Murray,

I second your call for discussion which could lead to better informed hardware buying decisions to run Mathematica 8 optimally.

For those of us with an overwhelming desire to continue using Apple's MAC OSX systems the current situation (as I understand it so far) is that Mathematica 8 GPU support strategy leaves behind a large installed base of Intel Mac systems built until very recently when MacBooks switched to nVidia GPUs and iMacs used the more recent ATI GPUs (ATI Radeon HD 4670, HD 5670 & HD 5750).

The wonderful part of Mathematica 8 strategy is the terrific incentive for one to buy some new hardware with (one hopes) scintillating speed improvements.

So lets look for some data to support the buying decision! Using MathematicaBenchmark8 with my current machine:

Machine Name:          sydsmacbookpro
System:                MacOS X V 10.6.5 Snow Leopard (64-bit)
Date:                  November 15, 2010
Mathematica Version:   8.0.0
Benchmark Result:      0.41

Compare this to the current overall best and the Apple system best:

3.07 GHz Core i7-950 (8 Cores)                1.00
Windows 7 Pro (64-bit) Desktop

3.06 GHz Core 2 Duo E8435 (2 Cores)           0.73
iMac OS X Snow Leopard (64-bit) Desktop

2 * 2.26 GHz Quad Core Xeon E5520 (8 Cores)   0.69
Mac XServe OS X (64-bit) Server 

The interesting fact is if I set up Mathematica 8 to use both of my MacBooks cores to I get a 61% MathematicaBenchmark8 improvement (an impressive result).

Machine Name:          2-node homogeneous cluster
System:                MacOS X V 10.6.5 Snow Leopard (64-bit)
Date:                  November 15, 2010
Mathematica Version:   8.0.0
Benchmark Result:      0.66

Note that Mathematica assigned the name  "2-node homogeneous cluster" to my "parallel MacBook".

So the disappointing first comparison from my current system (best result of .66 using Mathematica 8 to the best MathematicaBenchmark8 result of 1.0) apparently limits my available upgrade in performance to 52%.

Now this is clearly nonsense. But without getting into the benchmark writing business it is the best one can do currently with WRI tools.

I hope someone at WRI will recognize the importance of totally upgrading benchmarking to take into account support for GPUs (and address the CUDA vs OpenCL issues) and parallelism (multi-core, multi-thread) support. Without a serious benchmark upgrade I fear the general discussion will not lead to actionable information.

My current conclusion with Apple products is that the best and safest hardware upgrade (for me) to optimize use of Mathematica 8 would be to go Apple Mac Pro with the latest nVidia GPU cards to be able to use CUDA (assuming $$s unconstrained). 

My preference for a number of reasons (including cost), however, would be the 27" iMac which has ATI HD 5750 graphics. 

Mathematica 8 docs say the HD 5750 is "supported" with Mathematica 8 OpenCL. The problem is there is no documentation I have found so far that gives me a comparison of what that means vs going nVidia and CUDA.

Any input from WRI and other MathGroup aficionados would be greatly appreciated.

Yours truly ... Syd Geraghty

Syd Geraghty B.Sc., M.Sc.
sydgeraghty@me.com
San Jose, CA

Mathematica 8.0 for Mac OS X x86 (64-bit) (November 6, 2010)
Licenses: L2983-5890, L3028-2592
MacOS X V 10.6.5 Snow Leopard
MacBook Pro 2.33 Ghz Intel Core 2 Duo  2GB RAM

On Nov 18, 2010, at 4:05 AM, Murray Eisenberg wrote:

> I'd like to see some discussion in this group about 
> advantages/disadvantages of CUDA vs. OpenCL, especially when used in 
> conjunction with Mathematica 8?  I'll have to upgrade my hardware from 
> an ATI board that's too old to support OpenCL, so I want to choose 
> wisely between ATI and nVidia, with now Mathematica 8 in the picture.

