Can anyone confirm the level of CUDA computation in the new 480M mobile chip?

It is available in the UK from Kobalt in a 17in machine.

What I really want to know is whether this chip will finally provide double precision CUDA capability on a laptop - as far as I know no previous chip does.

It would also be very helpful if Nvidia would kindly put this type of information on their web spec areas as a matter of routine.

Thanks. But how many Double Precision processing units are there? My understanding is that a desktop GTX480 has 60 DP units but a C2050 has 224. What is number on 480M?

This is a pedantic language nitpick, but given how double precision seems to be implemented on Fermi, it no longer makes sense to talk about “DP units.” Instead, it looks like the double precision implementation joins two single precision units to do the calculation somehow, rather than using a dedicated double precision FPU like on GT200. It’s probably more accurate to talk about DP speed relative to single precision to ignore the implementation subtleties. So, GTX 480 runs double precision at 1/8 speed, and C2050 runs at 1/2 speed. It’s the same chip, just artificially capped in the GTX 480 case.

Anyway, yes, I have the same question. 1/8 for 480M? :)

Many thanks to you and seibert for the clarifications on this.

Sounds like the 480M is rather a breakthrough if we can finally replicate full scientific computation on a laptop. I will have to find the cash for one!

The lowering of the DP functionality on the GTX 480 and 480M cf the Tesla 20xx is a wee bit frustrating, but after going to a meeting where it was quite clear most of the academic staff were reporting results on GTX2xx series rather than C10xx series it’s not hard to understand Nvidia’s point of view. I understand the C20 series has better error correction as well.

If anyone gets a 480M laptop please could they post the results of e.g.

Yeah. But they should have put out a 1.2GB version of C2050 without ECC that sells for $700. I am sure many people in academia would buy. At $2,499 for 3GB and ECC, that’s too outrageous…

Even Intel is more reasonable, X5680 and 980X is less than 2x price difference.

I agree with you - I was only saying I could understand Nvidia’s plan! The company was much more reasonable during the Mad Science promotion and the final release prices are such that many of us will still buy rather more 480-class than take the leap to the Tesla. I would certainly find 700-1000 for a “C2030” over a 480 GTX, but have no plans to buy a 2050.

But back to the mobile chip. It is early days but in the UK kobalt have options for several GPUs on one of their machines. In GBP, including 175% VAT the options are