com/us/Graphics-Cards/ROG-STRIX-GTX1080TI-O11G-GAMING/HelpDesk_Download/
Have fun, John.


Fat1 X399 Pro Gaming, TR 1950X, RAID0 3xSamsung SSD 960 EVO, G.SKILL FlareX F4-3200C14Q-32GFX, Win 10 x64 Pro, Enermx Platimax 850, Enermx Liqtech TR4 CPU Cooler, Radeon RX580, BIOS 2.00, 2xHDDs WD
Back to Top
parsec View Drop Down
Moderator Group
Moderator Group
Avatar

Joined: 04 May 2015
Location: USA
Status: Offline
Points: 4996
Post Options Post Options   Thanks (0) Thanks(0)   Quote parsec Quote  Post ReplyReply Direct Link To This Post Posted: 20 Nov 2017 at 12:47pm
In simple terms a UEFI compatible video card has a VBIOS that supports a UEFI protocol called GOP, Graphics Output Protocol. It's hard to explain everything related to UEFI vs BIOS firmware, but our UEFI firmware is by default run in an emulated BIOS firmware mode by the CSM, Compatibility Support Module, which is part of the UEFI/BIOS, and enabled by default.

There are multiple levels of UEFI vs BIOS (BIOS is also called Legacy) support, the main ones being storage (SSDs, HDDs) video, and networking. Since you probably have not set the CSM option in the UEFI to disabled, you should only be using the UEFI storage Option ROM in the UEFI. That happened when you installed Windows 10 on an NVMe SSD, and the installation media was booted from an entry like this: "UEFI: <device name>". In the boot order, your OS drive is listed as Windows Boot Manager: Samsung 960 Pro, or similar, correct?

If you had CSM disabled, and the video source (GPU or iGPU) did not support the UEFI/GOP protocol, you would have a black screen, no video output when the PC started. Your symptoms using the 780 seem to indicate it may not be UEFI/GOP compatible, but I'm not 100% certain of that. The 1080 has UEFI/GOP support, and I was surprised, given the EVGA forum link above (thanks for that MisterJ), that a 780 video card did not have that support from the factory. I recall I had to update the VBIOS of a EVGA 650 for GOP support, but I don't recall doing that for my 760, perhaps I did. Confused  EVGA had GOP support in their product's VBIOS before other manufactures.

You could test the 780 for UEFI/GOP support on another PC where the 780 works with the GPU-Z tool from TechPowerUp:

https://www.techpowerup.com/download/techpowerup-gpu-z/

For example, see the box labeled UEFI below the NVIDIA logo:



If your 780 does not have that box checked, its VBIOS does not support UEFI/GOP, and that is most likely the issue. IMO, this is worth checking, at least to eliminate it from further consideration. Possibly the 1080 being GOP compatible does not like working with a non-GOP compatible card, or the driver installer does not like the difference. Sorry but this is an interesting problem, I'm curious what the source is, and the resolution.

I don't have your board, and the manual is outdated compared to the UEFI/BIOS updates, to find the TOLUD option, if you even have that option. I highly doubt that is the cause, since that is only an issue when using four or more video cards. You have so much DRAM memory, there is plenty for Windows to use.

That the 780 is not seen in the NVIDIA control panel is really bad, that is so basic and should not happen. If we check the X399 Taichi manual's SLI and CrossFire sections (NOT what you are doing, I know that) the setup procedure is to simply insert two, three, or four cards at the same time, not one at a time. It is expected that all the cards are identical, but that should not be a problem here AFAIK. I will try two different cards in my Ryzen system as a generally related test.

Beyond the above, I've got nothing else. Besides the 780 being broken, but that is not the situation. The potentially different driver in a post above is something to check on. You could try posting in the EVGA forum about this.


Back to Top
noktek View Drop Down
Newbie
Newbie


Joined: 17 Nov 2017
Status: Offline
Points: 10
Post Options Post Options   Thanks (0) Thanks(0)   Quote noktek Quote  Post ReplyReply Direct Link To This Post Posted: 20 Nov 2017 at 4:59pm
Originally posted by MisterJ MisterJ wrote:

noktek, seen this?
https://forums.evga.com/EVGA-780-SC-UEFI-post-error-m2503436.aspx


thank you! this is very interesting, and it is the exact same model i have, 03G-P4-2784-KR. Ill write EVGA again straight away... the last ticket was resolved as :  "contact mobo manufacturer for bios update"!

and here the CPU-Z screenshot


just wrote to the evga tech support, lets see if they have some good news for us!


