DreamPlug ARM Box Brings Power To Plug Computing 182
Blacklaw writes "UK GlobalScale distribution partner New IT has announced its latest ARM-based plug-top computer, the DreamPlug — and it's a major improvement over its predecessors, packing some serious hardware into a tiny case. The DreamPlug packs some serious power in its tiny case. The Marvell Sheeva ARM-based processor at its heart runs at 1.2GHz — a significant boost over the 800MHz version found in the SheevaPlug — and while 512MB of DDR2 memory might not sound very generous, if you need more then your project probably isn't suited to the plug computing model. Unlike the SheevaPlug, the DreamPlug goes all-out to impress, packing integral Bluetooth, 802.11b/g Wi-Fi, a 3Gb/s eSATA port, two USB 2.0 ports, a pair of gigabit Ethernet ports, and even analogue and SP/DIF digital audio ouputs. ARM developers will be pleased to hear that the JTAG-over-USB feature of the SheevaPlug has been replaced with full hardware JTAG and UART connectivity — although the breakout board is an optional extra."
Re:Serious Hardware in 1997... (Score:5, Informative)
The linpack benchmark focuses on floating point, whereas most ARM chips don't have hardware floating point units...
ARM chips tend to do much better at integer benchmarks, and most code you would run on a server is integer code.
That's just the UK reseller (Score:4, Informative)
The US version is supposed to ship this month. [globalscal...logies.com] The developer kit is $149, and $179 with a JTAG interface (recommended for development.)
The production version will probably be cheaper.
Hopefully they've fixed the overheating problem they had with their previous GuruPlug.
PogoPlug (Score:3, Informative)
Re:Windows... (Score:4, Informative)
Yes. But will it run Windows?
Your question was probably a joke.
Microsoft announced that they are developing Windows for the ARM chip, at the Consumer Electronics Show in January 2011. So I think the answer is no, not yet.
Will probably evaluate one, but (Score:4, Informative)
I bought the Sheevaplug and the Guruplug for some engineering applications, and was sorely disappointed. The Sheevaplug was a decent box, just needed more native IO. The Guruplug was a piece of crap. I had more issues with that box than any embedded box I've ever worked on, including some I've designed (which is saying something when you factor in initial debug time). The Guruplug had major heat issues, even when run from an external 5V supply. I removed the heat spreader (a thin piece of steel) and replaced it with a thicker copper spreader, and that made a big difference, but the unit was never completely stable and could not handle running two GigE interfaces at the same time. And they also had the niggling little problem of selling something different from what they advertised - the sale product did not have an I2C port (I think they finally changed the block diagram to reflect the truth).
By sticking with the same form factor I fear that the Globalscale product will continue to be plagued with heat issues. And based on the history of Globalscale's products, if you need a stable platform that does what the specs say they'll do - look elsewhere.
I'll probably get one to evaluate it, but this time I'm waiting. Someone else can be the early adopter.
Re:Wait, what? (Score:5, Informative)
But it's not perfect: the USB chipset is a bit dodgy. I have five hard disks, an ethernet widget, and a few other devices hanging off the SheevaPlug's one USB port and it's not happy --- I had to spend some time fiddling with it before it ran reliably, and there's still a nasty bug where every now and again the USB ethernet adapter stops processing packets (although the internal ethernet port is fine). (Replugging the USB ethernet adapter fixes it.)
I've been looking at the GuruPlug with great interest; real eSATA and two USB ports would make my life much easier, but I've held off getting one because of the heat problems. Maybe this DreamPlug will be the solution.
Re:Serious Hardware in 1997... (Score:4, Informative)
Most recent ARM chips (except the very low-end ones) do have hardware floating point, and even vector units. There are two issues, however:
The first is that most ARM Linux installs use a completely braindead ABI designed for to allow you to mix FPU and softfloat code. An ARM chip with no FPU traps on all FPU instructions, including loads and stores, so you can save a lot of overhead by passing floating point values as arguments in integer registers. Unfortunately, on ARM chips that do have an FPU, copying from an FPU register to an integer register causes a pipeline stall for each parameter. You can speed things up a lot by storing the floating point values on the stack and passing them by a pointer to work around this.
I find it somewhat ironic that a Free Software system would default to using an ABI that exists for the sole purpose of making binary-only distribution easier.
The second issue is that it is only 32-bit that has reasonable performance. 64-bit float (which LINPACK does a lot of) is very slow - typically a factor of seven or more slower than 32-bit - while it's close to the same speed on x86 (exactly the same speed on a P2, because both are done in the 80-bit x87 unit, may be up to half as fast in the P3 if you're using SSE instead of x87).
The reason for this focus is that most ARM chips come with a GPU and DSP on die - if you're doing anything really floating-point intensive, you're probably going to want to run it on the DSP or maybe the GPU. The FPU is just there to avoid floating point code from becoming a bottleneck in general-purpose code.