Start a Conversation

Unsolved

This post is more than 5 years old

616082

April 16th, 2013 08:00

Use a PCI graphics card on Poweredge T620

Purchased  Poweredge T620, with a NVIDIA Tesla C2075 CUDA card in for use as a high end workstation and CUDA machine.

Due to the programs I must run I have set the machine up dual booting Windows 7 and Linux (this in itself took some doing).

The graphics are currently being served by the onboard Matrox G200.

I have managed, on both systems to install the CUDA graphics card, for use in programming.  However I decided to attempt to use another graphics card (NVIDIA Geforce GT610) as on Linux, installing the NVIDIA drivers stops me being able to access my monitors full screen resolution (a problem that I think will be more easily solved if graphics are being handled by an NVIDIA chipset on an external card).  Additionally the Matrox G200 is not going to be sufficient for some of the programs I am going to be required to run in Windows.

On installing the GT610, no output was displayed (both before and after driver installation), I discovered this is also the case if I plug a monitor into the CUDA card.  It seems there is some BIOS/iDRAC setting preventing output through any additional PCI graphics card.

I have tried disabling the onboard graphics controller in the BIOS, however this merely left me with no display at all, then requiring me to reset the BIOS via the jumper, a process I would prefer not to have to repeat.  I also speculatively, adjusted the iDRAC network setting "Enable NIC" to "Disabled" but with no luck (though I have not tried disabling the onboard graphics via the BIOS since this change).  Windows sees all graphics cards as installed and working, however does not see any displays connected to the,

If anyone has any ideas as to which settings need to be changed to enable output from PCI graphics cards, that would be amazing!

Thanks

178 Posts

January 17th, 2014 08:00

Thanks for the info. I have two 1100W power supplies (obviously in redundancy) and I have no problem running my GTX 780 Ti with two monitors. Max resolution on my monitors is 1920x1080... not sure if higher resolution ones draw more power or not, but that is what  have and its working fine.

G

p.s. I am no computer wiz, but if your card causes the computer to crash when it only draws like 52 Watt max, doesn't that mean something else is wrong?! (Power Consumption Operational:51.1Watt)

4 Posts

January 17th, 2014 08:00

Just for everyones education.  The k2000 graphics card does not have an input for its own power supply.  When I booted the machine using this video card (internal disabled) the video card started to draw power off the mother board it would cause an internal power failure and the system would crash.  In essence the video card is to powerful. I then installed an old Radeon x600 the system booted up immediatly and without problem with dual monitors.  So if you plan on using a t620 with dual monitors be careful which video card you use in it. 

4 Posts

January 17th, 2014 09:00

I have the same 1100w power supplies.  I dont think it necessarly not enough power in the box its that the k2000 pulls to much power from the pci slots on the mother board.  I also have a raid controller and an additional 4 port nic. 

178 Posts

January 17th, 2014 09:00

Wait a minute... I read somewhere, that the pci slots are only good for 25 Watts, or something like that. That would explain it...  The T620 can be ordered with a GPU installation kit, which I have. It is a bunch of power cables to hook up the GPU's... 

thanks,

G

548 Posts

January 19th, 2014 01:00

A PCIe slot comes in two flavors so to speak, one flavor that is used by devices consuming up to 25W and another flavor intended for graphics cards that can supply 75W from the slot. The PCIe flavor is chosen at the time the motherboard is designed and presumably impacts the firmware (BIOS) while a x16 PCIe slot in and of itself does not mean it is designed to feed 75W to the inserted card.

Also one must note the PCIe system allows the cards power needs (from the slot) to be negotiated at boot time (but will still limited by the hardware design). It''s for this reason that low power (<25W) graphics cards can be placed in non graphics PCIe slots and still work.

However, the T620 supports graphics capable PCIe slots that can feed a full 75W to the inserted card should the card negotiate such needs at boot time. And since i don't believe the two variants of the T620 (GPU enabled server and non GPU server) impact the motherboard design/firmware where the difference seems to be limited to cabling , to know which slot to use you will have to check the hardware owners manual for clarification...

4 Posts

February 25th, 2014 06:00

Joe, can you tell me what you ended up using for a graphics card? I've got a Dell R620 and I want to try to get this working under Linux as well, but don't know which cards to try.

Thanks.

Jon

4 Posts

February 25th, 2014 07:00

Thanks. Were you driving multiple external monitors with those 610 cards or just using one for a monitor and the rest for CUDA?

Did you have any problems getting the card to fit? I notice that the slots use shorter brackets that most desktop cards that I'm used to using.

9 Posts

February 25th, 2014 07:00

I used a nvidia gt610 card(s).  I then played with xorg.conf to tell linux (xubuntu) which cards to use to output to which monitors.  Until it gets to X it still outputs via the onboard graphics, so if I need to do stuff before X I have to switch a cable over.  The advantage of Linux is it lets you be pretty explicit in which drivers are to be used with which cards and the cuda drivers are still there to be used in coding.  Whereas in windows I was getting lots of driver conflicts etc. I think you can make life easier by using a low end nvidia card which can use the same driver as the cuda card.

The relevant bit of my xorg is below, hope it helps, I think there is some other good information above, which may well allow you to end up with a better set up, I however have got it working (just) and it seems stable, so now I am leaving well alone!  Be careful in terms of updates as you want to make sure it doesn't upset the nvidia normal, or cuda drivers.

