cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

riza_guntur
Journeyman III

Installing ATI and NVIDIA in same machine, how?

How to install it? Any driver issue? Detecting issue etc?

How about programming issue?

0 Likes
24 Replies
Russian
Journeyman III

Originally posted by: riza.guntur How to install it? Any driver issue? Detecting issue etc?

 

How about programming issue?

 

 

I will give you answer tommorow. Today i will instal on Win7 GTX260 and HD 4850

0 Likes

Thank you

I'm going to experiment in CUDA too

0 Likes

Originally posted by: riza.guntur Thank you

 

I'm going to experiment in CUDA too



 

It will take longer. I have to reinstall Win7 (i have change motherboard, and now Win7 do not whant to run).

 

Regards,

Dmitry

0 Likes

I have tryed esterday CUDA (GTX260) and ATI Stream (HD 4850) on the same PC.

Unfortunatly it is not working. 😞

Only one of them are working, and it depends which one is the first one.

It is strange, Win7 in this case support 4 displays, but Catalyst not able to finde HD4850 on the second place.

Mabe i have to connect them trough CrossFire cable.


I do not know, but now it is not working. 😞

0 Likes

Thank you

0 Likes

Originally posted by: Russian

I have tryed esterday CUDA (GTX260) and ATI Stream (HD 4850) on the same PC.


Unfortunatly it is not working. 😞


Only one of them are working, and it depends which one is the first one.


It is strange, Win7 in this case support 4 displays, but Catalyst not able to finde HD4850 on the second place.


Mabe i have to connect them trough CrossFire cable.



I do not know, but now it is not working. 😞


Same result if ATI GPU is used with onboard GPU enabled.
0 Likes

Originally posted by: Raistmer
Originally posted by: Russian Only one of them are working, and it depends which one is the first one. It is strange, Win7 in this case support 4 displays, but Catalyst not able to finde HD4850 on the second place. . 😞
Same result if ATI GPU is used with onboard GPU enabled.


Confirmed, ATI Card is not recognized if onboard graphic is enabled (that sucks!), if onboard is disabled, Linux will hang if no monitor is attached (that also sucks, because it is a compute node in the server room without kvm)

[Hardware: Tyan Thunder, dual Opteron, Radeon 4350, OS: Linux]

0 Likes
Xiaoming_Cui
Journeyman III

one of dx11's feature is Heterogeneous graphics support - This was missing in Vista but was in XP. Basically you can take two different graphics cards with totally different drivers and have them run at once. But this isn't an alternative to SLI or Crossfire. Meaning if you run a game, it'll still run on one card or the other. The same thing will happen for stream computing.

0 Likes
mjharvey
Adept I

It is possible under Linux, at least, but the drivers need to be installed in a particular order. The following worked for me using 1.3 (not tried with 1.4 or 2.0beta)

1) fully remove any NV driver from the system

2)  Install ATI drivers

3) Configure the X server to use the ATI driver

4) Install Nvidia drivers, telling the installer to neither replace the GL libraries nor reconfigure X.

CUDA can be used without any display, ATI seemed to require an  X-server to be running.

YMMV, etc.

 

Can AMD/ATI tech comment on this?

 

1.  Can fglrx (linux) and stream SDK on the GPU work without an active display attached?

 

1b.  Can driver installer be reworked to allow X11 to pick up GL/glx from non-standard locations?

 

To previous post:  How are you able to *not* install GL/glx nvidia components?    I found that nvidia installer always moves libGL* and installs its own components.   Can CUDA apps render opengl output to the display running flgrx?

 

Thx

0 Likes

To confirm what mjharvey has stated, I have run several machines most of 2009 with this configuration.  ATI as the desktop and additional cards in the system which ran CUDA.  This is pretty easy in linux. I found installing nVidia drivers then CUDA and finally ATI drivers and Brook/CAL.  Then configuring X to run the desktop using fglrx drivers.

The nVidia card didn't need a display but sometimes needed the devices created so I made a script to create them when they weren't automatically done without a head attached.

Windows had issues in Vista and XP loading the nVidia drivers. I always ended up with Windows complaining of a conflict and disabling the nVidia card.

0 Likes

Originally posted by: mjharvey

 

4) Install Nvidia drivers, telling the installer to neither replace the GL libraries nor reconfigure X.

