I would like to mount a Tesla K20 on a Mac Pro Mid 2012 where I set a double boot MacOSX10.8 Mounting Lion and Ubuntu 12.04. At the moment I don’t have the 2x6pin PCIe power connector for Mac that seems difficult to find in Italy. Still I tried to mount the Tesla and I’ve got a kernel panic for MacOS.
I have three questions: the kernel panic will be fixed when I will conect the power cable on the Nvidia card? It is necessary to connect both plugs PCIe? And third; I have seen that the motherboard can provide only 300W for the all the PCI express slots, I have already an ATI Radeon GT 5500 graphic card, will my Mac be able to support both card, ATI and Nvidia Tesla K20?
Thankyou very much for your help ☺.
You might want to do a search yourself – it seems that a Radeon GT 5450 requires 19W to run… I wouldn’t think that’s the maximum wattage used, but gives an idea. So if the motherboard provides 300W for PCI-E slots you should be fine, but I wouldn’t stress the Radeon card at the same time as the K20, especially if it’s already near 100% load. This of course assumes Mac’s play happy with an ATI and NVIDIA cards together… do not have experience with Macs to know the answer.
Needless to say you do need the PCI-E power cable to have the Tesla K20 perform as designed… kernel panic is most likely because you didn’t connect the power cable. On some older cards you could get away with connecting a single PCIe plug, but I wouldn’t attempt it on hardware that expensive.
The PCI-Express specification only allows 75W to be supplied through a single card slot. A 6-pin PCI-E power connector is rated to provide another 75W. (The 8-pin connector does 150W, which is why you see it appear on graphics cards that need more than 225W total.) This is why the card requires two 6-pin connectors + the power from the card slot.
Thankyou for your replies, they helped me to understand something more.
Therefore, installing the Tesla card on a PCI Express 2.0 x16 card slot and connecting two 6 pin PCIe power cable is required for the card to be seen by the OS or some of this requirements are only recommened? For example if I use PCIe 2.0 x4 slot (with a mechanical support for 16-lane card) instead of x16, does the Tesla will work or not?
I am sorry but I’m trying to understand if my problems with the Tesla K20 installation is due to a hardware or software incompatibility.
I really don’t know. I neither use Macs nor Linux. I have a Tesla K20 workstation board (that’s the actively cooled version) running next to a Quadro K5000 under Windows 7 64-bit in my test system.
It just occurred to me that the product and driver download pages, which should be the first point for information about supported configurations, do not list Mac OSX support for Tesla boards.
What happens when otherwise correctly installing the board in a Mac desktop system running a Linux distribution is beyond my expertise.
As I mentioned earlier, you should power the card with both of the connectors to ensure that it functions correctly. If you place an x16 card in an x4 w/ mechanical 16 lanes it should* still work, just at x4 speeds, so you’d lose on bandwidth. Unless there’s some inherent incompatibility with the card and you’re motherboard, as long as you can see it in Linux on your Mac, you should be able to install the drivers and be on your way with the restriction on bandwidth as I mentioned.
Frankly, if you want to use it to it’s full potential, I wouldn’t run it on an x4 slot, and would look for an alternate system, especially if your CUDA application(s) will be memory bound or close to it.
Yeah, I think the conclusion here is that you are in unsupported territory, so you’ll have to experiment (with no guarantee of success) or switch to a supported system. If your plan is to run the K20 with Linux (and not OS X), putting together a dedicated computer with a proper power supply might be easier than trying to kludge it into the Mac Pro.
So far, I’ve only built CUDA systems from parts as needed. If building a computer today for a K20, I would probably build a Sandy Bridge-E (Core i7-3820) system with 16 GB of RAM, a cheap NVIDIA card to run the display (GeForce GT 640 Rev 2 is nice), a 128 or 256 GB Samsung 840 Pro SSD, and an 800W Antec power supply. I don’t have a motherboard recommendation since I haven’t assembled a Socket LGA-2011 system before, but I would look for one that uses PCI-Express 3.0 switches to provide a full x16 connection to three PCI-Express slots. Such a system would be a little more than you need for a single K20, but be able to take a second high end card if you need it. Should cost $1200 or less in the US.
Maybe someone reading this has a suggestion for an off-the-shelf tower appropriate for a K20?
Any of the HP Z workstations will do (except Z1) But i suggest for intensive computation and or graphics works use Z800 - Z820. There’s a lot of videos showing Z800 with Quadro K5000 & Tesla on Youtube.
I have MacPro 2008 and Tesla K20 (Graphic card is GT120).
I cannot use Tesla K20 when I connect or not connect the power cable.
I could mount the Tesla in 10.7.4 (but not use), but on 10.7.5 and later OS.X, kernel panic occurred.
Bootcamp and Windows 8.1 on MacPro 2008, I can use Tesla K20 normally.
So It is OK that I cannot use Tesla on Mac OS X, but I hope to avoid the kernel panic.
I want NVIDIA to make the Tesla drive for mac.
@Wakachan – from what I’ve seen online (no direct experience with Mac pro systems) it seems like the Mac Pro power supply is not sufficient to power a Tesla K20 (similar to GTX Titan/Titan Black/etc
Apparently there is a group that physically modifies cards to make them Mac compatible… but you’d still need external power of some sort.
Power supply might be a concern when it’s at full load. In re: kernel panics in OS X, that would seem to be card BIOS related… post on their forums with the kernel panic errors you get… chances are they might know what the issue is.