Parallella "supercomputer" :D
this looks very interesting
anyone ordered one of these yet?
should we order one?

Wow i’m impressed. All we need now is an ARM build of cycles :rolleyes: Do you think one would be able to run a home web/fileserver on one of these thanks to the highly parallelized coprocessor?

Buy one million of them at least

There’s little parallelism in web/fileserving. In fact, I can’t really think of any good use case for such a device. For graphics, it has too little RAM and I doubt it would even outperform CPUs or GPUs in any real-world scenario.
If there was a problem that this device solved any better, they wouldn’t have to resort to kickstarter to finance it. Seems more like a tinkertoy to me.

For those visual people:

Nice thing, someting like a beefed up Rasberry Pi, there 66core version can (once compleated) do 90 gigaflops with a power usage of 5w. You could power these with a Solar Plate :smiley:
The problem is that they only have 1gb of ram (well still a lot more then the Pi)

Here are the Specs:

Sure, but then again rendering on the GPU can also be considered a tinkertoy type of thing. Massively parallel computation on the GPU is still a side-benefit from a device that was designed for games. I’m still waiting for AMD to fix their shitty closed source drivers in order to be able to do anything with it. So being open-source, the Parallella has my full support. The strength of this little board is that they can be stacked up into a cluster. Even though a single unit with 1GB of RAM is not much, they can scale up for some pretty impressive performance per watt.

Weak and pointless.

Modern GPUs are designed for both “serious” computation as well as games, that’s why they are in most actual supercomputers built these days. Their parallel processors are general-purpose just like the parallela, the differences in design are probably not that significant.

The strength of this little board is that they can be stacked up into a cluster. Even though a single unit with 1GB of RAM is not much, they can scale up for some pretty impressive performance per watt.

The 1GB of RAM is the reason it doesn’t scale for a large amount of problems. So far, they have failed to demonstrate a viable use-case for these devices, yet they are marketing it as a “supercomputer for the home”, which is kind of ridiculous. If the Parallela is a supercomputer, then so is pretty much every modern GPU.

Lets compare this to the nVIdia Titan.

Parallella: 1GB Memory, 66 cores, 90GFlops with 5W power draw.
nVidia Titan: 6GB Memory, 2668 Cores, 4500 GFlops and ~250W power draw.

So the Titan is 50x more powerful, and uses 50x the power. The Titan is also probably cheaper than 50 Parallella, and a LOT smaller. The Titan is also probably easier to program for.

While I see it as an experiment to introduce people to parallel computing, learning to program via CUDA, or even good networked x86 applications would be more beneficial than this, while still teaching as well.

your just sad that you have amd cards lol

i thought geeks used blender not a bunch of cheap turds :confused:

I love the idea.

Details are what worries me.

What does that have to do with anything? I’m quite happy with my 7970’s. I use them for [email protected] and playing games most of the time.

I don’t use Cycles, I’ve used LuxRender since v0.7 and continue to do so today, and sometimes with my GPU’s.
They also cost me $420 each, I’m not complaining.

Given my example was SUPPORTING the Titan, your possibly a nVidia fanboy [removed unnecessary insulting text]

hey calm your tits :stuck_out_tongue:
wasn’t trying to be mean
and the cheap thing was about spending 100 on a nice little toy which any geek would do imho
100 dollars is spare change if your buying something cool :smiley:

If 6 of them use 30 Watts and add up to 6 gigs of.ram…
why cant I add another every time I have a hundred laying.around?

You can’t do that with a normal gpu?

The problem with that is that it doesn’t actually equate to a total of 6 gig. Just like when you have two gpu’s working they are only using as much ram in the scene as you have per card.

So, if I were running a game with this, the total geometry from.a scene.could not exceed 1 gig? Even if you had 10 gigs of chips?

It’s the same right now with GPU’s. Right now, there’s no technology that allows different fragments of data to reside in the VRAM of different chips (and then share that via. SLI or another setup type).

Nvidia’s Maxwell architecture however will pave the way for the ability of the GPU to share memory with the CPU, which the Parrelella team will have to do if they don’t want to have a potentially crippling memory limit.

This is quite neat - as problems arise, solutions are sure too follow. Amazing how far we have travelled since the humble first gears of the first computer were rotated.

I think the keyword here is a combination of open source hardware + parallel computing.

Its the same reason why Rasp Pi is popular despite there is a lot of other closed hardware by companies such as Parallax and all.

It makes a very good parallel computing study if I read correctly.

Anyway, it was enough for even Ericsson to invest in them.