99$ super computer, with the size and power consumption of a RPi
by Lukas Joergensen · in Torque 3D Professional · 12/19/2012 (8:30 am) · 15 replies
Yes you heard me, this little fella, is called Parallella and gives a whopping 90 GFlops while only taking 5W of power and costing the small sum of 99$..
What I am instantly thinking is that porting T3D dedicated server to linux was probably one of the best moves made by the community considering this little piece of tech.
You could literally have your whole server cluster in your home!
I can't stop getting excited over this! This is awesome!
Only downside, T3D is not the master of threading tbh, and since every core in a parallella runs at 700 MHz, it is not a simple plug in and go solution.
So a question to you guys that I am VERY interested in, do you think T3D would be able to take enough advantage of this PSC (Personal Super Computer) or do you think it is simply too much work?
The benefits of getting T3D running on one of those guys is unimaginable IMO.
Now do as I say and discuss!
Edit: Ah my bad, the 90 GFlops (or 45 GHz) would be on a 64 core board ( a stretchgoal but I think they said it would be released in around 2013 ) the 16 core, the one to 99$ only got about 1/4 of that. But the topic still stands! It would still be an incredibly cheap and power efficient server!
What I am instantly thinking is that porting T3D dedicated server to linux was probably one of the best moves made by the community considering this little piece of tech.
You could literally have your whole server cluster in your home!
I can't stop getting excited over this! This is awesome!
Only downside, T3D is not the master of threading tbh, and since every core in a parallella runs at 700 MHz, it is not a simple plug in and go solution.
So a question to you guys that I am VERY interested in, do you think T3D would be able to take enough advantage of this PSC (Personal Super Computer) or do you think it is simply too much work?
The benefits of getting T3D running on one of those guys is unimaginable IMO.
Now do as I say and discuss!
Edit: Ah my bad, the 90 GFlops (or 45 GHz) would be on a 64 core board ( a stretchgoal but I think they said it would be released in around 2013 ) the 16 core, the one to 99$ only got about 1/4 of that. But the topic still stands! It would still be an incredibly cheap and power efficient server!
About the author
IPS Bundle available at: https://www.winterleafentertainment.com/Products/IPS.aspx
#2
12/19/2012 (9:02 am)
Thats not nearly the same thing.. 
#3
12/19/2012 (9:16 am)
Wow, looks like it has awesome potential. I wish I had known about the kickstarter project earlier. Maybe I'd have one sitting on my desk already!
#4
It would be theoretically be the cheapest game server ever, if you manage to run a dedicated T3D server on it!
12/19/2012 (9:18 am)
@Scot I doubt it! They are releasing the first ones at 31st december, but yeah as soon as they begin selling then I will be jumping onto one!It would be theoretically be the cheapest game server ever, if you manage to run a dedicated T3D server on it!
#5
Fast forward to the present, I read your post about this 64 Core CPU and probably runs on the Minix 3 OS or possibly other small Linux release.
It is quite impressive for Power vs Size for sure. It's going to be the size of a credit card.
12/19/2012 (9:31 am)
I remember an article about Intel that had designed a 48 Core CPU several years ago. The problem that the article mentioned was that there are no Operating Systems designed or built to use it.Fast forward to the present, I read your post about this 64 Core CPU and probably runs on the Minix 3 OS or possibly other small Linux release.
It is quite impressive for Power vs Size for sure. It's going to be the size of a credit card.
#6
12/19/2012 (9:33 am)
@Scott it runs ubuntu or command-line linux (the one of your choice), and that is ootb 
#7
12/19/2012 (9:45 am)
Sounds cool for a server or desktop. The Intel 4000 GPU should run T3D reasonably well.
#8
The marketer pointed out something along the lines of saying 'this is why we make need 750k in 30 days to cover the production costs so we can make the part available to the masses and see what they can develop with it, as opposed to a few pieces at 10k each'. It's essentially like throwing bread crumbs to the ducks in a park and see how many flock in to eat.
Still think it's a bit early to be as excited about the news as I was just a moment ago. I couldn't glimpse the exact OS they used from reading or listening to the video but in my defense, if it is in there, I must have missed the details since I get interrupted with my kids every few minutes. Could you help me locate that Information about the OS that you found?
12/19/2012 (10:26 am)
I just finished watching all the videos and had a concern that the unit is basing the processor speed on Floating point processes with just a few Integer processes. I'm not sure exactly how this effects the usefulness but some of the Britains in the meeting room obviously had concerns.The marketer pointed out something along the lines of saying 'this is why we make need 750k in 30 days to cover the production costs so we can make the part available to the masses and see what they can develop with it, as opposed to a few pieces at 10k each'. It's essentially like throwing bread crumbs to the ducks in a park and see how many flock in to eat.
Still think it's a bit early to be as excited about the news as I was just a moment ago. I couldn't glimpse the exact OS they used from reading or listening to the video but in my defense, if it is in there, I must have missed the details since I get interrupted with my kids every few minutes. Could you help me locate that Information about the OS that you found?