Edited by noktek - 20 Nov 2017 at 5:09pm
Threadripper 1950, Asrock Taichi x399, Noctua 14s, Vengeance LPX 64gb 3066, Asus Strix 1080Ti, EVGA 780 SC, Corsair Hx1200, Samsung 960pro 512, evo 1gb, Barracuda 3gb, win10 64, bios 1.8
Back to Top
happyferret View Drop Down
Newbie
Newbie


Joined: 20 Nov 2017
Location: Pittsburgh
Status: Offline
Points: 2
Post Options Post Options   Thanks (0) Thanks(0)   Quote happyferret Quote  Post ReplyReply Direct Link To This Post Posted: 21 Nov 2017 at 1:31am
Originally posted by MisterJ MisterJ wrote:

happyferret, is there any chance you would be willing to try W10?  Please flesh out your specifications with drives, memory, SSDs and anything else.  I suspect you you are well aware of how to run W10, but will be glad to give you some pointers if needed.  Thanks and enjoy, John.

EDIT:  Fun User ID!

Hi MisterJ,

Thanks for the reply. I was planning on installing W10 on this machine eventually, so I used this problem as a good reason to buy a Samsung 960Pro 512Gb to install Windows on. The SSD will be here tomorrow and I will try to see if the same issue arises on Windows too, and will report my findings.

In the meantime, I thought of two other reasons that could explain my problem:

1. I use linux, but boot using BIOS legacy mode. I do not know why this could be an issue, and since both cards were recognized and usable some rare times, I do not think that can be the cause, but maybe ?
2. I do not enough 6-pins power cables to power both 1080 Ti AND the motherboard. I ordered one more cable, which should also arrive tomorrow, and I'll try to see if the problem is resolved when I can power both cards and the motherboard, my guts tell me this problem is power-related, but that's just an intuition.


For the cards, they are two different models, here are the details: (both cards have been swapped, and the problem persists in any configuration):

- PNY GeForce GTX 1080 Ti Graphic Card - 1.58 GHz Boost Clock
- GIGABYTE GeForce GTX 1080 Ti Turbo 11GD, GV-N108TTURBO-11GD

Will be back with more updates, thank you all for your help!
ASRock X399 Taichi, AMD Ryzen Threadripper 1950x, 2x NVIDIA 1080Ti, Seasonic 1200W Gold PSU
Back to Top
MisterJ View Drop Down
Senior Member
Senior Member


Joined: 19 Apr 2017
Status: Offline
Points: 1097
Post Options Post Options   Thanks (0) Thanks(0)   Quote MisterJ Quote  Post ReplyReply Direct Link To This Post Posted: 21 Nov 2017 at 5:39am
noktek, good, hope something gets resolved.  Seems like the 780 driver was 388.31 not 388.13.  Can you try this one.  I don't remember the vBIOS version, but GPU-Z checks UEFI.  Please keep us up.  Thanks and enjoy, John.
Fat1 X399 Pro Gaming, TR 1950X, RAID0 3xSamsung SSD 960 EVO, G.SKILL FlareX F4-3200C14Q-32GFX, Win 10 x64 Pro, Enermx Platimax 850, Enermx Liqtech TR4 CPU Cooler, Radeon RX580, BIOS 2.00, 2xHDDs WD
Back to Top
MisterJ View Drop Down
Senior Member
Senior Member


Joined: 19 Apr 2017
Status: Offline
Points: 1097
Post Options Post Options   Thanks (0) Thanks(0)   Quote MisterJ Quote  Post ReplyReply Direct Link To This Post Posted: 21 Nov 2017 at 5:43am
Thanks, happyferret.  I suggest you run GPU-Z on each of your cards and confirm that both are UEFI compatible.  If not, please see if you can find new vBIOS(s)  Enjoy, John.
Fat1 X399 Pro Gaming, TR 1950X, RAID0 3xSamsung SSD 960 EVO, G.SKILL FlareX F4-3200C14Q-32GFX, Win 10 x64 Pro, Enermx Platimax 850, Enermx Liqtech TR4 CPU Cooler, Radeon RX580, BIOS 2.00, 2xHDDs WD
Back to Top
parsec View Drop Down
Moderator Group
Moderator Group
Avatar

