Fastest Mac GPUs with full EFI boot screens, etc, GTX570 & GTX580

[freeflywing]

hey there, my first post here, but read the whole thread last night. intriguing stuff. I have a GTX 580 i brought for my 2012 Mac Pro. i used a splitter cable to garner the 8 pin from the standard 6 pin mac cable ( from the motherboard ). So in short my Palit 3GB gtx 580 is being directly powered by two 6 pin plugs from the motherboard and it seems to work well. However after reading some posts here, particularly from macvidcards, about the potential of burning out the motherboard i have taken it out ! So my question is thus . .

i am not trying to power more than one card and if my 580 wants to draw 244 w whats the best power supply to buy ?

it makes sense to me to try and put it in the extra CD drive bay because the graphics card cable lengths are so short. so i found this on amazon

http://www.amazon.com/450W-Graphics...392918&sr=8-1&keywords=juice+box+power+supply

so what do you think ? or should i just try to find a 2.5 GB 570 which seems to not to need the extra power ? thanks for your time all.

when you run game (D3) or 3d rendering, 580 draws more then 300w-350w normally. 570 requires 2 x 6 pins cable which you can just use the pci-e slot on the motherboard. the psu should be working and it wont hurt much on the air-flow, but i dont like it due to the small fan that make too much noise since mac pro is already a noisy machine. what i use to connect power to 580 is ocz zx, if you have better budget, corsair ax is better. they both are clean, no cable hanging around like the low end psu. right now, my psu is sitting next to the mac pro, i am thinking to get rid of everything in the dvd bay, and retrofit the psu in the dvd bay area. possibly i am not able to use the dvd shell if i do so, or else, i gotta do some modification on the shell.

 

[Asgorath]

when you run game (D3) or 3d rendering, 580 draws more then 300w-350w normally.

This is simply not true.

The TDP of a GTX 580 is in the 240W range. Slightly more than what the Mac Pro can deliver (i.e. 225W).

The TDP of a GTX 570 is in the 220W range. That is, a perfect fit for what the Mac Pro can deliver.

Honestly, at this point, I would recommend getting either a GTX 570 or a GTX 680. The 680 works out-of-the-box in Mountain Lion, and will be a much better choice in the long run. You're not giving up that much performance dropping down from a 580 to a 570, and it's definitely a safer choice overall.

 

[gegagome]

Installed 570GTX but not really

Hello fellow Mac Pro users

I bought an N570GTX (NVIDIA GeForce GTX 570) from Amazon last week and the mini pcie 6 pin to normal pcie 6 pin on ebay and at first glance everything seems to work fine.

Except for the fact that when rendering in Blender 3d (cycles) the compute device-NVIDIA GeForce GTX 570- (in Blender terms) doesn't seem to be making any difference in rendering time.

Everything's installed. I have the CUDA (CUDA Driver Version: 4.2.10) icon in system preferences and Photoshop CS6 has this GPU with OpenCL checked.

What I installed:
http://www.nvidia.com/object/macosx-270.00.00f06-driver.html
Both toolkit (required by Octane, which crashes on load) and cuda drivers from here: http://www.nvidia.com/content/cuda/cuda-downloads.html
Steps 2 and 3 from this url: http://tonymacx86.com/viewtopic.php?f=15&t=35148

More information:
This card has two dvi and one mini HDMI, and they all work, including the HDMI video output. Except for the audio through HDMI, which doesn't even recognize the HD TV device in system preferences > audio.

Finally I am doing all this using a Mac Pro 1,1
2 x 2.66 GHz Dual-Core Intel Xeon
9 GB 667 MHz DDR2 FB-DIMM
Mac OS X Lion 10.7.4 (11E53)
1 x ATI Radeon HD 2600
1 x NVIDIA GeForce GTX 570

Can anyone please help me determine why this card fails to crunch data faster, in 3d applications, such as Blender, and output audio through HDMI?

I can always return this card to Amazon and get something else
http://www.amazon.com/gp/product/B008809SUM/

Thanks!

 

[Asgorath]

Does Blender use CUDA or OpenCL?

 

[gegagome]

It uses both

Does Blender use CUDA or OpenCL?

