Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Coming to a nMP near you - CUDA !
- Thread starter MacVidCards
- Start date
- Sort by reaction score
Nando4,
I am eager to find any other folks trying eGPU with nMP
Sadly, this link now takes me to someone with sound problems
SOLD!!!! Perfect reason to buy a new Mac pro now with all the expensive options thrown in. Thanks OP for the warning and thank you Baclays for being there when i need you... 
Last edited:
Here's the revised link, it was moved to the 'Work In Progress' section.
http://forum.techinferno.com/provis...bps-tb2-akitio-thunder2-osx10-10-[sedor].html
Well now, this has been a most interesting read.
I got into all this simply because I needed to reduce render times.
It's not the first time. Back in "the day" when I was an Electric Image user and burned hours and days of my life tweezing a bajillion arcane settings just to shave time off renders when frame render times could be measured in hours, the only reasonable approach was to build a render farm. Unless one had ginormous amounts of capital, that meant buying up used G3 based macs off eBay and loading them up with 3rd party processor cards, kludged together cooling solutions, and basically spending as much time keeping it all running as I did creating content to render.
The whole GPU compute thing is a godsend. With cards like the 970 hitting a very nice sweet spot in the price / performance ratio, all of a sudden it's relatively practical to pull all this off. At some point I'll probably build another box or two and stuff a pair of 970's in those as well.
For those without access to cMP's, this whole mini thing has a fair amount of appeal. The host computer for rendering is pretty much irrelevant. The GPU grinds for a few minutes then pukes out a frame, so it doesn't matter if the hard drive is a 5400 RPM laptop drive, or if the ram is somewhat limited. If the render can't fit into the VRAM of the GPU in question it's a moot point anyways.
I'm aware that MVC probably chaffs a lot of people in the wrong way. The comments about "every few weeks he posts this fastest new card blah blah" and to some extent that's true, however, he's done his homework, and offers a service to a particular group of people with a particular set of needs. Very old school Apple in the best of the Think Different sort of way.
However, after fighting with OS X 10.10.3 and Server.crap 4.1 for the last 4 days (wherein if one enables a password policy for LDAP it changes the LDAP master's own password and craters the entire LDAP across the network in the process, not to mention the utter inability to force a password change to users any other way, or the 23, yes count em 23 times Server.Crap has crashed, and 4 complete reinstalls from scratch just trying to get it configured).
We often forget the essence of practical engineering, wherein the idea is to have something fulfill a function, and work reliably, and if one is lucky enough to be able to devote an appropriate amount of time to the task, elegantly.
There are also varieties of engineer types on this planet. Those that have a love of understanding how things work, and those that chose the profession as a means to a fiscal end. MVC falls into the former, and the people making the "shipshit" decisions in Watchertino these days seem to be falling into the latter.
I, for one will take 20 MVC type personas on a forum any day of the week over anything coming out of the hype machine of a company that used to embody the very notion of don't settle for the status quo, change it, but now seems to be little more than standing on it's own previous shoulders of what was once a righteous giant slayer and has now become that which it once refused to be.
I don't even have an MVC EFI card. For my needs having a lowly GT120 in the #4 slot works just fine for what I'm doing. However, I can appreciate the diligence applied to figuring out how to make it all work, and if as a result a bit of an ego manifests itself in the process, I can file that under the natural byproduct of being utterly human. It's that sort of focus that gets stuff done and just comes with the turf.
I got into all this simply because I needed to reduce render times.
It's not the first time. Back in "the day" when I was an Electric Image user and burned hours and days of my life tweezing a bajillion arcane settings just to shave time off renders when frame render times could be measured in hours, the only reasonable approach was to build a render farm. Unless one had ginormous amounts of capital, that meant buying up used G3 based macs off eBay and loading them up with 3rd party processor cards, kludged together cooling solutions, and basically spending as much time keeping it all running as I did creating content to render.
The whole GPU compute thing is a godsend. With cards like the 970 hitting a very nice sweet spot in the price / performance ratio, all of a sudden it's relatively practical to pull all this off. At some point I'll probably build another box or two and stuff a pair of 970's in those as well.
For those without access to cMP's, this whole mini thing has a fair amount of appeal. The host computer for rendering is pretty much irrelevant. The GPU grinds for a few minutes then pukes out a frame, so it doesn't matter if the hard drive is a 5400 RPM laptop drive, or if the ram is somewhat limited. If the render can't fit into the VRAM of the GPU in question it's a moot point anyways.
I'm aware that MVC probably chaffs a lot of people in the wrong way. The comments about "every few weeks he posts this fastest new card blah blah" and to some extent that's true, however, he's done his homework, and offers a service to a particular group of people with a particular set of needs. Very old school Apple in the best of the Think Different sort of way.
