New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Parallella boards as candidate for cheap CPU-intensive hosting offers
So, since Saturday I have my Parallella 2 board arrived. It was a long way, since the moment I became Adapteva's backer at Kickstarter.
It's indeed computer the size of a credit card, significantly more powerful than Pi. The one I hold in my hand has 16 CPU cores in its Epiphany accelerator. For $119, it's quite a deal. I plan to put it to tests/benchmarking shortly and will update those interested with the results.
I wonder, does anyone have plans on providing Parallella-based dedicated hosting, or CPU power rent as a service?
Comments
arm based? i been taking a look at that ECS LIVA that's an amazing bit of tech.
I remember seeing this. Benchmark please
ARM, yes. Their site contains details specs.
Yes, I'll do it as soon as possible.
So this turned out not to be a scam, well good for you.
While the ARM core is still more powerful than a Pi on its own, the way you put this is kind of misleading. It's not "CPU power" if you can't use it, and to take advantage of the additional 16 accelerator cores you need to write software specifically targeting this board. Has any such software been written yet? If so, what can it do? Aside from maybe special versions of *coin miners, few generally-usable applications of this parallel computing chip come to mind. It's not like this will accelerate your web server or database or even a game server.
Can someone please post the full specs?
Parallella-16 Micro-Server Specifications:
http://shop.adapteva.com/collections/featured-products/products/parallella-16-micro-server
Exactly where it's misleading?
CPU power is exactly what I meant - those Epiphany additional cores. This is what this all ado was about. However, in their Kickstarter details they were planning to put more cores. Well, 16 will do for a start, and if I can really put them all to work with decent speed, I'll look forward to more powerful Parallellas.
I don't know current state of tools/libraries supporting those cores (this is what is required if Parallella should be put for real use for Web servers/whatever else). Since all its software is community-supported, there is obvious place to start looking at.
I have posted link to its site, to have a look at all available info (in order not to post everything here).
It's not "CPU power" if you can't use it. And the Parallella coprocessor is not a CPU to begin with ("C" means "central", remember?). You keep repeating "CPU power", "16 cores", one might think it's just the same as doing a ''cat /proc/cpuinfo'' and just seeing all 16 there, being able to just run your regular apps on those, and it's all simple as that. But in reality this is more like a dual-core CPU with a specialized coprocessor attached. You have to write programs specifically for this board. And you can't use the 16 cores for pretty much anything at the moment, and even in the future, likely only for specialized tasks such as scientific computing. For everything else, the board is just a dual-core ARM, and if what you're doing you can only use ARM cores, there are much better and cheaper options, for example the ODROID.
You are entitled to opinion NOT to count those as CPU cores. I do not attempt to forbid you having this viewpoint.
After all, I haven't yet tried to build/install/run Linux distribution customized for Parallella. When I do, I will see how those cores are available and what is required to use them.
According to Parallella developers, goal is to allow using those cores as easily as possble, up to running regular applications built without specific Epiphany support.
I honestly believe there's better boards out there (better specs but don't know about stability). I had a list saved somewhere with some boards that I was looking at. There was one with quad and 2GB RAM.
I own a Pi as well - good device, but not extremely fast; however,very useful for many a task. I considered trying other devices as well, but at that moment I heard about Parallella.
I backed Parallella because I work in area requiring high computing power. The idea of having high computing power at moderate cost is very appealing.
Please wait for a while, until I have tests at hand. I intend to use the device for the mentioned purposes (TWIMC, it's cryptography and several mathematical problems requiring computing powers).
I am very interested to see your conclusion. I have several Pi boards too but they are really slow plus they just freeze sometimes (which makes them not very reliable). I am trying to find that list I had before but I'm afraid I've lost it with the tons of junk I keep in some of my folders (too many ideas not that much time).
Not cost effective in a DC to colo. Power costs + physical space costs + IP costs + bandwidth (+peering) + cooling costs + other various costs = little or next to no profitability for a colo provider.
It just isn't powerful enough yet. In saying that; if the it had more ram, more flash space & a cpu 5x better than what it currently is - it would be cost effective to colo for low-end colo plans.
I'd say give them 5 years or so to develop this further and it'll probably be cost effective and we should be seeing these popup.
Common problems with some of these micro computers are overheating and stability. As previously state, I've seen a lot of them freezing randomly. Although EDIS and other providers have Ras. Pi colo.
Both issues are common indeed. However, I'll test now, and I do not mind if they build something stabler and more powerful in a few years.
Does it actually need a fan to work optimally? Are you running it fanless?
It is strongly recommended on site to use both heat sink and fan, yes. Not too convenient, but currently it's the only option.
I think this will be less of a problem when it becomes more mainline for ARM servers. the biggest issue is not all software projects run on arm.
I like this!
Sorry for doing a necro. But it seems this thread was not updated with benchmarks. How is performance @Master_Bo?
No fresh data so far I'll post it when it's available.