 

 

 

 

Yes, I am able to get (non-graphical) CUDA to work if I install nvidia, then overwrite with ati. 

 

UBut how is step (4) done?   That is, how can I install nvidia driver and libs without installing the X driver and opengl GL/glx components?

 

 

0 Likes

Originally posted by: thehighhat

UBut how is step (4) done?   That is, how can I install nvidia driver and libs without installing the X driver and opengl GL/glx components?

 

 

During the nVidia driver install it asks if you want to install OpenGL libraries and then asks if you want the installer to modify the X configuration.

Answering No to both of those questions should give you the correct installation.

Most of the time I ended up following the install of nVidia drivers with the reinstall of ati fglrx drivers (custom built rpm package) to make sure the desktop loaded correctly.  But not updating the config.

 

0 Likes

Is there a way how to enable OpenCL on ATI (HD5770) which is plugged in as secondary GPU in Windows XP?

0 Likes

Hi, I know this thread is old and most people will not mix ATI and NVIDIA GPUs in the same host.

It seems that NVIDIA may have crippled it's drivers sometime last year so that the GPU is disabled if it detects an ATI graphics adapter: http://www.tomshardware.com/news/nvidia-ATI-physx-patch-gpu,8786.html . Before then, people had success mixing ATI and NVIDIA hardware if they used older NVIDIA drivers: http://www.golubev.com/blog/?p=70 .

My issue is that I have a GTX 480 (Fermi) and am trying to get this working on the same host that has a HD 5870. So I can not use older NVIDIA drivers as they won't work with Fermi. Everything seems right (just re-install ATI drivers after NVIDIA install), ATI stuff still works and NVIDIA deviceQuery runs. However, it returns: "There is no device supporting CUDA".

I am wondering if this is a software driver limitation that I can't really work around.

0 Likes
Ceq
Journeyman III

Try using Win7 x64 instead of WinXP, it offers less GPGPU performance but it worked for me. I installed the Radeon's driver first using the installer and then the GeForce, but using the device manager instead of the installer. PhysX didn't work, however OpenCL and CUDA do, so I don't mind at all about PhysX.

0 Likes
cjang
Journeyman III

Thank you for the suggestion. I am on Ubuntu 9.04 x86_64 however. Life is probably easier for Windows people in some ways...

I think that my issue is due to mainboard BIOS incompatibility: http://www.xtremesystems.org/forums/showthread.php?t=251412 . I have a Foxconn Bloodrage which has issues with the GTX 470/480. If the GTX 470/480 is the primary device for video, the system won't even reach the boot loader. When I have a HD 5870 as the video output device and have the GTX 470/480 in the secondary PCIe x16 slot, the system runs fine but the nVidia GPU is effectively dead.

I will try updating the BIOS and then see if I can get both GPUs (EVGA GTX 480 and Asus HD 5870) to work. Success or failure, this will at least be a data point with the current Evergreen and Fermi hardware from ATI and nVidia. I've been googling and can't find anyone else with this configuration!

0 Likes
Ceq
Journeyman III

Well, some time ago I had some skills using Linux and GNU compiler/tools. But then I met Visual Studio and I become addict due to its nice development environment.

About the card order, I had a similar problem... I have to use the Radeon as the primary display device, otherwise the system won't boot. However, using the Radeon as primary, under some circunstances it is like the GeForce's video BIOS isn't called and the card isn't initialized (not detected and noisy fan,?). The solution I'm using is to boot first in another Windows installation using Microsoft's reference driver and the Radeon disabled (GeForce's fan lowers RPM and now the card is ready for GPGPU,?).

0 Likes
cjang
Journeyman III

It is possible to use both ATI and NVIDIA GPUs from the same shell for CUDA/OpenCL with recent drivers. No hardware, BIOS reconfiguration or rebooting is required. However, I am on Linux and suspect this approach does not work for Windows. The reason is the need for hacked NVIDIA drivers to play games so that ATI is the GPU (graphics) and NVIDIA is the PPU (physics).

Basically, the NVIDIA driver needs to believe it is connected to a display and screen. So if the X server is started once, even if unsuccessfully, then the driver module resident in memory believes it is being used for graphics. From then on, the driver is "unlocked" and works for CUDA and OpenCL. The driver doesn't really need or use the X server. To my best knowledge, the only reason for the dependency is to prevent users from playing games using an ATI GPU with an NVIDIA PPU.