Section "Monitor"

# HorizSync source: edid, VertRefresh source: edid
Identifier "Monitor0"
VendorName "Unknown"
ModelName "Samsung SyncMaster"
HorizSync 30.0 - 81.0
VertRefresh 56.0 - 75.0
Option "DPMS"
EndSection

Section "Monitor"

# HorizSync source: edid, VertRefresh source: edid
Identifier "Monitor1"
VendorName "Unknown"
ModelName "DELL P2312H"
HorizSync 30.0 - 83.0
VertRefresh 56.0 - 76.0
Option "DPMS"
EndSection

Section "Monitor"

# HorizSync source: edid, VertRefresh source: edid
Identifier "Monitor2"
VendorName "Unknown"
ModelName "Samsung SyncMaster"
HorizSync 30.0 - 81.0
VertRefresh 56.0 - 75.0
Option "DPMS"
EndSection

Section "Device"
Identifier "Device0"
Driver "nvidia"
VendorName "NVIDIA Corporation"
BoardName "GeForce GT 610"
BusID "PCI:66:0:0"
Screen 0
EndSection

Section "Device"
Identifier "Device1"
Driver "nvidia"
VendorName "NVIDIA Corporation"
BoardName "GeForce GT 610"
BusID "PCI:67:0:0"
EndSection

Section "Device"
Identifier "Device2"
Driver "nvidia"
VendorName "NVIDIA Corporation"
BoardName "GeForce GT 610"
BusID "PCI:66:0:0"
Screen 1
EndSection

Section "Screen"
Identifier "Screen0"
Device "Device0"
Monitor "Monitor0"
DefaultDepth 24
Option "Stereo" "0"
Option "nvidiaXineramaInfoOrder" "CRT-1"
Option "metamodes" "CRT-1: nvidia-auto-select +0+0"
SubSection "Display"
Depth 24
EndSubSection
EndSection

Section "Screen"
Identifier "Screen1"
Device "Device1"
Monitor "Monitor1"
DefaultDepth 24
Option "Stereo" "0"
Option "nvidiaXineramaInfoOrder" "CRT-1"
Option "metamodes" "nvidia-auto-select +0+0"
SubSection "Display"
Depth 24
EndSubSection
EndSection

Section "Screen"
Identifier "Screen2"
Device "Device2"
Monitor "Monitor2"
DefaultDepth 24
Option "Stereo" "0"
Option "nvidiaXineramaInfoOrder" "CRT-0"
Option "metamodes" "CRT-0: nvidia-auto-select +0+0"
SubSection "Display"
Depth 24
EndSubSection

4 Posts

February 25th, 2014 08:00

So the low profile cards don't fit? I was thinking that would ensure they fit. What dimensions will fit? 

9 Posts

February 25th, 2014 08:00

I now have 3 external monitors running from two 610 cards (the cards had two outputs each).

The cuda card is not involved in any graphics output, it is I believe a very bad idea to use it for graphics output and simulations at the same time.

I had no fitting problems, but I do remember double checking when buying as the low profile cards do not fit (I think).

Also as I mentioned, I think you can possibly make your life a little easier by using a quadro rather than geforce card as I believe they share a driver see http://en.community.dell.com/support-forums/servers/f/956/p/19502455/20377597.aspx#20377597

9 Posts

February 27th, 2014 02:00

I believe, standard, non low profile cards, but perhaps your motherboard/box layout is different to mine?

May 12th, 2014 07:00

I have just installed a cheap PNY NVIDIA Quadro NVS 315 1GB in a 16x PCIe slot. It's working well with 2 big monitors in my Dell Poweredge T620.

The trick is to disable default Dell graphics card in BIOS menu and wait a little when restarting. But now BIOS menu has stopped working, and if I want to modify it I have to reset configuration via internal jumper what is a<ADMIN NOTE: Profanity removed as per TOU>.

Conclusion: external graphics cards work well, but custom Dell BIOS menu stops working.

4 Posts

May 12th, 2014 08:00

I got this card NVIDIA NVS 315 - Graphics card - Quadro NVS 315 - 1 GB DDR3 - PCIe 2.0 x16 low profile - DMS-59 (A7008623) and it worked fine once I disabled the onboard video in the BIOS.

June 4th, 2014 02:00

As I told in former post, BIOS and iDRAC menus dissapear when the NVIDIA NVS card was activated disabling internal graphics card, it's a minor issue for me (it can be restored via jumper).

But after trying a pair of days the card I noticed my computer very noisy, fans speed were around 2000rpm continuosly and when I restored the original graphic card the pc was totally silent (less than 1000rpm). After reading a lot in internet I reached the conclusion that the 'intelligent' Dell thermal algorithm is very silent in normal situations, but when it detects any card in a PCI slot fans increase speed (it thinks that is a TESLA big card, not a normal graphic card). There is no way to avoid this behaviour with the newest BIOS and iDRAC, so I came back to the crumpy Dell original card SILENCE IS IMPORTANT AT WORK :(


PD: It would be nice if Dell updates the thermal algorithm in next bios or iDRAC versions to keep silence with small PCI cards.

548 Posts

June 4th, 2014 02:00

@oligoelemento, all of the server class PCIe cards i have seen are passively cooled. And if you look into the specs for such server cards, they require big airflows to remain cool. On this basis, i very much doubt that Dell would expose any of it's cooling parameters in a firmware update as i suspect Dell would see it as risking card/machine warrrenty...

And as much as i would like to see such cooling parameters exposed in an official BIOS, the only way i see access to such parameters is if someone brighter than me managed a home grown BIOS hack for our Dell servers and documented how to reproduce the hack (as i'm not into blindly loading hacked BIOSes on my machines)...

Guess we can always hope [:?]

No Events found!

Top