OSX GPU limits

[3dworks]

just as tip: netstor builts solid and reliable mac compatible PCI expansion chassis at a much lower cost than cubix. also, if you compare a 'trashcan' mac pro with similar benchwell results with a refurbished and tuned 5.1 mac pro, the latter will always be cheaper and has much more expansion potential. things may change once a dual CPU mac pro will arrive, but for now buying the newest pro model from apple in no option for me.

markus

[Tutor]

just as tip: netstor builts solid and reliable mac compatible PCI expansion chassis at a much lower cost than cubix. also, if you compare a 'trashcan' mac pro with similar benchwell results with a refurbished and tuned 5.1 mac pro, the latter will always be cheaper and has much more expansion potential. things may change once a dual CPU mac pro will arrive, but for now buying the newest pro model from apple in no option for me.

markus

I’m no expert but my Hackintosh has been rock solid stable going on almost two years. My main decision at the time of build was due to the GPU restrictions of the new Mac Pro trash cans.

That said, I do have other Mac’s I can rely / fall back on if there were a serious issue as well as my Hackintosh being a dual boot Mac / Windows box. So I’m pretty well covered for work spaces.

Also before building my Hackintosh I tried to work with my existing 2008 Mac Pro and with a supplemental power supply successfully ran three GPU’s (one was an old Radon used for boot screen). Only one card was double slot though (had a GTX 570 and an even older single slot Nivida 800 or something).

I think the time, money, and effort spent trying to get a Mac Pro maxed out for Octane use is a losing proposition, especially if you need more than two cards. Even if I could afford it, I would never spend $5,674 on old technology and only have a single GPU (albeit a Titian X) to show for it.

If possible, I would at least wait till Octane 3 is released this summer to see how that will work with AMD cards. Then you can get a new Mac Pro trash can with the money you save not having to purchase an expander box or a separate $1000 GPU.

I love Mac’s but the new build I’ll be making this summer as a dedicated render box will probably be Windows based to allow possibility for 4 GPU’s at a cheaper cost than it would be to try and configure such arrangement on a Mac (I decided not to run a Hackintosh on the new build I'm planning but that is for reasons other than stability).

I am wondering how many GPU , card, OSX van andle, thinking about going on Windows of it can handle more....


Oxs mtlion |2 Geforce GTX780ti | i7 3770 | 32GB

Thumbs Up!

Question: But now, what Windows platform?

I too don't have a 2013 MP trash can, so I'm not sure what it can do "and frankly, my dear, ... ." (from Gone With the Wind). However, I've been hacking computer hardware since the mid-1980s. I have three 2007 MP2,1 that I've literally hacked [ http://forums.macrumors.com/attachment. ... 1392319138 & http://forums.macrumors.com/attachment. ... 1394607445 ]. I have one 2009 MP4,1 that I've hacked three times: (1) http://www.computerworld.com/article/24 ... -pro-.html & http://browser.primatelabs.com/geekbench2/266838; (2) then into a 2010 MP & (3) later into a 2012+ MP [ http://browser.primatelabs.com/geekbench2/422811 ]. A fourth hack is still in process on the 2009 MP4,1. I'll be posting that on my thread here: http://forums.macrumors.com/showthread.php?t=1333421 . I also know just a tiny bit about hackintoshing [ http://browser.primatelabs.com/geekbench2/500630 & http://browser.primatelabs.com/geekbench3/558552 / http://www.insanelymac.com/forum/topic/ ... ch-scores/]. I also run modified Linux and Windows, as well as Atari and Amiga, systems and currently have eight 8xGPU processor rendering systems (one of which I've benchmarked 29 times on OctaneBench for an Avg score of 895 and a high score of 905 and a low score of 867) [ https://render.otoy.com/octanebench/sum ... GTX+780+Ti ] . The characteristic that all eight of those systems have in common is that the GPUs within each system are identical to one another, tho' each system may have different groups of identical GPUs. For example one system is all classic Titans; another is all 780 TI ACX SC OCs, etc. is I am building two more massively parallel GPU computing systems with Amfeltec splitters cards (but not the GPUs themselves) inside of my two quad CPU/32-core SuperMicros [ see first pic, below ]. Amfeltecs products are the lowest cost GPU extension options by a long shot. Obviously, that rack of GPUs (in my pic, below) will not fit within that box. However, I like to think outside of the box and these 8x GTX 780 6Gs will too.

