Oh, so you didn't have to put the EFI Bios onto the card then? which is where Mac Pro folks are running into problems.
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.
PCs GTX285/275/260/295 working in any MacPro....
- Thread starter netkas
- Start date
- Sort by reaction score
Oh, so you didn't have to put the EFI Bios onto the card then? which is where Mac Pro folks are running into problems.
EVGA GTX 295 Co op Edition works
This home PC was driven entirely by wanting to play with CUDA. Again, I have both a PowerMac G5 2.5 (will need to retire this box) and a MacBook Pro.
I find the GTX 295 Single PCB edition very fascinating. Above all, I find Apple's Leopard OS to be the most spectacular OS out there. Here's what I witnessed with the GTX 295 card under Leopard:
I ran 2 instances of the nbody CUDA demo simultaneously as such:
./nbody --device=0 & This will utilize the first GPU
./nbody --device=1 & This runs on the 2nd GPU
Afterwards, I ran iTunes with the visualizer enabled inside the window.
This is where I witnessed the fluidness of Leopard's OS (possibly the kernel) when moving windows around, minimizing and resizing. It remained fluid. The interactivity of the OS shined.
Now, I did the same test under Windows 7 and SUSE Linux 11.1. The fluidness is not quite there. Kudos to Apple for it's amazing OS. Kudos to BSD. Kudos to all the amazing folks who made Leopard what it is today.
The power savings of utilizing one PCB versus 2 discrete cards installed in 2 slots is another plus. My system utilizes 150 watts in idled mode. With one GPU running nbody, the wattage goes up to about 280 and 350 when both GPUs are running nbody.
This home PC was driven entirely by wanting to play with CUDA. Again, I have both a PowerMac G5 2.5 (will need to retire this box) and a MacBook Pro.
I find the GTX 295 Single PCB edition very fascinating. Above all, I find Apple's Leopard OS to be the most spectacular OS out there. Here's what I witnessed with the GTX 295 card under Leopard:
I ran 2 instances of the nbody CUDA demo simultaneously as such:
./nbody --device=0 & This will utilize the first GPU
./nbody --device=1 & This runs on the 2nd GPU
Afterwards, I ran iTunes with the visualizer enabled inside the window.
This is where I witnessed the fluidness of Leopard's OS (possibly the kernel) when moving windows around, minimizing and resizing. It remained fluid. The interactivity of the OS shined.
Now, I did the same test under Windows 7 and SUSE Linux 11.1. The fluidness is not quite there. Kudos to Apple for it's amazing OS. Kudos to BSD. Kudos to all the amazing folks who made Leopard what it is today.
The power savings of utilizing one PCB versus 2 discrete cards installed in 2 slots is another plus. My system utilizes 150 watts in idled mode. With one GPU running nbody, the wattage goes up to about 280 and 350 when both GPUs are running nbody.
EVGA GTX 295 Co op Edition works
This is truly amazing to see 2 nbody demos running simultaneously from the single PCB edition containing 2 GPUs and computing near 650 GFLOP/s combined. All this utilizing 355 watts (not including the displays). The graphics card makes less noise than a busy Power G5 2.5 GHz.
Attaching the photo.
Technology has come a very long way since the Power G5 days.
This is truly amazing to see 2 nbody demos running simultaneously from the single PCB edition containing 2 GPUs and computing near 650 GFLOP/s combined. All this utilizing 355 watts (not including the displays). The graphics card makes less noise than a busy Power G5 2.5 GHz.
Attaching the photo.
Technology has come a very long way since the Power G5 days.
Attachments
This is all fascinating. The OpenGL scores are weak but in all other respects this is amazing, especially now we see a boot card is not needed.
Can you run the Cuda program MonteCarlomultiGPU as that will show the performance of (a) two cards (b) how the card is being clocked and power-managed. A Single 285 will show about 100,000 options per second, and my two 285s will show about 80000 each when both are up. I took a chance on going for a twin 285 set up but was always wondering about a 295.
10thz,
he is running a Hackintosh, so results vary from what we will likely see
(I'm gonna guess a Mac Pro will still need an EFI boot helper)
The slightly lower OpenGl numbers are in line with what could be predicted given numbers from 9800GX2 and a 4870x2. (slightly lower than base card)
Could you have a look at the trouble I am having in the "2 Gig Bug" thread and test yours with Halo, Doom3 or COD4? (or some other OpenGl game)
https://forums.macrumors.com/threads/756526/
he is running a Hackintosh, so results vary from what we will likely see
(I'm gonna guess a Mac Pro will still need an EFI boot helper)
The slightly lower OpenGl numbers are in line with what could be predicted given numbers from 9800GX2 and a 4870x2. (slightly lower than base card)
Could you have a look at the trouble I am having in the "2 Gig Bug" thread and test yours with Halo, Doom3 or COD4? (or some other OpenGl game)
https://forums.macrumors.com/threads/756526/
Rominator - I do not have those games - I just use these cards for CUDA apps and there the full 2G is recognized without difficulty in the calls to the card. The Hackintosh explains the boot success - thanks. For peak performance comparison against a pair of 285s, I just ran two lots of nbody targetting the separate GPUS as above, and this combo of a Mac 285 and a 2G PC 285 comes in at about 816 GFLOP/s, which is better than the 295, but obviously I had to spend rather more cash to get it. Shot attached (from Mac Pro 08, 8-core 2.8GHz)
Attachments
Here also is the result of CUDA deviceQuery and OpenGL EV for easy comparison with the new 295 results. EV is running off the Mac card. Note that the full (+)2G is recognized under CUDA on Device 1. I recall that when running EV off the PC card I got -2048M but all the tests worked with similar numbers. I am very impressed with the price/performance of the 295. To get the full 1.48GHz on the Mac card requires that a 3D app be up, as is the case here.