I think it uses both CUDA and OpenCL but not simultaneously. I'm not an expert in Blender though.

 

[G4DP]

gegagome, did you check CUDA is functioning properly through Terminal?

 

[MacVidCards]

fire up CUDA-Z and see if CUDA is functioning

Smallux GPU will tell you if OpenCl is working

 

[mamasboy]

Tech & experience question: If I purchase the gtx570 as a gpu card and place it in slot 2 of 5,1 mac pro, while the 5870 is the gui card in slot one...

1. Yes - I know the 570 will need external power & I will only have the top slot left for I/O card and no raid card room.
2. When I turn off/on computer - how do I wire it so the psu turns off/on with the computer?
3. If I don't wire as in #2 above - do you turn on the psu prior to the macpro or does it matter?
4. I fully believe macvidcards deserves his $$$ and commend him on his invaluable services - but if i'm not needing the 570 for gui displays do i still need his flashed cards or is it the same experience with a purchase from newegg? Let me know if I'm missing something David - because if I am, I have no problems purchasing a card from you via ebay...

Any advice or experience would be appreciated...
Lonnie

 

[mamasboy]

ahhhh... I believe macvidcards will provide "PCIE 2.0 5.0 GT/s speed" and Newegg purchase will NOT on Mac Pro 5,1 - is this correct, David?

 

[MacVidCards]

ahhhh... I believe macvidcards will provide "PCIE 2.0 5.0 GT/s speed" and Newegg purchase will NOT on Mac Pro 5,1 - is this correct, David?

Yes, the difference between our cards and the Newegg ones are very simple.

Our cards offer full boot screen through OS screens while Newegg ones will remain dumb, deaf *& blind until an OS with drivers for them is loaded. If you are trying to solve System issues, you NEED an EFI GUI card.

Our cards offer full use of PCIE 2.0 speed available in Mac Pros. Cards run by just the OS driver run at half bus speed (i.e., 2.5 GT/s instead of 5.0 GT/s)

 

[lixuelai]

Tech & experience question: If I purchase the gtx570 as a gpu card and place it in slot 2 of 5,1 mac pro, while the 5870 is the gui card in slot one...

1. Yes - I know the 570 will need external power & I will only have the top slot left for I/O card and no raid card room.
2. When I turn off/on computer - how do I wire it so the psu turns off/on with the computer?
3. If I don't wire as in #2 above - do you turn on the psu prior to the macpro or does it matter?
4. I fully believe macvidcards deserves his $$$ and commend him on his invaluable services - but if i'm not needing the 570 for gui displays do i still need his flashed cards or is it the same experience with a purchase from newegg? Let me know if I'm missing something David - because if I am, I have no problems purchasing a card from you via ebay...

Any advice or experience would be appreciated...
Lonnie

HD5870 just for the GUI is overkill. You would save more money either 1) buying the modded card and selling the HD5870 or 2) sell the HD5870 and get a different card like say a HD2600 for the GUI. TBH upgrade from HD5870 to GTX570 isn't really worthwhile unless you need CUDA or play a game the HD5870 is particularly bad at.

If you are keen on keeping that HD5870 and get a new card along with it most if not any GTX570 will work. Your PCI-E throughput will be halved. You won't notice the difference however unless your apps require the bandwidth (you would know if you needed it then) or if you play Flight Sim X.

 

[mamasboy]

Thanks lixuelai - a thoughtful reply! I need Cuda & Mercury Playback for CS6 (AE & PP), and for Davinci Resolve. So question:

1. Would the "halved experience" (non-macvidcard) be noticed for CS6 or Resolve with 5870/570 combo?
2. How would a macvidcard 570 solo (no-additional gpu - if i sold the 5870) perform compared to #1 above scenario?

Thanks in advance - and thanks to you too David for the clarity!
Lonnie

 

[lixuelai]

Thanks lixuelai - a thoughtful reply! I need Cuda & Mercury Playback for CS6 (AE & PP), and for Davinci Resolve. So question:

1. Would the "halved experience" (non-macvidcard) be noticed for CS6 or Resolve with 5870/570 combo?
2. How would a macvidcard 570 solo (no-additional gpu - if i sold the 5870) perform compared to #1 above scenario?

