GTX 670 4 GB - How to enable CUDA and openCL

[xav8tor]

Sorry for a new thread, but the question got buried elsewhere in another thread due to all of the excitement.

First issue: 4 GB 670 installed. Runs fine but CUDA Z returns CUDA not found error. How to I enable CUDA? I can't find a clear answer. Dumb it down for me because I am no driver expert.

Secondly, with 4 GB I understand I can enable openCL with a fix but netkas is a bit over my head too. Luxmark returns openCL device not found error. One fix I saw there had a link to a "dirty" download in Ru and that scares the crap out of me. I do have one or two apps that may take advantage. Can I enable openCL on this card? If so, how?

Thanks

[Asgorath]

Originally Posted by xav8tor

Sorry for a new thread, but the question got buried elsewhere in another thread due to all of the excitement.

First issue: 4 GB 670 installed. Runs fine but CUDA Z returns CUDA not found error. How to I enable CUDA? I can't find a clear answer. Dumb it down for me because I am no driver expert.

Secondly, with 4 GB I understand I can enable openCL with a fix but netkas is a bit over my head too. Luxmark returns openCL device not found error. One fix I saw there had a link to a "dirty" download in Ru and that scares the crap out of me. I do have one or two apps that may take advantage. Can I enable openCL on this card? If so, how?

Thanks

The Apple OpenCL framework explicitly disallows cards that have more than 2GB of RAM. As you can see here:

http://netkas.org/?p=1161

Netkas figured out a way to get around that check. Note that this isn't a driver thing, it's in the Apple framework itself. I believe that link is to a modified version of the framework file. If you don't trust the download, you can just make the modifications yourself.

[xav8tor]

Originally Posted by Asgorath

The Apple OpenCL framework explicitly disallows cards that have more than 2GB of RAM. As you can see here:

http://netkas.org/?p=1161

Netkas figured out a way to get around that check. Note that this isn't a driver thing, it's in the Apple framework itself. I believe that link is to a modified version of the framework file. If you don't trust the download, you can just make the modifications yourself.

Yeah, I saw that but the instructions, such as they are, confuse me. He says:

"P.S. doesnt applies to gtx670/gtx680, they have a different problem"

If it doesn't apply, where's the fix?

Also, what about CUDA?

[Asgorath]

Originally Posted by xav8tor

Yeah, I saw that but the instructions, such as they are, confuse me. He says:

"P.S. doesnt applies to gtx670/gtx680, they have a different problem"

If it doesn't apply, where's the fix?

Also, what about CUDA?

No idea, sorry. I would check the netkas.org forums, I think I've seen info about how to get CUDA working for the Kepler cards.

[xav8tor]

Originally Posted by Asgorath

No idea, sorry. I would check the netkas.org forums, I think I've seen info about how to get CUDA working for the Kepler cards.

Tried that too already. Registered so I could DL a fix but never got the confirmation email. Man, I am so ready for the Pro update next year!

[netkas]

Originally Posted by xav8tor

Tried that too already. Registered so I could DL a fix but never got the confirmation email. Man, I am so ready for the Pro update next year!

Mayebe you didnt read in the head of my forum, a link for registering users with this bold text - So, if u dont want to be automaticaly flagged as spambot and then wait weeks for activation - choose complicated nickname, and you will be fine.


Also, if u open my site, you will see that there is a fix for gtx670/680 opencl, in you case you need both kepler fix and 2gb+ fix, god, when will people learn to read..

[xav8tor]

Originally Posted by netkas

Mayebe you didnt read in the head of my forum, a link for registering users with this bold text - So, if u dont want to be automaticaly flagged as spambot and then wait weeks for activation - choose complicated nickname, and you will be fine.


Also, if u open my site, you will see that there is a fix for gtx670/680 opencl, in you case you need both kepler fix and 2gb+ fix, god, when will people learn to read..

Trust me, with two grad degrees, one in languages, and one in law, I can read, write, and even do some simple arithmetic! I did read the registration guidelines, followed them (using same handle as here), and no, nothing went into my spam folder.

You deserve great, HUGE credit for your efforts but no need for the condescending tone. I saw the fix but you must admit, clarity is lacking for the less knowledgeable folks like myself. As I said above, I am unable to dechiper the fix how to's, like the following:

-----------
long story short – by default opencl doesnt work on gk104 but works on apple’s gk107 from new mbps.