Edit: I just reran the twin nbody test on the 2X285 under XP32 under bootcamp. Despite being a mac fan I have to confess that (under CUDA 2.2, the Mac was 2.3) under Windows I was getting just over 900Gflop/s. I am not sure what to make of the difference.
Edit: I just reran the twin nbody test on the 2X285 under XP32 under bootcamp. Despite being a mac fan I have to confess that (under CUDA 2.2, the Mac was 2.3) under Windows I was getting just over 900Gflop/s. I am not sure what to make of the difference.
Attachments
The card is being managed by the driver I believe. I'm not runnning a 3D app when running this if this is helpful.
Here's the result from MonteCarloMultiGPU:
$ ./MonteCarloMultiGPU
main(): generating input data...
main(): starting 2 host threads...
main(): waiting for GPU results...
main(): GPU statistics
GPU #0
Options : 128
Simulation paths: 262144
Time (ms.) : 1.801000
Options per sec.: 71071.626912
GPU #1
Options : 128
Simulation paths: 262144
Time (ms.) : 1.804000
Options per sec.: 70953.436019
main(): comparing Monte Carlo and Black-Scholes results...
L1 norm : 3.083966E-06
Average reserve: 361.436058
TEST PASSED
Shutting down...
Press ENTER to exit...
Running MonteCarlo to utilize only 1 GPU returns:
GPU options per sec.: 80782.583541
I removed the device-od from the NVDAResman.kext plist. You're correct. Thanks. Initially, I ran an egrep for 05e0 -- the one for the GTX 295 (dual PCB design) and found it here as well as NVDANV50Hal.kext plist.
The 05eb entry is needed in NVDANV50Hal.kext plist though.
Updated my previous post #236 and removed the entry for NVDAResman.kext plist.
Thanks.
This is quite a setup you have. I always wondered about the power utilization for a setup like this -- especially both N-Body demos running simultaneously. The watts used for the Mac Pro system -- not including the display(s).
Somewhere in a box I have a power meter - when I find it I will report usage with 0,1,2 copies of nbody running. My case is open with an external PSU to power the 2G 285 - I switch it off when not using it as my room gets rather warm. Thanks for the MonteCarlo numbers: twice 70k is a very good number for the outlay and my guess is you are heating your room a lot less.
Here are the wattage used from the GTX 295 setup: (excluding the display)
Note: I updated this. I thought I had my display connected in another plug behind the UPS. I moved it to ensure it wasn't included. The watts used are as follow.
System Idled: 100 watts
One N-Body demo on GPU1: 201 watts
Two N-Body demos on GPU1/2: 278 watts
~ 650 GFLOP/s
It's a lot of computation power for 280 watts. This is where technology amazes me.
I found my meter. Mac Pro 08 with GTX 285 driving 1920x1200 desktop, 16M RAM, 8 cores 2/8GHz, otherwise idle 210W (fluctuating around this value). With one nbody on Mac GTX 285 I get a draw of 332W.
The second 2Gig GPU is powered by a separate ATX PSU, which itself draws 25W with the fan going around, and about 50W with the GPU on and idling. This goes up to 140W running nbody. So the total power with both cards is about 472W fluctuating +-10 around this value. My guess is that a twin 285 config powered off a single PSU would run at around 450W.
I am now wondering why my injected (netkas tool) Palit 2G card consumes an extra 90W to run nbody whereas the EVGA 1G Mac card wants 120W more. Same clock speed, twice the memory, 25% less power. it also has two fans.
The second 2Gig GPU is powered by a separate ATX PSU, which itself draws 25W with the fan going around, and about 50W with the GPU on and idling. This goes up to 140W running nbody. So the total power with both cards is about 472W fluctuating +-10 around this value. My guess is that a twin 285 config powered off a single PSU would run at around 450W.
I am now wondering why my injected (netkas tool) Palit 2G card consumes an extra 90W to run nbody whereas the EVGA 1G Mac card wants 120W more. Same clock speed, twice the memory, 25% less power. it also has two fans.