The thing I do not know is if the NVIDIA driver on Windows really needs an attached physical display and screen. The driver probes for an attached monitor. If it doesn't find one, it then fails. On Linux, this is ok and in fact desirable. All I need to do is unlock the driver module so the GPU works. On Windows, you may need to attach a physical monitor (or a KVM switch) to the NVIDIA GPU to make the driver work.

What's funny is that while the ATI OpenCL implementation is dependent on a running X server, the NVIDIA implementation is not - yet there is an introduced dependency that forces initializing the windowing system in order to unlock the driver.

So the workaround I have now is:

  1. install NVIDIA driver
  2. edit xorg.conf so it uses the NVIDIA GPU
  3. startx (this will fail but that's good as we only need to unlock the driver)
  4. install ATI driver
  5. edit xorg.conf so it uses the ATI GPU
  6. startx

This is naturally inside a script. Unfortunately, now that both the ATI and NVIDIA implementations write to common X runtime libraries (i.e. OpenGL), they conflict, resulting in the ugly workaround above. It probably wouldn't be too hard to figure out exactly where the two driver installs intersect to come up with a cleaner solution.

Also, some of my earlier problems (with NVIDIA GPU in primary PCIe slot, system does not boot) were due to a Foxconn Bloodrage mainboard BIOS incompatibility with the GTX 470/480. Fortunately, last month, a Lord Trident came out with the G40 beta BIOS that fixes this on the Bloodrage. Ceq, I think your boot failure problem may be something like this. It's a device/BIOS incompatibility.

0 Likes

We are trying to install a NVIDIA Geforce GTX 480 and an ATI Radeon HD5870 in a CentoS 5.5 (Our goal is to use it as a server, connected by ssh and using OpenCL to compute)

We encounter difficulties to make things working.

We started with installing Nvidia driver 256.44 (cuda Toolkit 3.1 / Opencl 1.0)

Nvidia parts works properly but we couldn't see the Ati's GPU.

We then installed the Ati catalyst 10.7 driver (OpenCL 1.1 support).

Now everything is broken, we can't startx and we can't see both cards anymore.

- can you post your Xorg.conf, it may probably help us

- Do you think it will cause problems to mix a nvidia driver 1.0 with an ati 1.1?

Thank you very much for you time

 

 

 

 

 

 

 

 

 

 

 

 

 

0 Likes

Originally posted by: Ceq Try using Win7 x64 instead of WinXP, it offers less GPGPU performance but it worked for me. I installed the Radeon's driver first using the installer and then the GeForce, but using the device manager instead of the installer. PhysX didn't work, however OpenCL and CUDA do, so I don't mind at all about PhysX.

 

hi Ceq,

I was trying to install both HD5870 and GTX460 on my Win7 x64 system just as you mentioned, but unfortunately ever since I run the

status = clGetPlatformIDs(0, NULL, &numPlatforms);

to get platform info, the code crashes with an access violation in a given reading location (0x0000013c). What driver versions you used?

I have hooked both cards in my system and each one is cabled to a different monitor. After boot, windows prepared the cards and I rebooted. After reboot, I cleared every Radeon/Geforce software (display drivers, opencl sdk and so on). After rebooting, 1- I installed the ccc 10.10 with display drivers and the included opencl sdk (v2.2), 2- I went into the device manager and updated the gtx 460 driver using the 260.99 version.

Any suggestions?

ps: i had previously installed two amd cards together (HD5870 and HD5770) successfully (could run opencl on both cards separately).

0 Likes
sanjib
Journeyman III

Microsoft OS's can't handle two diffrent display drivers at once, which is one of the reasons why windows needs to be re-designed from scratch.Charter Medical

<a href=" http://www.chartermedical.ie "> Charter Medical </a>

0 Likes

sry, but win7 can. With 2 cards, 2 monitors, i can install both amd and nvidia drivers. And as Ceq said above, he could even do it for opengl/opencl api's. I just would like to know how it can be donne, in which order, drivers, sdk's, pci-e slot order...

0 Likes

someone recomend install nvidia card first and then ati. and ati have as primary card.

0 Likes