A classic MacPro (cMP) has four PCIe slots, only one of which was intended by Apple to run one double width PCIe video card. However, one can hack those cMPs to run internally up to 3 double width GPUs - IF ADEQUATELY POWERED. cMP can run at least four GPU processors with additional hardware, such as the Amfeltec Clusters and Splitters [ http://forums.macrumors.com/showthread. ... 21&page=49 - See posts #1222 - 1241]. Mavericks*/ can support between 4-5 CUDA GPUs depending on the particular card's IO space implementation [ see discussion of variables here: viewtopic.php?f=40&t=43597&start=140 ]. I can run, at least, 2 GTX 590s (dual GPU video cards) in any of my cMPs**/ . [See second pic, below]

Answer: Tyan and Supermicro have the best preconfigured platforms for massively parallel GPU computing systems. But they're not cheap unless you're willing to do much research and learn to live outside of the box containing others ideas of beauty and forgo that new computer smell, or learn to do your homework and build your own.

*/Given that I have 24 rendering systems, I don't have the time to play that game where each Apple update breaks the Nvidia web drivers (since the Nvidia web drivers are tied to a particular Apple update). That's why I'm not in a hurry to upgrade my four Apple systems to a later version of Mavericks or even Yosemite. But keep in mind that on Macs the OS level determines which drivers will work, and thus that determines which GPUs you can run in them successfully.

**/ I run the overwhelming majority of my 116 GPU processors under Windows and Linux in systems that I've built.

P.S. 1. For the hackintoshers - as indicated in my thread here - viewtopic.php?f=40&t=43597&start=140, not only does the motherboard determine how many GPUs can be run successfully, but, among other factors, the OS also affects that. Since GPU processing is a new frontier, you have to be willing to take risks to determine what will work with what. As an alternative, you may have to learn to borrow GPUs from your nearby friends for your experimentation.
P.S. 2. As to which video cards to use, I recommend that one's GPU cards be of the same version/model/feature set to avoid IO space, i.e., lack of resource issues.
P.S. 3. I am not aware of any GPU expansion box solution, cluster, riser, splitter, etc. that can extend a computer system's ability to handle a particular number of GPUs above the ability of the motherboard's IO implementation to support that number of GPUs; such add-on solutions appear currently to only provide additional slots into which to plug GPUs (and/or other PCIe devices) where the inhabitants of the physical PCIe slots of that motherboard (and other non-obvious PCIe users such as, but not limited to, memory, SAS and SATA) have not already exhausted all IO reserves.
P.S. 4. Of the GPUs that I own, the GTX 780 TI ACX SC OC appears to be a IO resource hog and, in particular, does not play well with different GPUs as far as IO resource limitations are concerned.

[balt]

Hey Tutor,

that's an impressive setup you've got! And I thought I was rivalling the NSA with my quad 780Ti setup... haha.

Have you ever come across a Yosemite solution for the GPU limitation? I just upgraded to Yosemite and the bloody OS only finds 3 GPUs now (of 5). Fortunately I was anticipating problems and have a boot mirror with 10.8.5...

Also, on 10.8.5 I get kernel panics fairly often, always the same driver stack, usually the same addresses (see below - usually NVDAResman is the kext on top). This is using the latest released nvidia drivers still running on 10.8.5. Any suggestions on how to get this to be more stable? This is just having octane standalone running rendering...

Cheers

- Balt

Kernel Extensions in backtrace:
com.apple.iokit.IOSurface(86.0.5)[650AA894-6B30-3299-A9BB-DAD71712551B]@0xffffff7f95e4e000->0xffffff7f95e5ffff
com.apple.NVDAResman(8.1.6)[39D35403-42FB-3F08-999C-9866938D6B3A]@0xffffff7f9555f000->0xffffff7f95802fff
dependency: com.apple.iokit.IOPCIFamily(2.8)[6C1D646D-7E5E-3D7F-A557-2CBA398FF878]@0xffffff7f9517d000
dependency: com.apple.iokit.IONDRVSupport(2.3.7)[F16E015E-1ABE-3C40-AC71-BC54F4BE442E]@0xffffff7f9554b000
dependency: com.apple.iokit.IOGraphicsFamily(2.3.7)[9928306E-3508-3DBC-80A4-D8F1D87650D7]@0xffffff7f95508000
com.apple.nvidia.gk100hal(8.1.6)[347201ED-EC77-3189-B256-B3403CFCBB06]@0xffffff7f9580e000->0xffffff7f95b3ffff
dependency: com.apple.NVDAResman(8.1.6)[39D35403-42FB-3F08-999C-9866938D6B3A]@0xffffff7f9555f000
dependency: com.apple.iokit.IOPCIFamily(2.8)[6C1D646D-7E5E-3D7F-A557-2CBA398FF878]@0xffffff7f9517d000
com.apple.GeForce(8.1.6)[D77AC325-C0E3-33AB-A64C-6020322B9A15]@0xffffff7f963fd000->0xffffff7f964cafff
dependency: com.apple.NVDAResman(8.1.6)[39D35403-42FB-3F08-999C-9866938D6B3A]@0xffffff7f9555f000
dependency: com.apple.iokit.IONDRVSupport(2.3.7)[F16E015E-1ABE-3C40-AC71-BC54F4BE442E]@0xffffff7f9554b000
dependency: com.apple.iokit.IOPCIFamily(2.8)[6C1D646D-7E5E-3D7F-A557-2CBA398FF878]@0xffffff7f9517d000
dependency: com.apple.iokit.IOGraphicsFamily(2.3.7)[9928306E-3508-3DBC-80A4-D8F1D87650D7]@0xffffff7f95508000