On Nov 19, 9:07 pm, Syd Geraghty <sydgerag...@me.com> wrote:
> Murray,
>
> I second your call for discussion which could lead to better informed hardware buying decisions to run Mathematica 8 optimally.
>
> For those of us with an overwhelming desire to continue using Apple's MAC OSX systems the current situation (as I understand it so far) is that Mathematica 8 GPU support strategy leaves behind a large installed base of Intel Mac systems built until very recently when MacBooks switched to nVidia GPUs and iMacs used the more recent ATI GPUs (ATI Radeon HD 4670, HD 5670 & HD 5750).
>
> The wonderful part of Mathematica 8 strategy is the terrific incentive for one to buy some new hardware with (one hopes) scintillating speed improvements.
>
> So lets look for some data to support the buying decision! Using MathematicaBenchmark8 with my current machine:
>
> Machine Name:          sydsmacbookpro
> System:                MacOS X V 10.6.5 Snow Leopard (64-bit)
> Date:                  November 15, 2010
> Mathematica Version:   8.0.0
> Benchmark Result:      0.41
>
> Compare this to the current overall best and the Apple system best:
>
> 3.07 GHz Core i7-950 (8 Cores)                1.00
> Windows 7 Pro (64-bit) Desktop
>
> 3.06 GHz Core 2 Duo E8435 (2 Cores)           0.73
> iMac OS X Snow Leopard (64-bit) Desktop
>
> 2 * 2.26 GHz Quad Core Xeon E5520 (8 Cores)   0.69
> Mac XServe OS X (64-bit) Server
>
> The interesting fact is if I set up Mathematica 8 to use both of my MacBooks cores to I get a 61% MathematicaBenchmark8 improvement (an impressive result).
>
> Machine Name:          2-node homogeneous cluster
> System:                MacOS X V 10.6.5 Snow Leopard (64-bit)
> Date:                  November 15, 2010
> Mathematica Version:   8.0.0
> Benchmark Result:      0.66
>
> Note that Mathematica assigned the name  "2-node homogeneous cluster" to my "parallel MacBook".
>
> So the disappointing first comparison from my current system (best result of .66 using Mathematica 8 to the best MathematicaBenchmark8 result of 1.0) apparently limits my available upgrade in performance to 52%.
>
> Now this is clearly nonsense. But without getting into the benchmark writing business it is the best one can do currently with WRI tools.
>
> I hope someone at WRI will recognize the importance of totally upgrading benchmarking to take into account support for GPUs (and address the CUDA vs OpenCL issues) and parallelism (multi-core, multi-thread) support. Without a serious benchmark upgrade I fear the general discussion will not lead to actionable information.
>
> My current conclusion with Apple products is that the best and safest hardware upgrade (for me) to optimize use of Mathematica 8 would be to go Apple Mac Pro with the latest nVidia GPU cards to be able to use CUDA (assuming$$s unconstrained).
>


Syd I have a one month old Macbook Pro with an Nvidia GPU but CUDAlink
doesn't seem to like this particular card (the standard GPU shipping
with this model mac, GeForce 320M)

In= CUDAQ[]
Out:= False

So it looks like this must work only on selected nvidia cards???