#9
Zynq-7010 Dual-core ARM A9 CPU Epiphany Multicore Accelerator (16 or 64 cores) 1GB RAM MicroSD Card USB 2.0 (two) Two general purpose expansion connectors Ethernet 10/100/1000 HDMI connection Ships with Ubuntu OS Ships with free open source Epiphany development tools that include C compiler, multicore debugger, Eclipse IDE, OpenCL SDK/compiler, and run time libraries. Dimensions are 3.4'' x 2.1''
I expect a bit more ram from the 64 bit version (1 gig is a bit hard to get by with)
Of course, doing floating point calculations only (like a normal GPU if you have ever written shaders or other kinds of GPU Computing) will have its issues, but it seems like floating point calculations is the standard in coprocessors so it doesn't seem like a major obstacle.
12/19/2012 (10:52 am)
@Scott under: Parallella Computer SpecificationsI expect a bit more ram from the 64 bit version (1 gig is a bit hard to get by with)
Of course, doing floating point calculations only (like a normal GPU if you have ever written shaders or other kinds of GPU Computing) will have its issues, but it seems like floating point calculations is the standard in coprocessors so it doesn't seem like a major obstacle.
#10
The video mentions that each chip uses 32k memory but they all share a bus of 4Gb for processing. I think he mentions something like 1k for accessible memory for programs. To be sure I'd have to watch the videos over again. I am likely wrong.. interruptions at home while I read.
Try looking into that other research for GPU processing of programs. Sorry I don't have a link to it but it might offer some insight.
Thank you for pointing out the OS for me.
Exciting stuff this is. Hope you plan to blog your milestones. This is intriguing.
12/19/2012 (11:33 am)
There is research to swap processing to a GPU is already out in Google land. I didn't read much of it, but the info is there.The video mentions that each chip uses 32k memory but they all share a bus of 4Gb for processing. I think he mentions something like 1k for accessible memory for programs. To be sure I'd have to watch the videos over again. I am likely wrong.. interruptions at home while I read.
Try looking into that other research for GPU processing of programs. Sorry I don't have a link to it but it might offer some insight.
Thank you for pointing out the OS for me.
Exciting stuff this is. Hope you plan to blog your milestones. This is intriguing.
#11
That would really be a shame. But either way you could always use a Parallella instead of a standard webserver (sure as hell would do it better than my Raspberry Pi!)
12/19/2012 (11:38 am)
One thing that is critical for T3D tho is that it doesn't do a whole lot of threading afaik, so it might need a major rewrite to run efficiently on a Parallella :/That would really be a shame. But either way you could always use a Parallella instead of a standard webserver (sure as hell would do it better than my Raspberry Pi!)
#12
12/19/2012 (12:08 pm)
if a core is 700mhz like the Pi, then most likely it will run similar to running on a Pi without updating torque to take advantage of threading more. Have you tried getting it running on the Pi?
#13
12/19/2012 (1:19 pm)
@Tim never did that because I figured it wouldn't be powerful enough anyway.. Can try and see if I will be able to do it, will just have to install Ubuntu on my pc to compile for linux then (i suggest you can't do it on windows)
#14
I've done a good bit of work with the dedicated server, one area of problem you will likely run into is assembly. If there is not code already to disable it and use C backup code for arm then you may have to do that (or write arm assembly replacements :) )
I would also recommend getting my pull request from github that removes SDL from the dedicated server just to give you one less dependency.
The dedicated server uses a lot less resources than client obviously, so its possible it may run on the Pi, you may not be able to play with a lot of people though.
12/20/2012 (8:06 am)
You should be able to compile it on the Rasp Pi itself. You'll have to cross compile it on linux (which you could do on windows if you get toolchain setup) since the Pi is arm based.I've done a good bit of work with the dedicated server, one area of problem you will likely run into is assembly. If there is not code already to disable it and use C backup code for arm then you may have to do that (or write arm assembly replacements :) )
I would also recommend getting my pull request from github that removes SDL from the dedicated server just to give you one less dependency.
The dedicated server uses a lot less resources than client obviously, so its possible it may run on the Pi, you may not be able to play with a lot of people though.
#15
Unfortunately a lot of people will be disappointed with this thing when they get it and try to do all the usual stuff and get awful performance out of it.. I applaud him for trying to push the issue of parallel programming, but as it stands most of the basic level computer tech goes underutilized in every day applications. Hopefully this doesn't end up being a flop when released, although I have a feeling there will inevitably be a lot of disappointment.
12/20/2012 (10:36 am)
Hate to say it, but I have my own reservations on the whole thing, mostly because it's just nothing really revolutionary and at the heart of it, kind of obvious. Parallel processing is indeed faster but there are reasons we don't go that route currently.Unfortunately a lot of people will be disappointed with this thing when they get it and try to do all the usual stuff and get awful performance out of it.. I applaud him for trying to push the issue of parallel programming, but as it stands most of the basic level computer tech goes underutilized in every day applications. Hopefully this doesn't end up being a flop when released, although I have a feeling there will inevitably be a lot of disappointment.
Pedro Pereira
Default Studio Name