[philliplakis]

Hey Tutor,

that's an impressive setup you've got! And I thought I was rivalling the NSA with my quad 780Ti setup... haha.

Have you ever come across a Yosemite solution for the GPU limitation? I just upgraded to Yosemite and the bloody OS only finds 3 GPUs now (of 5). Fortunately I was anticipating problems and have a boot mirror with 10.8.5...

You will never have them back unfortunately in Yosemite, Mavericks might give you 4. And I'm pretty sure if you took 2 cards out leaving 3 only 2 will show... Thats what was happening in my testing...

Can i ask which Expansion you are using?

[balt]

Has Apple actually made a statement to that effect that they're never going to support big numbers of GPUs Not in El Capitan either? If so, that will be the end of Apple in my work environment.

I'm using an Amfeltec cluster.

Cheers

- Balt

[philliplakis]

Has Apple actually made a statement to that effect that they're never going to support big numbers of GPUs Not in El Capitan either? If so, that will be the end of Apple in my work environment.

I'm using an Amfeltec cluster.

Cheers

- Balt

Yeah exactly.. The interesting thing is i had 4 in a eGPU solution that works in yosemite and another had 2 titan z's working through thunderbolt but the cMP is crimped


. Ive just PM'd you a work around.

[cbmotion]

Hi Folks - i had the same ugly Problem - came from Mountain Lion with perfectly working Netstor NB255A over PCIe and good speed with 4 GTX titan inside.
Yesterday updated to yosemite and didnt heard about the GPU limitations before.
Now i only see the GTX 680 ind the Mac and only two Titan in the external box.
Apple said they can not help - it is third oarty problem - thank you apple - but before updatetd it works perfect.
the problem is that i need for expl apple motion updatetd but this only work with the newer system.
i read that mavericks before 10.9.4 maybe can handle both - 4 x titan and the new motion
so i will try
but i got an terminal line for Mac pro 6.1 that will stop the limitations for the GPU - but only when you use thunderboldt

here it is - maybe it helps someone

nvram tbt-options=\<00\>

i think apple will not have any Professional user in the future - so i had to change to windows after 24 Years in the Mac world - what a ....

cheers
carsten

[sparkworld]

If I understand this above discussion it would seem our Macs are SOL. We upgraded to Mavericks to get the two Titans to work with the latest Nvidia drivers. I had hoped that the Amfeltec racks that we use with our PC's would be the solution, but they crash on boot and the very nice support guy at Amfeltec has no idea what to do with Macs. Maybe, by the tone of the conversations around this, it's just as well, as Apple clearly doesn't want to play with us anymore. Makes me nuts to have to switch systems but we need more GPU's for our animations.

If any of you have a clue how to get the Amfeltec system to work with Mavericks or Yosemite please let me know. Much appreciated,

[cbmotion]

Hi - it is me again
i installed Mavericks 10.9.3 because the supporter from Netstor said after this version the GPU limitation starts.
and surprise - it didnt work - so i think only the 10.9. will be limited free - but that doesnt help me because i need a least 10.9.2 to run some new programms.
but the work goes on and i need all my 4 titans to render - and i went back to my old good an stable Mountain Lion
and see it works fine again.
So my hope is that apple comeing back to the old believes and take the Professional back in their programm and El Captain has no GPU limits any more
but i didnt really believe it.

i don't want to go on the windows side ;-(

cheers
carsten

[balt]

Hey guys,

have any of you tested this with OSX El Capitan? Are we still limited to one GPU? Or is it properly supporting this, or is there a workaround for multiple GPU's?

Cheers

- Balt