Mobile GPU capable of 64bit arithmetic

I am intending to vectorize an FE code using CUDA/OpenCL. I am going to be out of the office whilst doing this development work so it would be very convenient if I could find a suitably capable laptop machine.

I was considering an Alienware M17X with a GTX 280m but I now see that this isn’t actually based on the GTX280 - Is it capable of Double Precision Arithmetic?

Is there a mobile NVIDIA graphics card capable of 64-bit arithmetic using CUDA/OpenCL?

The NVIDIA web page for the GTX 280M links to the DELL site for the Alienware M17X, when I try to configure it it only has the option of ATI 4870 graphics. Would that give DP arithmetic with OpenCL?

There is no NVIDIA mobile part with double precision support. The Mobility Radeon 4870 can, theoretically at least, do double precision (although AMD don’t actually have dedicated double precision hardware). Last time I checked, AMD don’t support the double precision extensions to the OpenCL specification.

I’ll re-try here with asking related question (unanswered in other thread): does anyone have a list of new mobile GPUs that are actually CUDA 1.2 capable?

Yes, any of the G215 G216 G218 parts. You can just look for the 40nm process in the parts list. The fastest currently is the GTS260M.

Documentation for compute capacity is uniformly awful for anything other than Tesla. The following should be Compute 1.2 capable:

Geforce G210M (GT218 based 16 cores)
Geforce GT230M/GT240M (both GT216 based 48 cores)
Geforce GT250M/GT260M (both GT215 based 96 cores)

but take that with a large grain of salt.

GT21x is 1.2, GT200 is 1.3.

Thanks for the replies. What about new Quadro mobile GPUs (for example, FX 880M) - any of these 1.2 capable?

At a guess (and I stress guess), the FX380M, FX880M and FX1800M probably are also compute 1.2 parts. The FX1800M is the one with the biggest question mark over it - the official NVIDIA specs show it as having 72 cores. So perhaps it is a GT215 with three MP fused off, otherwise it isn’t obvious what it is. But all three are listed as DX10.1 compatible, which means they should be GT21x based.

The one mobile part that has got be really interested is the new Ion 2. Anyone know anything about it? I am guessing GT218 based…

At first glance, Ion 2 + pinetrail could be a killer mobile cuda system. Driving the display using the pinetrail IGP with CUDA running on the ION 2 could be a great way to do development and demos on the go. Acer have already announced a pretty tasty looking 10" netbook using Ion 2 + pinetrail that could be a perfectly serviceable CUDA rig weighing less than 1.25kg.