Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

OpenCL, Radeon 4870, & 2006 Mac Pro

The Rominator said:
I have now tried the linked OpenCl benchmarking app on both OEM 4870 and flashed XFX 1 Gig...locks up with either when it comes time to test card.

Either App is flawed or not designed to run in 421a

Will hopefully have 432 sometime later today.
Click to expand...

Thanks Rom. Would you post back in this thread when you can test the 2006 Mac Pro & 4870 with 432?
 
I was expecting that result on the current status of the 3870. But itmay not be the end of the road. Does anybody here has contacts with Ati that can be approached for driver development questions and requests?
 
gugucom said:
I was expecting that result on the current status of the 3870. But itmay not be the end of the road. Does anybody here has contacts with Ati that can be approached for driver development questions and requests?
Click to expand...

yeah it sucks. But I'll bet dollars to donuts that a future rev of the drivers supports the 3870. Hmm, donuts. Now I want to go to Krispy Kreme. Dammit.

;-)
 
gugucom said:
I was expecting that result on the current status of the 3870. But itmay not be the end of the road. Does anybody here has contacts with Ati that can be approached for driver development questions and requests?
Click to expand...

At one time I did but sadly no more.
 
The Rominator

its not 421a fail, its apple fail, gui freezes when app starts testing amd gpu.

ssh still works well, so I made testing via ssh, (10a432, 4850)

CL_DEVICE_NAME: Intel(R) Core(TM)2 Quad CPU Q9450 @ 2.66GHz
CL_DEVICE_VENDOR: Intel
Now computing - please be patient....
time used: 14.841523
Number of elements computed: 2097152
CL_DEVICE_NAME: Radeon HD 4870
CL_DEVICE_VENDOR: AMD
Now computing - please be patient....
time used: 28.595200
Number of elements computed: 2097152
Now checking if results are valid - please be patient....
:) Validate results test passed - GPU=CPU :)


from dmesg, after test finished
...
0x00000000
0x00000000
0x00000000
0x00000000
0x00000000
0x00000000
** GPU Debug Info End **


clearly a drivers fault
 
netkas said:
a little bench from apple's examples, the only one which works on radeons

run via terminal, otherwise it cant find .cl file

http://rapidshare.de/files/48210843/transpose.zip.html

$ time ./transpose
Performing Matrix Transpose [256 x 4096]...
Bandwidth Achieved = 2.711743 GB/sec
Results Validated!

real 0m0.439s
user 0m0.071s
sys 0m0.026s
Click to expand...

./transpose
Performing Matrix Transpose [256 x 4096]...
Bandwidth Achieved = 35.047244 GB/sec
Results Validated!


Daddy like :D
 
OK, so we have OpenCL, even if we don't have a card...

It occurs to me that even if we have 3870s, and the drivers don't support them yet, we STILL can run OpenCL apps.

The OS supports using our CPU cores for OpenCL. Slower than a GPU, but at least there's support for it, right?

From that OCLInfo app at InsanelyMac:

1 OpenCL platform found!

[Platform 0]
Name: Apple
Vendor: Apple
Version: OpenCL 1.0 (Jul 15 2009 23:07:32)
Profile: FULL_PROFILE


1 OpenCL device found!

[Device 0]
Name: Intel(R) Xeon(R) CPU X5355 @ 2.66GHz
Vendor: Intel
Type: CPU
Device Version: OpenCL 1.0
Driver Version: 1.0
Compute Units: 8
Work Group Size: 1
Clock: 2660 MHz
Global Memory (Total): 6144 MB
Global Memory (Host): 6144 MB
Global Memory (PCIe): 0 MB
Local Memory: 16 KB
Cache Size: 4096 KB
Cache Line Size: 64 Bytes
Available: Yes
Double-Precision: Yes
Extensions:
cl_khr_fp64
cl_khr_global_int32_base_atomics
cl_khr_global_int32_extended_atomics
cl_khr_local_int32_base_atomics
cl_khr_local_int32_extended_atomics
cl_khr_byte_addressable_store
cl_APPLE_gl_sharing
cl_APPLE_SetMemObjectDestructor
cl_APPLE_ContextLoggingFunctions
Click to expand...