Now it works on gk104 (gtx670-gtx690)
Update: obviously ML drivers are much better (still need a patch so)

-----------

1) for 32-bit part:
find 3d 01 00 00 80 73 e7
replace with 3d 01 00 00 80 90 90
yes, its just last two bytes to disable the check


2) for 64-bit part, for lion 
find ff 50 68 48 39 d8 73 e5
replace with ff 50 68 48 39 d8 90 90


for mountain lion
find ff 50 68 4c 39 e0 73 e7
replace with ff 50 68 4c 39 e0 90 90


P.S. doesnt applies to gtx670/gtx680, they have a different problem
--------------

OK, maybe with the above I am mixing apples and oranges, but I was told there is a manual alternative to the "dirty download" below. If dirty download is anything like a dirty b*mb, I don't want to go near it!
----------------
got opencl working on gtx670!

so far its only for mbp2012 drivers, replace file inside GeForceGLDriver.bundle with this one http://rghost.ru/39181304

that's a dirty hack workaround just disables the check for some error, which i believe lays inside kernel side of drivers, so be carefull with it

Now also for ML GM - http://rghost.ru/39181502

-------------

I've done the MS tech support thing years ago, so I am not completely clueless troubleshooting computers, even built a few, but maybe I am just stupid on these matters. I don't get the above. ML needs a patch, but the patch won't work. What am I missing here?

And again, any clues ref CUDA?

Thanks in advance. I really am trying here. Really I am. All I am asking for is something like:

1) go here (link) and download this.
2) install (how/where)
3) then do this and you're good to go.

[netkas]

>Trust me, with two grad degrees, one in languages, and one in law, I can read, write, and even do some simple arithmetic! I did read the registration guidelines, followed them (using same handle as here), and no, nothing went into my spam folder.

Good for you.

Actualy you need both fixes, for 2gb+ and patched libclh.dylib


Then if you are on mountain lion, then there is a thread on enabling cuda, on my forum. You dont need a registration to read it.

Lastly, if you pm me your username, i can manualy activate you.

[MacVidCards]

yeah, CUDA fix is simple as all get out

as posted by Rominator

no download, just read

takes under 30 seconds to do

[xav8tor]

Originally Posted by netkas

Actualy you need both fixes, for 2gb+ and patched libclh.dylib

Then if you are on mountain lion, then there is a thread on enabling cuda, on my forum. You dont need a registration to read it.

Lastly, if you pm me your username, i can manualy activate you.

PM sent.

OK, for openCL, you're saying I need to manual make the changes with a hex editor AND install this libclh file, correct?

Ref CUDA, I'll look for the Rominator post.

Thanks for being patient. This is new ground for me. On Win7 via BC, everything works - all of my recent upgrades.

If I get this stuff working, maybe you can help me get a High Point USB 3 card going too! Like everything else, in Win, it just works, and if not, I can almost always fix it. IMHO, with the dumbing down of OS X after SL. Lion, for example, is harder to use for some things. Took me ten minutes to find my OS disk (Libraries) in Finder! And what's up with this bassackwards mouse scrolling? Rant over.

Update: OK you guys were right. CUDA fix was as easy as can be. Now for a try at openCL.

[1ricca]

god, when will people learn to read..

I really appreciate all of the great things that have been done thanks to you but, I really don't get the attitude. Your site layout is an abomination of site design. I do wordpress for a living and could not find anything without scrubbing through hundreds of posts. You don't make links to your previous posts either when you refer to them in new ones.

And frankly, you could have done a re-wrap of what to do in this guy's situation right here in this forum in less time it took to insult his intelligence. I bought the same card he got and this is the first thing that shows up on google about getting this card to work on a mac pro. You could have at least posted the link we needed in between your smug replies.

If you ever need a web guy to fix that site and justify the arrogance, PM me.

[chiefroastbeef]

Originally Posted by 1ricca

I really appreciate all of the great things that have been done thanks to you but, I really don't get the attitude. Your site layout is an abomination of site design. I do wordpress for a living and could not find anything without scrubbing through hundreds of posts. You don't make links to your previous posts either when you refer to them in new ones.

And frankly, you could have done a re-wrap of what to do in this guy's situation right here in this forum in less time it took to insult his intelligence. I bought the same card he got and this is the first thing that shows up on google about getting this card to work on a mac pro. You could have at least posted the link we needed in between your smug replies.

If you ever need a web guy to fix that site and justify the arrogance, PM me.

agree 100%

[ClassObject]

Originally Posted by chiefroastbeef

agree 100%

+1

[Davirus]

I'm experiencing similar problems.

I recently bought an ASUS GTX 670. When I patch the .kext to make OpenCL works, it works, but it disable any CUDA technology.

When I restore the kext to the original one, I get CUDA but OpenCL its disabled.

[GeneralAntilles]

Originally Posted by 1ricca

I bought the same card he got and this is the first thing that shows up on google about getting this card to work on a mac pro.

This always seems to be how it happens.