Thanks in advance - and thanks to you too David for the clarity!
Lonnie

1. It is hard to say without having someone test for you. Typically there is very little difference (average is like 3%). But most testing are done with games. For GPGPU usage it might make a difference. I've seen it discussed in the context of distributed computing (via CUDA) where workloads are probably similar and difference was like 3% as well.

2. It will be at least as good as 1. The HD5870 does absolutely nothing really.

 

[xav8tor]

Yes, the difference between our cards and the Newegg ones are very simple.

Our cards offer full boot screen through OS screens while Newegg ones will remain dumb, deaf *& blind until an OS with drivers for them is loaded. If you are trying to solve System issues, you NEED an EFI GUI card.

Our cards offer full use of PCIE 2.0 speed available in Mac Pros. Cards run by just the OS driver run at half bus speed (i.e., 2.5 GT/s instead of 5.0 GT/s)

What about native support in Mountain Lion (but without boot screens)? Will 600 series cards run full or half speed? If half, can you mod a 680?

 

[MacVidCards]

What about native support in Mountain Lion (but without boot screens)? Will 600 series cards run full or half speed? If half, can you mod a 680?

I imagine that the 600 are running in Mac Pros at 2.5 GT/s, which is 1/4 of what they are capable of.

I might be able to get them to run at 5.0 GT/s with some fiddling but hardly worth the bother until I get EFI working.

And my senior genius tells me that the rMBP Kepler EFI does NOT include boot screen support. The Intel 4000 is set up to run display until the OS loads. So EFI boot screen secret may be further away than we make like, at least until something else with Kepler comes out for Mac.

 

[xav8tor]

I imagine that the 600 are running in Mac Pros at 2.5 GT/s, which is 1/4 of what they are capable of.

I might be able to get them to run at 5.0 GT/s with some fiddling but hardly worth the bother until I get EFI working.

And my senior genius tells me that the rMBP Kepler EFI does NOT include boot screen support. The Intel 4000 is set up to run display until the OS loads. So EFI boot screen secret may be further away than we make like, at least until something else with Kepler comes out for Mac.

OK, thanks. I'll benchmark with a couple of cross-platform apps when the card arrives.

 

[gegagome]

fire up CUDA-Z and see if CUDA is functioning

Smallux GPU will tell you if OpenCl is working

Hello MacVidCards and thank you for chiming in

Please see the following screenshots:

----------

gegagome, did you check CUDA is functioning properly through Terminal?

How would I check through terminal? Is there a command to do so?

 

[MacVidCards]

The CUDA-Z numbers show a 1.3 GB 570 running at normal speed (1300 SPGF) and at 2.5 GT/s.

I don't know exactly how to gauge the smallux scores as I don't use those benchmarks.

You did realize that the "tonymac" post is from last October, yes?

I don't know why people read that site, all they ever do is re-post things from Netkas.

And if you read up at Netkas, you will see that step 2 isn't needed anymore. So you may very well have buggered up OpenCl with some out of date "tonymac expertise"

And step 3 probably isn't needed either.

Get your news from the source, there are several peole who WISH they were smart, but all they ever do is repost and repackage things they read elsewhere. The learning curve at Netkas is rather steep, but at least it is "the real deal" and current.

 

[mamasboy]

To Macvidcards,

Hello David - I'm sure this info would be helpful for many other would be purchasers:

1.) Would you send me the details (here or via PM) on how I would send a new egg purchased gtx570 to you for your modification please, along with your fees and how you prefer payment. (check with card, paypal, etc...)

2.) from page 10 of this thread, is it that the 2.5GB card needs the additional OpenCl fix or is it that OpenCl does not currently work with or without "the fix"?

3.) If I go with the 1.28Gb version, and overclocking is not important - do you know which cards keep cooler?

Thanks,
Lonnie

 

[gegagome]

The CUDA-Z numbers show a 1.3 GB 570 running at normal speed (1300 SPGF) and at 2.5 GT/s.

I don't know exactly how to gauge the smallux scores as I don't use those benchmarks.

