CUDA on new MacBooks Anyone use CUDA on GT 330M chip?

Word is that the new MacBooks have GT320M and GT330M chips driving them. Anyone used these for CUDA? How’s the performance? Any issues? All the info I can get from nVidia’s web site says that this can happen, and is supported.

Regards,
Martin

That’s a good question. I’m also very interested in this topic. It would be nice if anyone

can post some information

That’s a good question. I’m also very interested in this topic. It would be nice if anyone

can post some information

what can a 48cuda core can do… it even lower then 9600gt…

48 CUDA cores at 1.2 GHz is still not bad, especially compared to a laptop CPU. I would rather be transcoding a DVD to H.264 with a GT330M than a mobile Core 2 Duo.

Plus the 330M is a compute 1.2 part… you get doubled registers, much better coalescing, zero-copy memory, and shared atomics.
And finally it uses about half the watts/flop than older mobile parts.

I’d like one!

Edit: followup. I am behind the times. Maybe I’d like one of these Asus “gamer” laptops more.
These have the GTS360M, more than double the GPU power over the Apple macbook Pro. And at $1400, it’s actually not a bad deal… 2/3 the price and double the GPU power.

SPWorley,

Where do you find this info? Not that I don’t believe you, but the only info I have on Compute Capability is from the OpenCL programming guide, appendix A, and it doesn’t list the 330M.

It would be a good machine for demoing CUDA code before running it on a bigger card. I use Macs for most of my regular stuff, and linux boxes for computation. Best of both worlds, but it costs more sometimes.

Regards,

Martin

This is a useful list for released parts. (Don’t trust it for unreleased parts though!)

The 330M is a 40nm GT206 die. This is the exact same chip as the GT220 board, which is definitely compute 1.2.

There’s a bunch of us here on the forums (myself included) who are fond of the single-slot GT240… they are single slot, very cool and very low wattage, yet pack a punch.

They’re nothing special for graphics (Anand called it “the card that doesn’t matter.”) but are great for GPGPU.

Verification is at NVIDIA’s page on the 330m. This claims DX10.1 support, which is only supported by the G200 (and GF100) chips… ie, compute 1.2, 1.3, and 2.0. If you look at say the G92b based GTX260M, it only supports DX10.0, and it’s compute 1.1.

It’s Compute 1.2.

Thanks. I suppose your being a nVidia employee counts as confirmation. (Not to discredit SPWorley work, I believe him too. I just want to know where I can find out about other options.)

Now, where can I find this info for myself, all in one neat table. It would be great if it were at:

http://www.nvidia.com/object/graphics_cards_buy_now.html

It would also be great if older cards were on that table.

Regards,

Martin

Hi,

I made some benchmarking on my 17" 2.53 GHz MBP with CUDA N-body. The result ws quite disappointing, about 60 Gigaflop/s. It is far worse result what 330 GT is cpeable of, 182 Gflops, as NVIDIA claims.

Could it be a driver problem?

Regards.

Screen_shot_2010_05_09_at_8.53.09.png

Did you take a look at the memory bandwidth associated with the GeForce 330M ? :lol:

Memory is expensive, especially for Apple, that prefer to put 28GB/s bandwidth on a GPU that need at least 56GB/s bandwidth to express himself…

PS: compare that to your CPU, and you will find that the GT 330M is largely faster for n-body simulation, and that even integrated solutions, GT 320M or 9400M are usually real faster, and could be compared to quad-core desktop CPU!
Demoing the performance-level of a $2000 iMac Core i5 desktop on a $1000 laptop (white MacBook) is not so bad!!!