When GPUs are added to the mix, they SHOULD, in theory, distribute tasks to those additional "compute units" to the degree that an extension is supported in the Compute Units available.

So I guess it's better than nothing, right?

JP
 
Sorry guys, I'm a little lost. I have a MacPro 1,1 with a 3870 in it right now, has it been confirmed that it's NOT OpenCL compatible?

I have a good deal lined up on a 4870, but I don't want to go thru with it unless I know it will be OpenCL compatible in my MacPro 1,1. Has anyone confirmed this to be the case yet? :confused:
 
I believe it's been confirmed that it does not work right now. That does not mean ATI won't update drivers and make it compatible. I believe that's where it's at right now.
 
bozz2006 said:
I believe it's been confirmed that it does not work right now. That does not mean ATI won't update drivers and make it compatible. I believe that's where it's at right now.
Click to expand...

Thanks, I guess I'll pass on the 4870 until I know for sure it will work with OpenCL on my MacPro 1,1.
 
slicecom said:
Sorry guys, I'm a little lost. I have a MacPro 1,1 with a 3870 in it right now, has it been confirmed that it's NOT OpenCL compatible?

I have a good deal lined up on a 4870, but I don't want to go thru with it unless I know it will be OpenCL compatible in my MacPro 1,1. Has anyone confirmed this to be the case yet? :confused:
Click to expand...

I made a post about opencl in SL, read it carefully http://netkas.org/?p=164
 
I am having trouble following as well. I see people are asking about both the 3870 and the 4870 now. This thread is about OpenCL and the 4870, specifically in the 2006 Mac Pro. Are we still waiting for a clear answer? Perhaps those running tests could better explain the results.
 
benborman said:
I am having trouble following as well. I see people are asking about both the 3870 and the 4870 now. This thread is about OpenCL and the 4870, specifically in the 2006 Mac Pro. Are we still waiting for a clear answer? Perhaps those running tests could better explain the results.
Click to expand...



I don't use ATI hardware, but OpenCL definitely works in the 06/07 Mac Pro, as per my post above.
 
grue said:
./transpose
Performing Matrix Transpose [256 x 4096]...
Bandwidth Achieved = 35.047244 GB/sec
Results Validated!


Daddy like :D
Click to expand...

Re-ran when my CPUs weren't pegged out in the middle of video compression:

Loki:transpose srg$ ./transpose
Performing Matrix Transpose [256 x 4096]...
Bandwidth Achieved = 38.837567 GB/sec
Results Validated!
 
THe other benchmarking tool has been updated and now works right with multiple GPUs:




Number of OpenCL devices found: 3
OpenCL Device # 0 = GeForce 8800 GT
Device 0 is an: GPU with max. 1500 MHz and 112 units/cores
Now computing - please be patient....
time used: 0.683 seconds

OpenCL Device # 1 = GeForce GTX 260
Device 1 is an: GPU with max. 1400 MHz and 216 units/cores
Now computing - please be patient....
time used: 0.365 seconds

OpenCL Device # 2 = Intel® Xeon® CPU X5365 @ 3.00GHz
Device 2 is an: CPU with max. 3000 MHz and 8 units/cores
Now computing - please be patient....
time used: 3.094 seconds

Now checking if results are valid - please be patient....
Validate test passed - GPU results=CPU results
 
Something isn't right:


Number of OpenCL devices found: 3
OpenCL Device # 0 = GeForce 8800 GTS 512
Device 0 is an: GPU with max. 1650 MHz and 128 units/cores
Now computing - please be patient....
time used: 0.652 seconds

OpenCL Device # 1 = GeForce GTX 285
Device 1 is an: GPU with max. 1476 MHz and 240 units/cores
Now computing - please be patient....
time used: 3.176 seconds

OpenCL Device # 2 = Intel(R) Xeon(R) CPU 5160 @ 3.00GHz
Device 2 is an: CPU with max. 3000 MHz and 4 units/cores
Now computing - please be patient....
time used: 6.024 seconds

Now checking if results are valid - please be patient....
:) Validate test passed - GPU results=CPU results :)
logout

[Process completed]
 
The Rominator said:
Something isn't right:


