Raspberry Pi interview: Eben Upton reveals all (Linux User) [LWN.net]

Raspberry Pi interview: Eben Upton reveals all (Linux User)

Posted Mar 6, 2012 5:06 UTC (Tue) by Imroy (guest, #62286)
Parent article: Raspberry Pi interview: Eben Upton reveals all (Linux User)

Why is the Beagle/Panda boards so much more expensive than the Raspberry Pi? For one, they have a lot more parts on their boards. They're sold as developer boards. So it's unfair to just talk about the price of the SoC and say the rest of the price of the boards must be profit.

Also, Broadcom are subsidising the price of the RPi. I think TI also subsidises the Beagle/Panda boards, so maybe there's just a difference in how much each manufacturer subsidises their board(s).

As for performance, I want to see the benchmark results. The multi-issue super-scalar Cortex-A8 and A9 cores are reported to be able to perform 2.0 and 2.5 DMIPS/MHz, respectively. The aging ARM11 core only does about 1.0 DMIPS/MHz. So the claim that a 700 MHz ARM11 (~700 DMIPS) is only 20% slower than a 600 MHz Cortex-A8 (~1200 DMIPS) just doesn't "add up". Maybe something like a faster memory bus is helping it perform better on a certain benchmark.

to post comments

Raspberry Pi interview: Eben Upton reveals all (Linux User)

Posted Mar 6, 2012 5:36 UTC (Tue) by alison (subscriber, #63752) [Link] (2 responses)

Upton's comments are unfair just because the Raspberry Pi is newer and, stop the presses, hardware tends to drop in price over time. The Pandaboard that I'm familiar with has far more ports and peripheral chipsets than the R.P. The development of the R.P. was subsidized by both the British government and the Qt Foundation, while T.I. paid for Pandaboard development out of pocket. T.I. has huge numbers of staff engineers on IRC and mailing lists and wikis patiently answering the same Beagleboard and Pandaboard questions over and over. We'll see how good support for R.P. is.

I strongly suspect that contrary to "making piles of money" on these products that T.I. has actively subsidized them, but I am not party to inside information.

Raspberry Pi interview: Eben Upton reveals all (Linux User)

Posted Mar 6, 2012 6:54 UTC (Tue) by koenkooi (subscriber, #71861) [Link]

The panda is subsidized, the beagle isn't. You can take the BOM, order the parts yourself and manufacture a beagleboard for slightly less than we're selling it right now. The key part is "order the parts yourself", you can't do that for a lot of chips, but you can for everything on the beagleboard and beaglebone.

Raspberry Pi interview: Eben Upton reveals all (Linux User)

Posted Mar 6, 2012 8:10 UTC (Tue) by kragil (guest, #34373) [Link]

Could you provide some proof for all those claims?

Raspberry Pi interview: Eben Upton reveals all (Linux User)

Posted Mar 6, 2012 11:26 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link] (1 responses)

Heh. I remember paying a few years ago more than $700 for a dev. board with the end-user device costing about $50.

Raspberry Pi interview: Eben Upton reveals all (Linux User)

Posted Mar 7, 2012 16:51 UTC (Wed) by juliank (guest, #45896) [Link]

And for NVIDIA, boards still do cost that much. There newest Tegra board costs about $1000, the older unsupported ones cost about $400.

Raspberry Pi interview: Eben Upton reveals all (Linux User)

Posted Mar 7, 2012 5:41 UTC (Wed) by geuder (subscriber, #62854) [Link]

> As for performance, I want to see the benchmark results.

I would be satisfied with real user experience. If it is a desktop used for browsing, how fast does it feel.

> The multi-issue super-scalar Cortex-A8 and A9 cores are reported to be able to perform 2.0 and 2.5 DMIPS/MHz, respectively.

The problem with all the nice multi-ussue, super-scalar (and you don't mention multicore for the A9) is that it needs a suitable workload and excellent compiler support or even hand crafted assembler (e.g. NEON) to perform optimally. All the nice high tech does not help if it waits for the network or the mass memory. Unfortunatley it often enough even waits for its "own" cache. I have seen A8 running at less than 30% of its theoretical performance far more often than I would have liked.

> The aging ARM11 core only does about 1.0 DMIPS/MHz. So the claim that a 700 MHz ARM11 (~700 DMIPS) is only 20% slower than a 600 MHz Cortex-A8 (~1200 DMIPS) just doesn't "add up"

Well maybe the imbalance between a too fast CPU and a two slow everything else is just not as bad on the older systems. So I would definitely not rule out a smaller observed difference in performance than what the theoretical maximum figures tell.