The difference between LART and PLEB: The Long Story^TM

History

The plans to make LART and PLEB are approximately equally old. About two years ago Adam sent a message to linux-kernel discussing processor choices for an embedded Linux board. At the time I had just noticed that no existing (as in: buyable) board matched my own requirements, and was planning to design my own PCB based on an embedded PowerPC. Adam formed the PLEBList, I joined. Through the PLEB ideas, I saw the One True Way (SA-1100 ;-) and decided to use the same processor for my LART.

We've had a lot of discussions over the years, and it became clear that there are some fundamental differences in our design philosophies. If this hadn't been the case, I wouldn't have bothered putting in the many man-months to make my own PCB. Note that in design philosophies there are no absolute truths, and that PLEB and LART are both `right' from their respective point of view.

The main differences are outlined below.

Size vs. `independence'

For me it is absolutely vital that the LART board can function as a minimal computer, without any daughter boards. This means that at least the processor, boot flash, (some) RAM and RS232 has to be mounted on one PCB, yielding a board size of 75×100mm. PLEB considers size more important and put RAM and flash on a daughtercard, reducing the board size to 90×60mm. This does mean that Photon has to use buffers in the RAM address lines, which increases access times.

Low weight (size) vs. low power

The LART power supply is vastly overdimensioned. It eats about 20% of the total board space and weight; it can deliver around 16W of output power on the core and IO voltage nets combined (I've never managed to push the power usage of a single LART beyond 1W). Because of this oversizing I could use components with very low internal resistance, thus minimizing losses in the power supply. The Photon supply is a much closer fit, decreasing system size and weight but increasing losses.

Bleeding edge vs. stable modularity

I'll keep the electrical and mechanical specs of LARTs to come identical to the current design. If there will ever be an SA-2 LART, you should be able to use it directly with existing modules. Adam feels less tied to the past, and will thus be able to make faster use of new technologies.

Hardware design

The LART board family is expanding rapidly. I've finished the design of the kitchen sink board, which offers an ATA interface, quad Ethernet, netboot, dual PS/2, 16-bit 44k1 stereo sound out, IrDA, video out, USB client, mono mic/speaker, touch screen interface and a telephony interface. More boards are yet to come, including a fast SRAM board (450MB/s read speed), GPS, PCMCIA, a frame grabber and a 2.4GHz RF interface (with datarates up to 150Mbit/s).

The biggest hardware toy will be the crossbar switch, which will allow message-passing between multiple LARTs with latencies in the order of one microsecond. I plan to build a ClusterLART which will hold around 100 LARTs, with an aggregate communication bandwidth of about 40GB/s (and yes, that's gigabytes per second, i.e. the same capacity as 320 Gigabit Ethernet streams). The ClusterLART will be used to do signal processing on large real-time data streams.

I don't know whether PLEB has any plans for Photon expansion boards. Adam?

Availability

I'd love to be able to sell LART boards, but selling things as a university is dodgy at best. I wouldn't have clue #1 about handling things like DOAs or foreign money transfers. I'm looking at building & selling LARTs externally, but haven't gotten very far.