However, after fighting with OS X 10.10.3 and Server.crap 4.1 for the last 4 days (wherein if one enables a password policy for LDAP it changes the LDAP master's own password and craters the entire LDAP across the network in the process, not to mention the utter inability to force a password change to users any other way, or the 23, yes count em 23 times Server.Crap has crashed, and 4 complete reinstalls from scratch just trying to get it configured).
We often forget the essence of practical engineering, wherein the idea is to have something fulfill a function, and work reliably, and if one is lucky enough to be able to devote an appropriate amount of time to the task, elegantly.
There are also varieties of engineer types on this planet. Those that have a love of understanding how things work, and those that chose the profession as a means to a fiscal end. MVC falls into the former, and the people making the "shipshit" decisions in Watchertino these days seem to be falling into the latter.
I, for one will take 20 MVC type personas on a forum any day of the week over anything coming out of the hype machine of a company that used to embody the very notion of don't settle for the status quo, change it, but now seems to be little more than standing on it's own previous shoulders of what was once a righteous giant slayer and has now become that which it once refused to be.
I don't even have an MVC EFI card. For my needs having a lowly GT120 in the #4 slot works just fine for what I'm doing. However, I can appreciate the diligence applied to figuring out how to make it all work, and if as a result a bit of an ego manifests itself in the process, I can file that under the natural byproduct of being utterly human. It's that sort of focus that gets stuff done and just comes with the turf.
Are there any news about nMP and eGPU?
Id like to use this card http://www.asus.com/Graphics_Cards/STRIXGTX980DC2OC4GD5/ which i actually only need in windows.
Would be nice if someone have some info on that matter.
Also, a general newb question, can i use asus or gigabyte cards or card has to be evega.
cheers
Id like to use this card http://www.asus.com/Graphics_Cards/STRIXGTX980DC2OC4GD5/ which i actually only need in windows.
Would be nice if someone have some info on that matter.
Also, a general newb question, can i use asus or gigabyte cards or card has to be evega.
cheers
Glad you asked !
Netkas and I have made some wonderful progress. We are set to turn the eGPU world in a new direction.
On the good news side, we have not only got GTX980 and Titan-X running on TB2 Macs, we can enable full screen output over the card's ports. In past efforts most TB2 Macs wouldn't allow screen output on Nvidia cards. We can enable it for 9/10 boots. We have also been able to greatly simplify the kext mods that are currently required.
In the "meh" department we are only getting bootscreens on Maxwell cards 30-40% of the time. In most cases the machine continues boot process and the card self-inits as in the long Nvidia thread at top of stickies. KEPLER cards have better odds. We are still hopeful to get better response from Maxwell.
But in the bad news department I have still been unable to get cards to work in eGPU and Windows on nMP. And I have spent a GREAT deal of time in EFI shell trying to make a tunnel through the various PCI bridges to get card to work. On 2014 Mini it is just a matter of WHEN to plug in TB2 cable. I have been running Far Cry 4 and Witcher 3 on Mini at 4K, something completely impossible without eGPU.
It takes awhile to figure things out, especially when testing partner is in a very different time zone. He is all excited and eager to test when I am sitting dazed at the keyboard at 3AM.(and vice versa)
I'll post some recent screenshots tonight.
I'm really glad that you two guys are putting so much effort into this. It is even better to hear that you are making such a good progress.
I hope you will crack nMP soon since I'm checking GPUs every day and waiting for D-day to come.
Looking forward to see the screenshots.
Awesome work! Does this currently require modified kexts or are you using a vanilla osx?
Did you also try out AMDs or only Nvidias at this time? I know, most flashable AMDs wouldn't be an improvment over the FirePros but would be interesting though...
Some easy usable eGPUs might actually get me back to buying Apple stuff since a long time.
Did you also try out AMDs or only Nvidias at this time? I know, most flashable AMDs wouldn't be an improvment over the FirePros but would be interesting though...
Some easy usable eGPUs might actually get me back to buying Apple stuff since a long time.
7970 worked as well, but pretty slow, looks like issue similar to mp3,1, or not. Didn't investigate amd that much.
Okey Doke
Attachments
Really awesome! How many cards can you connect to a single nMP? If i understand correct, we need something with your medicine in order for this to work, right? Will you have plug-in solutions for sale?
thank you,
C
More Apple control-freakism. eGPU machines have limits on pixel clocks that make no sense. Same card in Mac Mini in Windows does 4K MST with ease.
This is on MST display, on SST it would likely do 4K 60 Hz or 5K on the Dell.
Could it be the limited pcie bandwidth? The MP has no lanes left where the mini will have most of its.