Any Excel 2010 -> Windows H/PC 2008 R2 users/triallers?

[vbuggy]

Um... I wasn't quite sure where to put this, so I've stuck it here. Apologies if it's not a right fit.

I do a lot of crunching in Excel, and I'm seeing a lot of stuff about how Excel 2010 integrates with Windows H/PC in order to accelerate calculations in order of magnitudes.

I do a lot of financial and energy related work as well as a few other things - but the energy use alone would actually help hugely. I have branched into GPGPU based computing and I have done some useful stuff in life sciences and engineering with Tesla arrays I already have, but these have been ground-up developments - and above all, it's just a fact that I have a lot of really valuable, total-sod-to-replicate-elsewhere, 'legacy crunching' locked up in Excel.

I can certainly make room in my tech budget next year for a couple of adequately specced Octane III / CX1-class installations for example, and I was wondering whether it was all it was cracked up to be. I don't want to look like an idiot when talking to a Cray / SGI / etc rep and am still researching exactly what this all entails, but I also wanted to be forearmed with a few anecdotes from actual users.

It's a very long shot here, but I'm guessing that some of you work in maintaining or implementing larger-scale computing systems and I was wondering if there was anyone here who could shed some light on WinHP/C 2008 R2 Server + Windows 7 workstation + Excel 2010 scenario - in terms of things such as the implementation process, what changes need to be made in the workbooks, and common pitfalls. Thanks

 

[PsiStar]

Excel & HPC are an oxymoron.

 

[Mark R]

Not tried it, and not something I'm interested in setting up - but I heard about it a while ago. AFAIK, it only accelerates user defined functions (UDFs) which have been compiled with an external compiler.

If you recompile the UDFs with cluster support, and install them on your HPC cluster, then it's just a matter of setting up HPC off-loading in excel, and pointing it at your cluster.

 

[vbuggy]

That alone is fairly attractive, but what I'd been led to believe is that the acceleration is now more widely implemented. Is that BS?

 

[libertysyclone]

I would ask on the nvidia CUDA forums there are a lot of people having good success with the Monte Carlo simulations within excel there.

I work on the hardware end of what you are looking for so, if you have any questions for that I would be happy to help. Most of these HPC resellers have software teams that would answer your "sill question" happily.

 

[PsiStar]

The cuda stuff makes sense although that would likely be an addin written in some language & compiled ... which would apply to almost any addin and it sounded like that was what you were hoping to avoid, "... ground-up developments".

It sounds to me like you should investigate Matlab or MathCAD. They can be made to do everything with significant effort on your part OR because they are OLE applications as is Excel, you can use Excel directly within those programs (which are better described as environments).

Using Excel inside those programs gives you all there is about Excel as well as significantly improved crunch power not to mention graphics (most like Matlab for the graphics). I am pretty sure that Matlab can use GPGPUs but may require some expensive option and the same for MathCAD. Personally I think MathCAD maybe the easier route to investigate ... if it has GPGPU access capability.

 

[libertysyclone]

The cuda stuff makes sense although that would likely be an addin written in some language & compiled ... which would apply to almost any addin and it sounded like that was what you were hoping to avoid, "... ground-up developments".

It sounds to me like you should investigate Matlab or MathCAD. They can be made to do everything with significant effort on your part OR because they are OLE applications as is Excel, you can use Excel directly within those programs (which are better described as environments).

Using Excel inside those programs gives you all there is about Excel as well as significantly improved crunch power not to mention graphics (most like Matlab for the graphics). I am pretty sure that Matlab can use GPGPUs but may require some expensive option and the same for MathCAD. Personally I think MathCAD maybe the easier route to investigate ... if it has GPGPU access capability.

Matlab at this point hasnt released a GPU client, but accelereyes has and it doesnt tax you on the amount of GPU (15 day trial too IIRC)

http://www.accelereyes.com/

 

[vbuggy]

Matlab is a separate point - and yes, I am looking at developments there with interest as well. It was really the question of how easy it was to literally give Excel a kick up the backside using the H/PC integration without the development work that we've had to put in on other solutions.

Anyway, I'm going to be in conversation with sales teams so yes, I guess I'll act dumb. I just wanted to know the basics before the hard sell started, oh well.