The threads and posts that contain only the question that needs answering and the knowledgable party's reference to some other thread or post with the answer (never with any links, though) or just some snarky reply always end up at the very top of searches.

I get why people like netkas and Rominator react the way they do. I've been in that situation, and the deluge of of PMs, emails, and threads asking the same questions over and over is exhausting. Some folks will never be able to help themselves no matter how much you help them. They will always want their hands held.

This doesn't help the people willing to put in the due diligence to try to answer their own questions, though, and this is why in other communities I've been involved in we've always had wikis in place to collect and summarize the important information (in this case, how to modify OpenCL, which drivers to install, and which program files to change to allow cards with >2GB of VRAM to run well).

My first question: Is there a wiki? If not, is there an existing wiki we could latch on to for GPU-related how-tos?

I'm quite willing to put in the time to summarize the steps needed to get a 4GB GTX 600-series card working happily in OS X if someone will point to the information relevant to the task.

[GermanyChris]

Originally Posted by GeneralAntilles

This always seems to be how it happens.

The threads and posts that contain only the question that needs answering and the knowledgable party's reference to some other thread or post with the answer (never with any links, though) or just some snarky reply always end up at the very top of searches.

I get why people like netkas and Rominator react the way they do. I've been in that situation, and the deluge of of PMs, emails, and threads asking the same questions over and over is exhausting. Some folks will never be able to help themselves no matter how much you help them. They will always want their hands held.

This doesn't help the people willing to put in the due diligence to try to answer their own questions, though, and this is why in other communities I've been involved in we've always had wikis in place to collect and summarize the important information (in this case, how to modify OpenCL, which drivers to install, and which program files to change to allow cards with >2GB of VRAM to run well).

My first question: Is there a wiki? If not, is there an existing wiki we could latch on to for GPU-related how-tos?

I'm quite willing to put in the time to summarize the steps needed to get a 4GB GTX 600-series card working happily in OS X if someone will point to the information relevant to the task.

We could use a wiki like OSX86, just because your gigabyte worked OTB doesn't mean a power color or something else will..

So call that a +1

[MacVidCards]

http://forums.macrumors.com/showthread.php?t=1440150

This guy started the thread just to answer these sorts of questions. Fermi & Kepler are cousins so anything posted there should be relevant.

[1ricca]

I think that's a wonderful idea doing a wiki. There should really be a wiki for all of the hackintosh stuff and then we could latch onto that but that whole gang has the same issues we've described here, maybe with a few less issues and some real sticky threads. I'll look around a bit and maybe some others here can too. I would definitely contribute.

[GermanyChris]

Originally Posted by 1ricca

I think that's a wonderful idea doing a wiki. There should really be a wiki for all of the hackintosh stuff and then we could latch onto that but that whole gang has the same issues we've described here, maybe with a few less issues and some real sticky threads. I'll look around a bit and maybe some others here can too. I would definitely contribute.

The Hackintosh and alt video cards folks are "kissin cousins" they just don't realize it yet.

[dia3olik]

hi netkas!
is the opencl fix for gtx680 4gb card still needed in 10.8.2 with the new drivers?
thx!

[xav8tor]

Originally Posted by dia3olik

hi netkas!
is the opencl fix for gtx680 4gb card still needed in 10.8.2 with the new drivers?
thx!

As far as I can tell, 10.8.2 does not remove the >2 GB restriction. Although one forum member gave a good shot at writing a guide to getting OpenCL (and CUDA, which is easy) running, it is still a bit unclear and a number of posts continue to pop up around the WWW wherein people describe bricking their Mac Pros trying to enable OpenCL, which happened to me. Fortunately I had a backup handy, but have not yet tried enabling OpenCL again because there still is not a totally clear step by step guide for doing it and I can't risk another disaster with multiple projects underway.

The guide needs to avoid technoslang and be patently clear on exactly what files to download/change/delete/leave where they are, etc. The hex edit process should also be clearly explained. Ideally, screenshots of each and every step would be included.

I now have an app that would greatly benefit from OpenCL, so if anyone is up to the challenge of writing the process as described, I'm sure many GTX 6XX users with more than two GB would be very grateful.

[thevode]

I have the GTX 670 in a Macpro 3.1

I want to activate cuda for blender but I can't.

Where is the rominator post to fix cuda??, I'm looking for it.

xav8tor, did you get it?.

xav8tor = xavi?, are you from catalonia???, porque si es así me ahorras el mal trago de hablar inglés como el culo.

Thanks and regards

[Asgorath]

http://forums.macrumors.com/showthread.php?t=1440150

[dtechlogic]

Espero que hay solucion para esto. Yo estoy usando un gtx 670 2GB