Thank you for this info on the power utilization. I have a single socket quad core processor. The GT200 GPUs do require power. I'm amazed by Apple's hardware with 2 sockets and the powerful GTX 285 at idled state being around 210W. The numbers are about right in which 2 GTX 285 cards will require 100 more watts versus a GTX 295 card and considering that you have 2 populated CPU sockets.
For anyone interested, I can confirm after a bit of playing with a visiting Quadro 4800 in an 08 Pro today that
(a) a PC Quadro CX 4800 works fine with netkas injector with an 8800GT as boot card
(b) replacing the 8800 with an ATi card causes the sticking at a blue screen just before Finder window.
So no surprises there. I was surprised to see the Q needed only ONE 6-pin connector, which means that no power rejigging is needed for the 8800+Q combo in a Pro - this is efficient for 192 cores, though is reflected in clock speed of just under 1.2GHz. BUT, the Q peaks at about 285 Gflop/s running nbody under CUDA. It also likes to power down to 0.6 GHz when not running 3D stuff, in common with the Mac 285, and is double precision capable. Nice power-efficient card, though it reminded me we still need a driver update to get these cards going into fast mode more easily. I might have the time later this month to try that thing loading the 8800 ROM with Refit alone - has anyone done that and produced idiot-guide?
(a) a PC Quadro CX 4800 works fine with netkas injector with an 8800GT as boot card
(b) replacing the 8800 with an ATi card causes the sticking at a blue screen just before Finder window.
So no surprises there. I was surprised to see the Q needed only ONE 6-pin connector, which means that no power rejigging is needed for the 8800+Q combo in a Pro - this is efficient for 192 cores, though is reflected in clock speed of just under 1.2GHz. BUT, the Q peaks at about 285 Gflop/s running nbody under CUDA. It also likes to power down to 0.6 GHz when not running 3D stuff, in common with the Mac 285, and is double precision capable. Nice power-efficient card, though it reminded me we still need a driver update to get these cards going into fast mode more easily. I might have the time later this month to try that thing loading the 8800 ROM with Refit alone - has anyone done that and produced idiot-guide?
Hello mroy,
I own a Gainward gtx 295 with single pcb but this method is not working for me. I have invalid ATY_Init.kext message and invalid NVDANV50Hal.kext message when rebooting
I verified and Gainward has same string of evga co-op.
Any idea?
Thank you
I own a Gainward gtx 295 with single pcb but this method is not working for me. I have invalid ATY_Init.kext message and invalid NVDANV50Hal.kext message when rebooting
I verified and Gainward has same string of evga co-op.
Any idea?
Thank you
you need to reset the permissions on those kexts once you have modded them
even moving them in and out of extensions folder will make them useless to the System
an easy way is to download "kexthelper" and use it to reinstall those kexts
there are other ways too
as always, research is your friend
and you must still have another Nvidia card if you are on a Mac Pro.
even moving them in and out of extensions folder will make them useless to the System
an easy way is to download "kexthelper" and use it to reinstall those kexts
there are other ways too
as always, research is your friend
and you must still have another Nvidia card if you are on a Mac Pro.
The Rominator, first of all thank you!
My problem was that I didn't know what I have to search. Totally new Mac OSX for me, and for me was strange the message for ATY_Init.kext that I haven't modified.Maybe some internal call from NVDANV50Hal.kext that is the only file I modified.
Now I have all the informations (I hope)
I'll try
thank you again!
My problem was that I didn't know what I have to search. Totally new Mac OSX for me, and for me was strange the message for ATY_Init.kext that I haven't modified.Maybe some internal call from NVDANV50Hal.kext that is the only file I modified.
Now I have all the informations (I hope)
I'll try
thank you again!
I solved with kexthelper as The Rominator says, then I started the upgrade with 10.5.8 apple update and apart a problem with ps2 keyboard (solved with an old ps2 voodo kext reinstall) all went fine. GTX 295 is working!
The only little malfunction that I have is with usb logitech optical mouse that seems not perfect in its movement.
Maybe I have to search better if this is a common problem with hackintosh.
Thank you again The Rominator for your help
The only little malfunction that I have is with usb logitech optical mouse that seems not perfect in its movement.
Maybe I have to search better if this is a common problem with hackintosh.
Thank you again The Rominator for your help
Thank you 'The Rominator' for helping out fabio67. I added the following to my post #236 to ensure folks have the permissions set correctly:
Perform the following: (very important steps, especially after the mod to the kext above)
1. sudo su ;inside the terminal and if not already in sudo root
2. chown -R root:wheel /System/Library/Extensions/ATY_Init.kext
3. chmod -R 755 /System/Library/Extensions/ATY_Init.kext
4. touch /System/Library/Extensions
Wait a minute -- Leopard will refresh the cache.
I was doing this script based on an Apple web page for a while, but the latest kext helper apps make it a lot easier than this, and essentially use the same commands buried in a script.