Reply With QuoteVery interesting indeed!
I'm definitely looking forward to more ARM benchmarks!
Senior Member
Except that Michael's one clearly states "NORPRO" right on the cover...Originally Posted by oliver
Senior Member
Very interesting indeed!
I'm definitely looking forward to more ARM benchmarks!
Senior Member
how are you going to connect all those together?
Senior Member
"This six-board cluster will be running an Ubuntu 12.10 ARM OMAP4 snapshot for the noticeable performance improvements it offers via the Linux 3.4 kernel and GCC 4.7. "Michael, you really should checkout the Linaro GCC backlog now and again and use their latest version directly when you test as there's some interesting options in the backlog blueprint
https://launchpad.net/gcc-linaro/+milestone/backlog
34 blueprints and 0 bugs targeted
https://launchpad.net/gcc-linaro
Linaro GCC 4.7-2012.06
Senior Member
probably through the slow On board 10/100 Ethernet ports but he may get better thoughput with the
integrated 802.11 b/g/n chip , i don't know if it can do better real throughput than the 100Mbit Ethernet though
http://pandaboard.org/content/pandaboard-es
Last edited by popper; 06-07-2012 at 10:16 PM.
Junior Member
Why? I am dubious about how useful this really is.
1. Does the 30W quoted include an ethernet switch? Presumably you'd need at least 7 ports on the switch (6 for each of the units and 1 uplink port). Or are you assuming that the devices will be connected to a professional switch, in which case, what is the pro-rata power usage for 6 ports on a normal switch (e.g. cisco or something) ? I wouldn't be surprised if it exceeds 30W.
2. A modern x86 server would have at least 16 cores capable of running 32 threads, but would use a lot more power. But I suspect the x86 would still win on most workloads, especially floating point.
Anyway, I still think it's a fun project if not a useful one.
I have a friend who once built a cluster of 5 i386 boards (in ~ 1996, when they were long obsolete) . It was cool.
Senior Member
I don't have much experience on this matter, but it makes me curious... What method do you use to allocate the workloads to all the boards? Through what protocol - SSH? How will the nodes be managed - through LinuxPMI, or will they remain individual? What programs are well-clusterable?
I'm thinking about whether I could do something similar on a smaller scale with x86_64 hardware. I'm going to upgrade my PC soon, which will leave some capable hardware behind (in essence a headless terminal - motherboard, CPU, RAM), and since I often need to do tasks that are very computing-intensive (FFmpeg video manipulation, mostly), it would be pretty awesome to utilise it.
Junior Member
Expensive
This kind of bugs me because BeagleBoards come out at over $100 when there is silicon available at a fraction of this price.
I can't understand why these boards need to be so expensive? Lets just take the Rasberry Pi, granted it is under 1GHz and single core, but it is $35, doubling the cores and increasing the CPU speed won't cost more than $5-10.
We need some $50 Marvell* Armada 1500 based machines to be built and if I keep seeing these expensive boards appearing I might have to do something about it myself....
* These things run cooler than the Kirkwood of the PlugComputer fame.
Junior Member
not that cheap but quite interesting...
BTW: HDMI (1.4?) with a/v data support Ethernet/IP 100Mb/s ... hmm KVM/Ethernet someone?
on other side i don't see any benefits compared to LAN except processed video output(remote Renderfarm/DVD players/stuff ).
how about a bulk of Sitara AM335x (5$) or 25$ rasberyy PI usb powered? and hdmi out.. flashdisk size
Last edited by SunnyDrake; 06-08-2012 at 06:57 AM.
Posting Permissions