Ultimate Edition Forum

The Jetson TK1 is a Compute(GPGPU) solution from Nvidia with a sub 10 Watt power usage.
The compute power is 300 GFLOPS. This kind of compute power is more than the Xbox 360 and ps3. In technical terms the Jetson TK1 is a supercomputer which uses Ubuntu 14.04(with nvidia modules for the TK1) out of the box. It uses an NVIDIA kepler chipset with 192 CUDA cores. In other words flagship and enthusiast parts of less than 8 years ago gets dwarfed in performance by this SOC, all possible by linux and the open nature of its kernel.

Agreed.
But $192 is still unaffordable for most. The price trend should be less than $100. The sitting shelf price would be $150 on mass consumption.

As far as FLOPS is concerned, there is no way to test it on a home computer. But the phoronix test suite is still the best bench out there for linux. http://www.phoronix-test-suite.com/ As to what FLOPS is; its a performance benchmark of pooled floating point performance or in supercomputers; using multiprocessor multicore CPU's over a network that is just as fast. There is always overhead involved in consolidating computer resources. Theoretically you can still calculate it...

As per wikipedia:
One can calculate FLOPS using this equation:[1]

{FLOPS} = {cores} x {clock} x {FLOPs} / {cycle}

Most microprocessors today can do 4 FLOPs per clock cycle.[1] Therefore, a single-core 2.5 GHz processor has a theoretical performance of 10 billion FLOPS = 10 GFLOPS.

On my six core CPU, that would be 108 GFLOPS(only in theory).

A real world scenario would be; an array of CPU's and memory put together, connected to each other using infiniband. The hardware would be extremely powerful and scalable in performance. Couple this with protocol refinements in the network and better algorithms wherever needed, massive overheads can be reduced. Its no longer got to do Anything with CPU's and RAM, its the network that becomes the computer. FLOPS is a metric of complex computation and the higher it goes, on the same hardware and only because of refinements in software and the networks, determines the power and scalability of the array. That is needless to say that older CPU's will be dumped for new CPU's, older arrays will be dumped for newer workstation cards(Tesla/Quadro/Firepro/Xeon-phi).

The Radeon 295x2 has 11.2 TFLOPS.

Im no fan of Nvidia. Im an AMD guy. Dont get me wrong, Im not a hypocrite.

Nvidia for now is the most straight faced company around and does more than what they say.
AMD's OpenCL and HSA(using SP-stream processors) when it 'materializes' will beat CUDA into the dirt....and then NVIDIA will do something you cant imagine.....healthy cOmPeTiTiOn is Good.

Strange thing, for me now its not an option to use Radeon parts. I use Blender. It means I have both CPU and GPU at my disposal. With a Radeon I'd be in depressed state. Blender lusts for CUDA cores and additional CUDA cards. INX from Oz Unity forums is a Blender pro with years of experience. He'd prolly need 3x GTX 780Ti for the stuff he's capable of rigging. A processor too can do a job in case of some disaster but today its only ideal as a backup plan.

The TK1 has a 4x ARM A15 cores for its CPU part. That isn't much because CPU's are being superseded in real tasks by GPU's at very low power with even lesser heat. If I ran blender on a CPU for 4 hours rendering something, the electric bill will be too high and I would have stressed the system out, capacitor's doing overtime, with the heatsink untouchable and a scary situation to be in(Hardware is no longer cheap anymore) It'd be impossible for me to replace.

Imagine having low cost, low powered, high performance multi gpu's in a small form factor....AMD'S 7850K and Nvidia's Jetson TK1 are just the stepping stones...