Intro & Faq

[Da_Blitz]

using the keyboard scaning logic makes it eaiser to go to low power mode, sure the usb spec can handel it but turning off the USB logic in suspend means greater power saving

as i see it there will be a power button for turnig on the device initially (or perhapes power on when power is applied) and let any keypless wake from suspend

however a dedicated usb device does allow for some wierd posabilities, for example using the usb "bridge" feature to connect the usb keyboard directtly to a host computer

if i remeber correctly there should be 24 gpio's free if i use the keyboard handeling stuff

as for the "backplane" idea, usb is fine for loading the firmware and you could use it for comms but i had a better idea in mind

make a pcb with some flash, ram, a fpga, an ethernet port and multiple controlable powerdomains on it with several PCI connectors, then make stripped down cards with the cpu/ram/fpga combo on it. basically you can swtich power to any slot on and off, allows hotpluing of cpu cards (usb handels detection) and use a fpga to implement a layer 1 and perhaps 2 OSI comms model. eg 32bit DR @ 133Mhz (233Mhz effective) the fpgas have this logic built in to talk to ddr ram but it would make for a nice way for these things to talk

its a cluster in a box of sorts. but its a better way of giving these things gigabyte comms in some ways (or @ 64bit, 400Mhz you get 3.2GB) you mem map it as sram and you get low latency comms between cards with enough speed to make rebote hardware (in this case hardware not attached to the chip) appear as local, which is my goal as well as some cluster and load balence stuff. its cheap

as for the actual bus, could be wishbone, wont be pci but i would rather make it seems like an ethernet card with RDMA capabilities and some ultra heavy offloading capabilites (eg scatter grab and assembley of packtets done by the card instead of the kernel)

but its a fpga so chuck whatever you want in there, and wihle your at it you could add a cryto offload egine in the spare space

lets just say i have refound my intrest in the project as well as finding a way that this design may be useful to more people

[Ragnorok]

- "Cluster in a box" sounds nice, but that seems like it'll be a bit of overkill for a proto system to slap an OS on for testing.  (crinkly grin)  If they are that low latency we should make a "hypercube" out of them and market the whole thing as an affordable Grid computer.  "Yes sir, I know the CPUs are slow, but there are hundreds in this one suitcase-sized box.  Your brain switches at 1Khz and look what it does.  Throughput on this will knock your socks off."
- But I digress.
- Let's say hypothetically I decide to try to port an OS to this beast.  What would I expect to receive as a prototype and how would I interface to it?  Pretend I have a pretty decent familiarity with programming in C/C++ and know hardware well enough, but haven't designed squat with electronics for 20 years so my understanding is almost hopelessly out of date.  Would this be one board and an USB connector to load the firmware?  Something else?  If it's just a CPU board how does one interface to it for "keyboard" and "display" to see it do something?
- Or am I just going off on a tangent?  (shrug)

[Da_Blitz]

you would get a pcb with a cpu and 64MB of ram, ram size may change but for the dev version i doth think there is needs for alot

it would have a pinout for a usb connector but wuold be fitted with a edged card like the PCi socket, so you can solder to that or plug it into a back board

there would most likly be a fpga on board for a bit of fun and IO stuff. and thier would be a serial line as well

the hope is that it would all plug into a backboard and they would all have an ethernet equivelent connection between each other and nfs thier root fs off node 0, all force fed over the devices usb client port, if you wanted to run it without a back board then you lose the network connection and do everything over usb, includhing loading firmware

so load firmware then move to usb-net in the downloaded (into ram) rom

[Ragnorok]

- Cool.  What about keyboard emulation and such?  If the system should boot and load drivers how will the dev board do that?  Or will it simply not do that initially to get the basic OS whipped up and bootable?  If it doesn't worry about trivialities like keyboard and display initially, what about later?  All these devices are supposed to be part of the final unit, but none will be on the prototype board.  Will those be added later, somehow, then the drivers bolted on then?
- I presume this CPU board would be a precusor to the bona-fide Pocket Penquin.  Perhaps that's off the mark and you're concentrating more on "blades" than on the PPC itself?

EDIT:
- Just read up on FPGAs.  Wowzers!!  Are we using a small one glue, or is this thing going to be built from heuristic hardware??  (wolfish grin)

[Da_Blitz]

I am concentrating on the blades more, they are the base for the other designs i have but i was betting on a device comming out that would match my needs or wants closly, and it was anounced today (http://www.engadget.com/2007/06/14/compal-showcases-prototype-umpc-running-windows-ce/ http://translate.google.com/translate?u=ht...language_tools)

time will tell if it makes it to market but i have a feeling it will, but i also feel they will shave the ram down a bit and make the case nicer (in black please: http://64.233.179.104/translate_c?hl=en&ie...language_tools)

and it supports USB host ( http://64.233.179.104/translate_c?hl=en&ie.../language_tools , look at the specs)

i was thinknig about a spartan 3, you can get some heafty devices in that range in many diffrent price rages that are pin for pin compatible

for anyone else reading this, look up what a fpga does, i can give you a brief but its mind blowing stuff.

essentially its the equivelent of what software is to a cpu only for hardware. its dynamically reconfigurable hradware that can one seccond be a tv tunner, the next a distributed key cracker and the next a PC. and they are MASSIVLY parrelel, one of these doing an optimised fast furior transform is over 50x faster than the best x86 computer you can put together

x86 can only exectue one stage at a time of the fft, this thing can do all stages in parrelel (so around 2048 operations in parrelel per clock)

[Ragnorok]

device comming out that would match my needs or wants closly

- Interesting do-dad, but I missing what the CPU speed is, though 256MB RAM looks nice.  It also only has a single micro-SD slot and no hard drive; the former (currently) doesn't support as much storage as full SD nor does it have peripheral capability, but that may not be an issue with WiFi & BT built-in.  The latter is more of a restriction imho because any "full" Linux will run better with swap than without it.  Even pdaXrom runs faster and is happier with the hard drive in a 3200 compared to no drive/swap on a 1000, based on how T3_slider says it runs on his Z and how mind-numbingly slow it is on mine.

i was thinknig about a spartan 3

- ???

[Da_Blitz]

spartan 3 is a xilinx fpga, has small caches of ram and some embedded logic in it. on the pda design i would use a cpld as i wouldnt need the higher logic fuctions, only things like adress decoding (CPLD's are lower power but much simpler devices, you couldnt cram a uP in a cpld)

still undecided, i have a couple of things happening all at once that are related to this but i should have had a design by now. i think the one thing that has held me back is trying to satisfy everyone. i should just skim the features to what i want, add in anything else i can and run with it.

one of my "wants" however is a profesinal looking case

[Ragnorok]

- I hear ya.  I finally lost Hiroshi and ordered an UMID BZ to replace it, so I'm Z-less until further notice.  There it is...