Number of OpenCL devices found: 3
OpenCL Device # 0 = GeForce 8800 GTS 512
Device 0 is an: GPU with max. 1650 MHz and 128 units/cores
Now computing - please be patient....
time used: 0.652 seconds

OpenCL Device # 1 = GeForce GTX 285
Device 1 is an: GPU with max. 1476 MHz and 240 units/cores
Now computing - please be patient....
time used: 3.176 seconds

OpenCL Device # 2 = Intel(R) Xeon(R) CPU 5160 @ 3.00GHz
Device 2 is an: CPU with max. 3000 MHz and 4 units/cores
Now computing - please be patient....
time used: 6.024 seconds

Now checking if results are valid - please be patient....
:) Validate test passed - GPU results=CPU results :)
logout

[Process completed]
Click to expand...


That a side effect of Apple's good ol ramp-up effect?
 
Darned Throttling

grue said:
That a side effect of Apple's good ol ramp-up effect?
Click to expand...

Good question...or results are transposed. The 8800GTS is in a 4X lane, would make more sense if it had that score.

Anyhow, in line with OP...here is now working result of First Gen with OEM Apple 4870:

................ OpenCL Bench V 0.25 by mitch ...........
...... C2D 3GHz = 12 sec vs Nvidia 9600GT = 0,93 sec ......
... time results are not comparable to older version! .....
...........................................................

Number of OpenCL devices found: 2
OpenCL Device # 0 = Radeon HD 4870
Device 0 is an: GPU with max. 750 MHz and 4 units/cores
Now computing - please be patient....
time used: 4.184 seconds

OpenCL Device # 1 = Intel(R) Xeon(R) CPU 5160 @ 3.00GHz
Device 1 is an: CPU with max. 3000 MHz and 4 units/cores
Now computing - please be patient....
time used: 6.177 seconds

Now checking if results are valid - please be patient....
:) Validate test passed - GPU results=CPU results :)
logout

And this is in 421a...I'm just gonna buy The Final Golden Release on friday. (there's 2 Apple Stores within 5 miles of me...I would be supremely lazy if I didn't just drive down and buy it for $30 on Friday)

EDIT:

Yep, you were right, it's the throttling.

I used a G70 card as "Helper"...so it won't show up in OpenCL stuff.

First run with GTX:

OpenCL Device # 0 = GeForce GTX 285
Device 0 is an: GPU with max. 1476 MHz and 240 units/cores
Now computing - please be patient....
time used: 3.165 seconds

2nd run, hard on heels of first:

Number of OpenCL devices found: 2
OpenCL Device # 0 = GeForce GTX 285
Device 0 is an: GPU with max. 1476 MHz and 240 units/cores
Now computing - please be patient....
time used: 0.258 seconds


Guess I should try that 4870 again, see if it's throttling.
 
Code: 
...........................................................
.................. OpenCL Bench V 0.25 by mitch ...........
...... C2D 3GHz = 12 sec vs Nvidia 9600GT = 0,93 sec ......
... time results are not comparable to older version! .....
...........................................................

Number of OpenCL devices found: 3
OpenCL Device # 0 = Radeon HD 4870
Device 0 is an: GPU with max. 750 MHz and 4 units/cores 
Now computing - please be patient....
time used:  4.127 seconds

OpenCL Device # 1 = GeForce GT 120
Device 1 is an: GPU with max. 1400 MHz and 32 units/cores 
Error: clBuildProgram for device # 1 
ERROR NUMBER = -11
Sadly 4870 is really SLOW and I don't think it is throttling. From the benchmark it seems the driver really ONLY use 4 unites/cores as it claims to.
 
If it technically "works" I'm not too worried. ATI will update drivers. And at this point, i doubt it matters how well OpenCL-supported software runs, as it will take time to optimize it. hopefully ATI can update drivers soon, and hopefully software writers can update their code soon. I am just happy that the 2006 Mac Pro and the Apple OE 4870 are OpenCL compatible.

Thanks everyone
 
Hopefully so on the ATI part....

If not... it looks like some are getting the Nvidia GT260 series and beyond up and running. However results sound more like a mixed bag and not as straight forward as the ATI ROM flash..
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back