Last November, NVIDIA announced CUDA on ARM during the unveiling of the Barcelona Supercomputing Center, which is powered by combination of ARM based Tegra 3 SOC and the CUDA GPUs (prelude of Project Denver). Today, the company launched a developer kit based on this combination named CARMA.
In 2013, NVIDIA plans to launch a desktop, notebook and more importantly, a server chip based on the Project Denver, it’s first custom CPU based on the in-house 64-bit ARM-compliant architecture.
In order to allow developers to build the applications for the new chip, which will combine multiple ARM cores (allegedly) with 256 CUDA cores – NVIDIA has rolled out a development kit which combines the NVIDIA Tegra 3 chip, featuring five ARM Cortex-A9 CPU cores with a Quadro 1000M GPU. The kit is manufactured by SECO, and is consisted out of Quadmo747-X/T30 daughterboard, NVIDIA Quadro 1000M MXM board (very similar to one used by HP in their revolutionary upgradable workstation), and the motherboard, also built by SECO.
The specifications are as follows:
• NVIDIA Tegra 3 Quad-Core ARM Cortex-A9 processor with 2GB SOC Memory
• NVIDIA Quadro 1000M GPU with 96 Cores and 2GB GPU Memory
• 270 Single Precision GFLOPS, 135 Double Precision GFLOPS
• 4x PCIe Gen1 link connecting processor to GPU
• Gigabit Ethernet Network Port
• Direct attach storage support with SATA port
• HDMI and DisplayPort display connectivity
The board was in development for quite some time and the company decided to do a contest in order to get the name for the part. Daniel Holt came up with the name CARMA (CUDA meets ARM) and won one of the first parts that will go off the line.
The price for the development kit isn’t known at current point in time, but you can expect to be a typical low volume part, with the price in upper XXX to a very low XXXX figure (we’d wager between $800 and $1200).