Did I download the correct smallux? If not, which one should I use?

Also, are the CUDA-Z results any good?

Thanks

----------

The CUDA-Z numbers show a 1.3 GB 570 running at normal speed (1300 SPGF) and at 2.5 GT/s.

I don't know exactly how to gauge the smallux scores as I don't use those benchmarks.

You did realize that the "tonymac" post is from last October, yes?

I don't know why people read that site, all they ever do is re-post things from Netkas.

And if you read up at Netkas, you will see that step 2 isn't needed anymore. So you may very well have buggered up OpenCl with some out of date "tonymac expertise"

And step 3 probably isn't needed either.

Get your news from the source, there are several peole who WISH they were smart, but all they ever do is repost and repackage things they read elsewhere. The learning curve at Netkas is rather steep, but at least it is "the real deal" and current.

I may have dismissed the post date, but it actually enabled OpenCL, before I did this it was greyed out in Photoshop.

But as you said, I may have bugged something.

Is there a way to reverse that change?

 

[gegagome]

The CUDA-Z numbers show a 1.3 GB 570 running at normal speed (1300 SPGF) and at 2.5 GT/s.

I don't know exactly how to gauge the smallux scores as I don't use those benchmarks.

You did realize that the "tonymac" post is from last October, yes?

I don't know why people read that site, all they ever do is re-post things from Netkas.

And if you read up at Netkas, you will see that step 2 isn't needed anymore. So you may very well have buggered up OpenCl with some out of date "tonymac expertise"

And step 3 probably isn't needed either.

Get your news from the source, there are several peole who WISH they were smart, but all they ever do is repost and repackage things they read elsewhere. The learning curve at Netkas is rather steep, but at least it is "the real deal" and current.

Does anyone know how to make this card render at fast speeds in Blender?
Thanks!

 

[freeflywing]

This is simply not true.

The TDP of a GTX 580 is in the 240W range. Slightly more than what the Mac Pro can deliver (i.e. 225W).

The TDP of a GTX 570 is in the 220W range. That is, a perfect fit for what the Mac Pro can deliver.

Honestly, at this point, I would recommend getting either a GTX 570 or a GTX 680. The 680 works out-of-the-box in Mountain Lion, and will be a much better choice in the long run. You're not giving up that much performance dropping down from a 580 to a 570, and it's definitely a safer choice overall.

lolz, well, tried it urself, killawatt always tells the truth. with the number you mentioned, the draw is still too heavy for the 6pin pci-e slots in the MB.

 

[Asgorath]

lolz, well, tried it urself, killawatt always tells the truth. with the number you mentioned, the draw is still too heavy for the 6pin pci-e slots in the MB.

What were you measuring exactly? The total system load on the power supply can easily be in the 300-350W range, but that doesn't mean it's all going to the GPU.

Given the fact that the TDP of 240-245W of the GTX 580 is more than the 225W that the Mac Pro can deliver by itself, then yes, the GTX 580 can and will draw too much power if you are just using the 6-pin connectors on the motherboard (with a 6-pin to 8-pin converter cable). This is why most people recommend getting the GTX 570, which has a TDP of ~220W and is pretty close to the performance of the GTX 580.

 

[Cindori]

guys, you can do SLI GTX 680 without any external or extra internal power supply. if you want. the Mac Pro PSU is 1000w. you only have to be clever with routing power. the wattage adds up.


But yes, the 2008-2010 Mac Pro's are tricky to route power for, that is why many people (including me) just say "go with a less powerful card" or "get an extra internal PSU if you want to run that card". it is purely because of the limitations of PCIe 6-pin power outlets (the 2 on the mobo) that we say this, not because we cant count watts.

its not worth it to play around with the internal power in the Mac Pro, I fried my mobo once just by connecting my aftermarket GPU cooler incorrectly.

 

[gegagome]

Blender won't render using CUDA

read blend: monkeys.blend
CUDA error: Unknown error
CUDA error: Invalid context in cuCtxDetach(cuContext)

Error that comes up in Blender when rendering using CUDA in Blender, despite the fact that it appears to be properly installed. See screenshots above.

Thanks