Re: Mathematica 8.0.4 now available
- To: mathgroup at smc.vnet.net
- Subject: [mg122646] Re: Mathematica 8.0.4 now available
- From: Bert RAM Aerts <bert.ram.aerts at gmail.com>
- Date: Fri, 4 Nov 2011 06:01:52 -0500 (EST)
- Delivered-to: l-mathgroup@mail-archive0.wolfram.com
- References: <j89uqj$60$1@smc.vnet.net> <j8tkif$evr$1@smc.vnet.net>
On 3 nov, 09:49, DmitryG <einsch... at gmail.com> wrote:
> I have tested Mathematica 8.0.4 for CUDA because now I have a new
> laptop with a new nVidia Geforce 555M card. As before, CUDADot is
> slower than regular Dot on reals, so I do not know how I could benefit
> from GPU calculations
>
On my 4 years old notebook with Core 2 Duo T7300 and nVIDIA GeForce
8600M GT I get the following difference:
Timing[CUDADot[Table[i, {i, 500}, {j, 500}], Table[i, {i, 500}, {j,
500}]]] gives 0.031995
Timing[Dot[Table[i, {i, 500}, {j, 500}], Table[i, {i, 500}, {j,
500}]]] gives 0.218967
So CUDADot is faster than Dot. My graphics card only supports integers
and single precision floats.
I use Mageia 1 linux with latest nVIDIA driver 285.05.09 and CUDA
toolkit 4.0.17, Mathematica 8.0.4 with CUDAlink 8.0.4.1
Concerning compilers: I use GCC and needed to adapt a header file to
get my version supported, see
https://groups.google.com/group/comp.soft-sys.math.mathematica/browse_thread/thread/b2796ec532fa3b12/b916d81a4ec2e636?hl=nl&lnk=gst&q=cuda+bert+aerts#b916d81a4ec2e636