On Nov 20, 5:29 pm, truxton spangler <truxtonspangle...@gmail.com>
wrote:
> On Nov 19, 9:07 pm, Syd Geraghty <sydgerag...@me.com> wrote:
>
>
>
>
>
>
>
>
>
> > Murray,
>
> > I second your call for discussion which could lead to better informed h=
ardware buying decisions to run Mathematica 8 optimally.
>
> > For those of us with an overwhelming desire to continue using Apple's M=
AC OSX systems the current situation (as I understand it so far) is that Ma=
thematica 8 GPU support strategy leaves behind a large installed base of In=
tel Mac systems built until very recently when MacBooks switched to nVidia =
GPUs and iMacs used the more recent ATI GPUs (ATI Radeon HD 4670, HD 5670 &=
 HD 5750).
>
> > The wonderful part of Mathematica 8 strategy is the terrific incentive =
for one to buy some new hardware with (one hopes) scintillating speed impro=
vements.
>
> > So lets look for some data to support the buying decision! Using Mathem=
aticaBenchmark8 with my current machine:
>
> > Machine Name:          sydsmacbookpro
> > System:                MacOS X V 10.6.5 Snow Leopard (6=
4-bit)
> > Date:                  November 15, 2010
> > Mathematica Version:   8.0.0
> > Benchmark Result:      0.41
>
> > Compare this to the current overall best and the Apple system best:
>
> > 3.07 GHz Core i7-950 (8 Cores)                1.00
> > Windows 7 Pro (64-bit) Desktop
>
> > 3.06 GHz Core 2 Duo E8435 (2 Cores)           0.73
> > iMac OS X Snow Leopard (64-bit) Desktop
>
> > 2 * 2.26 GHz Quad Core Xeon E5520 (8 Cores)   0.69
> > Mac XServe OS X (64-bit) Server
>
> > The interesting fact is if I set up Mathematica 8 to use both of my Mac=
Books cores to I get a 61% MathematicaBenchmark8 improvement (an impressive=
 result).
>
> > Machine Name:          2-node homogeneous cluster
> > System:                MacOS X V 10.6.5 Snow Leopard (6=
4-bit)
> > Date:                  November 15, 2010
> > Mathematica Version:   8.0.0
> > Benchmark Result:      0.66
>
> > Note that Mathematica assigned the name  "2-node homogeneous cluster"=
 to my "parallel MacBook".
>
> > So the disappointing first comparison from my current system (best resu=
lt of .66 using Mathematica 8 to the best MathematicaBenchmark8 result of 1=
..0) apparently limits my available upgrade in performance to 52%.
>
> > Now this is clearly nonsense. But without getting into the benchmark wr=
iting business it is the best one can do currently with WRI tools.
>
> > I hope someone at WRI will recognize the importance of totally upgradin=
g benchmarking to take into account support for GPUs (and address the CUDA =
vs OpenCL issues) and parallelism (multi-core, multi-thread) support. Witho=
ut a serious benchmark upgrade I fear the general discussion will not lead =
to actionable information.
>
> > My current conclusion with Apple products is that the best and safest h=
ardware upgrade (for me) to optimize use of Mathematica 8 would be to go Ap=
ple Mac Pro with the latest nVidia GPU cards to be able to use CUDA (assumi=
ng$$s unconstrained).
>
> Syd I have a one month old Macbook Pro with an Nvidia GPU but CUDAlink
> doesn't seem to like this particular card (the standard GPU shipping
> with this model mac, GeForce 320M)
>
> In= CUDAQ[]
> Out:= False
>
> So it looks like this must work only on selected nvidia cards???

While CUDA is supported on your card (I believe), you'll still need to
download the relevant CUDA programming software!

So... (1) update to the most current version of the NVIDIA drivers at
nvidia.com, these will enable support for the latest version of CUDA
on your card, and (2) go to the developer section of NVIDIA's website
and grab the needed CUDA stuff (SDK and examples).

See my other thread on CUDA and Current Laptops.

-RG

For what it is worth, on my new, pimped out 27" iMac:

2.93 GHz Intel Core i7
8 GB RAM
Mac OS X Snow Leopard 10.6.5

I am getting a MathematicaMark8 benchmark of 0.98 after a fresh start
up.

Considering the top entry is a 3.07 GHz machine with the same
processor, I appear to be running at an efficiency that is 2.7% higher
than their reference Intel Core i7 machine, (3.07/2.93)*0.98 = 1.027.
I have to say, I was pleased to see this and am taking it as evidence
that Wolfram has worked to level the playing field between platforms.

This being said, as noted earlier, the new 27" iMacs are using the ATI
Radeon 5750 and will not be able to take advantage of CUDA. Just the
same, the bang for the buck for this machine is nice.

I had the same CUDAQ[]=False when I tried to use CUDALink on my month
old macbookpro 17''.

After some research I found my computer has two cards (one is the
intel integrated, the other the NVIDIA). I installed gfxCardStatus and
then I could select it to use only the NVIDIA card, it works fine
now.

I just don't get it why the CUDA examples from NVIDIA don't have a
problem detecting the correct GPU while Mathematica, which uses the
same drivers, has the problem.

Best
Felipe

Hi

On Nov 21, 12:29 am, truxton spangler <truxtonspangle...@gmail.com>
wrote:

> Syd I have a one month old Macbook Pro with an Nvidia GPU but CUDAlink
> doesn't seem to like this particular card (the standard GPU shipping
> with this model mac, GeForce 320M)
>
> In= CUDAQ[]
> Out:= False
>
> So it looks like this must work only on selected nvidia cards???

I had the same problem.  No worry, you can use CUDA but you need to
install the driver that you will find there :
http://www.nvidia.com/object/mac-driver-archive.html

The documentation is not quite clear thereabout.

Could you please try the CUDA tutorial and check if acceleration
occurs as expected ?