Joined: 04 May 2015
Location: USA
Status: Offline
Points: 4996
Post Options Post Options   Thanks (0) Thanks(0)   Quote parsec Quote  Post ReplyReply Direct Link To This Post Posted: 21 Nov 2017 at 12:10pm
noktek, your 780 is UEFI/GOP compatible, in case you are still wondering, the GPU-Z screenshot confirms it.

I put a second video card in my Ryzen PC today, end of the story first:



A little story here, I did not do my usual procedure of clearing the UEFI/BIOS when installing new hardware, just connected the 650 video card and started the PC. I updated its VBIOS to UEFI/GOP compatible years ago.

The first time I started the PC, I did not have a monitor connected to the 650. Win 10 got to the loading screen circle of dots, and then froze. I shut off the PC.

I then connected a monitor to the 650, borrowing one of the three monitors I had connected to the 960. I did nothing else at all, and started the PC. I have the ASRock logo screen disabled, so instead get a few lines of text displayed only on the primary monitor, with the two digit POST code. Ryzen's POST time is long, particularly in RAID mode, so that screen is visible for almost 10 seconds. I noticed I also had the same text and POST code display on the monitor connected to the 650. This time Win 10 booted to the Desktop fine.

But the Desktop was not really active, as the little blue activity wheel instead of the mouse pointer was spinning away for at least 30 seconds. The startup/POST text on the 650's monitor disappeared finally, and my usual background picture appeared on the 650's monitor. Win 10 was obviously loading the driver for the 650, which you cannot see above, but is the same version I installed for the 960, 23.21.13.8813.

The two video cards have been running fine now for about two hours. It seems to me that the normal behavior in the situation of adding a second video card to an existing PC would be for Windows to load a driver for the card. That was simple in my case since the same driver used by the 960 could be applied to the 650, so no download of another driver was necessary, I assume. Even in the worst case situation, say of a PC not connected to the Internet, without a compatible driver for the second card available, the generic Windows video driver that would be installed during a new Windows installation would be used, and installed on the 780. Do you agree?

Just to confirm for me, the GPU-Z screenshot was done on another PC, not on the X399 PC, correct? I only ask because I see it shows PCIe 2.0 as the interface. Not that it should make a difference.

The highest level question is, why wasn't a driver installed for the 780 video card? We don't have the same information from happyferret since he is not using Windows. But it is interesting (and telling) that he seems to have the same problem. How can we know specifically if it is the board's UEFI/BIOS that is the problem, not that I'm dismissing that. It is all to easy for another manufacture to pass off problems on the board's UEFI/BIOS, but it seems to be heading in that direction.

I have nothing new to suggest as a fix. I suppose an inadequate PSU problem with both users at 1200W seems improbable, but who knows? While my two video cards are not power hungry, I'm using a Seasonic 660W Gold PSU in my Ryzen X370/1700X PC. Can you run a monitoring program to check the PSU rails, to see if your +12V rail seems to be drooping below 12V, possibly as an indication of a power issue? That is shown in the H/W Monitoring screen in the UEFI.

I cannot find a TOLUD option in your board's manual, but as I said the manual is not up to date with all the UEFI updates. In the Advanced\AMD PBS screen there is the PCIe x16 Switch option, and the Promontory PCIe Switch option. What those options do, I don't know, but their default settings should allow multiple video cards, if these options were related to that. They are worth checking just in case.

One experiment for you, in the Boot screen in the UEFI, find the CSM option. Click on it to see its sub-options. Find the Launch Video OpROM Policy option, and set it to UEFI only. Then Save and Exit, boot into Windows, and see what happens. If you cannot boot after changing this option, you'll need to do a UEFI/CMOS clear to restore sanity, and that will work I promise.


Back to Top
 Post Reply Post Reply Page  <123>
  Share Topic   

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 12.04
Copyright ©2001-2021 Web Wiz Ltd.

This page was generated in 0.188 seconds.