It is also going forward with project Denver and Boulder, as far as we know

Nov 2, 2012 07:28 GMT  ·  By

The ARM architecture is really expanding its uses, even without ARM Holdings doing much directly. Calxeda, Advanced Micro Devices and NVIDIA, among others, are doing it instead.

This is the latest piece of proof that ARM only really needed to make a 64-bit capable architecture in order to enter markets beyond that of smartphones / portable consumer electronics devices.

After AMD said it would create ARM CPUs, and with the first 64-Bit ARMv8 cores released, NVIDIA has made a move as well.

To be fair, the Santa Clara, California-based GPU maker already has a couple of development initiatives going. One of them is the Tegra SoC series, while the other is known as project Denver.

Project Denver is the idea that a GPU can be given one or more ARM cores, so that it may run an OS without the need for a CPU from AMD or Intel.

Then there is Project Boulder, which might just turn out to be a genuine ARM-based CPU, possibly bereft of a GPU even.

The new idea that NVIDIA got, though, is for ARM Tesla GPU compute accelerators. In other words, NVIDIA is thinking of including ARM cores in its parallel computing modules for supercomputers.

"Tegra is going to become GPU computing capable in the not-so-distant future. Sometime this decade we are also going to start bringing integrated CPUs and GPUs together in the Tesla line," said Steve Scott, chief technology officer for the Tesla product line at Nvidia, according to InfoWorld.

Now that 64-bit instructions are supported by ARMv8, the better performance of x86 is no longer a strong enough argument against the use of ARM cores, especially with GPUs accounting for most of the processing power anyway.

For those who want proof, 90% of those 20 Petaflops produced by the Titan supercomputer are accounted for by Tesla GPUs. AMD Opteron x86 CPUs provide the rest.

No date has been given for when Tesla cards, or chips with both ARM cores and GPUs, will be available.