New mobile cards performance


Does anyone know where i can find some performance estimates for some of these new mobile cuda capable cards:

I noticed the bandwidth is a bit lower, what about the shader clock? It would be interesting to see the performance/watt ratio.

One question on my mind, how is the host <----> device latency on these cards? :confused:

If using pinned memory the mobile GPU will have direct access to the RAM so the host to device transfer is basically gone ?


Since these mobile GPU cards are connected by PCI Express, they have no direct access to the (PC)RAM. You have

to copy the data to the GPU RAM, like on non mobile cards.

Really? I read this in the BPG

“Zero copy is a new feature as of version 2.2 of the CUDA Toolkit. It enables GPU threads to directly access host memory. For this purpose, it requires mapped pinned (nonpageable) memory. On integrated GPUs, mapped pinned memory is always a performance gain because it avoids superfluous copies as integrated GPU and CPU memory are physically the same” -P. 28

Or have i misunderstood something here?

Only compute 1.3 devices and ION/9300M/9400M support zero copy (and even then the PCI express bus is still involved). These mobile parts are neither.

ok, really interesting. Despite it still involving the pci, do you know of and/or have any reference to what sort of bandwidth one might expect?

GT21x chips support zero-copy (1.2 devices are the bulk of the GT200 changes, 1.3 just adds double precision). They are also PCIe based, even though they’re in a laptop.

(you can max out whatever PCIe connection you have with zero-copy no problem)

so that includes the GT230 and GT240 ?

Anyways I’ve contacted some Nvidia ppl in the UK for more info…


well there are hundreds of parts, each which are essential. But some of the major parts include, battery, speaker, microphone, transceiver, antenna, flash memory, RAM, an LCD display, and the keypad.

Wireless Spy Camera