Hacker News Reader

Comment by EwanToo | Hacker News Reader

One interesting thing about the Cubieboard is the SATA port, otherwise it's very similar to a lot of other ARM boards out there (e.g. Beagleboard, Pandaboard, etc). This makes it much easier to stick a low power SSD on the device, rather than rely on a hugely slow SD card.

What I'm looking forward to next are Cortex A15 boards, the first CPUs are shipping (In the new Chromebook), so hopefully we'll start seeing them on small hackable boards soon.

View on HN

Another board is say: https://www.miniand.com/products/Hackberry%20A10%20Developer...

With WiFi. Other alternatives are at the same site.

I have a few Mali devices next up on the chopping block after I finish reversing the VideoCore IV.

I have a few Mali devices next up on the chopping block after I finish reversing the VideoCore IV

Hehe. Reversing a GPU is a lot of work. From my experience it usually involves zillions of commands, state bits that subtly change handling, and different instruction sets for different kinds of shaders and other sub-processors... (even figuring texture formats can be lots of work) I can't even imagine starting such a project again, though it was fun, wish I'd still have the time...

I wish you the best luck of course!

Wladimir, I respect your view here - I agree with everything you write. My field of expertise is more ISA encodings, overall functional block architecture and decompilers. Once I hold the encodings, I point at the blob and say there is your source & specification. ie Binary blobs + ISA encodings = specification to me.

>> after I finish reversing the VideoCore IV

I've only just stumbled across details of your efforts in the past couple of days.

I wish you every success, if you were to be successful, you would open the door to impressively cheap possibilities.

Hey napierzaza, you've been hell banned.

I didn't realize there was a significant speed difference between SD and SSD technology! After some googling, I found a good video that gives the relevant figures:

http://www.youtube.com/watch?v=5fxAQv_KDV0

In short, the class 10 SD card I got for my Raspberry Pi is rated for 10MB/s. An SSD drive can do 1GB/S (I imagine the numbers vary for reads vs writes). Also, SD cards seem to be able to take a lot fewer writes per block before they wear out.

1GB/s is a bit much for an SSD, since SATA currently only goes up to 6Gbit/s. PCIe-based SSDs can do it, but for a good SATA SSD, you can expect speeds ranging from 100MB/s to 500MB/s depending on the workload.

Most of the speed advantage SSDs have comes from having a dozen or more flash chips that are internally treated as something like a RAID array. Memory cards and USB sticks usually have only one or two chips.

It's quite significant. SD card is fine for the hacker community though. They have different intentions I suppose. SD cards weren't really designed for OS (particularly if you forget to disable a pagefile which will constantly write old memory to "disk" and drastically shorten its life)

To be fair, a class 10 SD card could be ten times faster than the figure you quoted. Still a big difference, of course, and I imagine there are some write queueing feature differences as well.

In theory. On the PandaBoard I max out a class10 card at 30 MB/sec.

SATA port also means it's viable as a small NAS solution. Shame it doesn't come with gigabit Ethernet.

Absolutely, there's all sorts of things that can be done which aren't really workable without some better performance/capacity storage than an SD card (even the nice new 64GB ones)

IO performance on these CPUs is fairly poor, so gigabit is unlikely to actually be worthwhile.

On my TS-7800 [1] which has GigE and a 500MHz (!) CPU, Samba outperforms the Raspberry Pi (15MB/s vs 7MB/s). Note that on the RPi not even the whole 100Mb link is utilized despite the CPU being slightly faster in general (yes, I did some benchmarks) - and the CPU on the TS-7800 was actually designed for storage applications. So there may be something on to your claim, and it would appear that CPUs that include graphics and stuff are not that good at I/O.

[1] http://www.embeddedarm.com/products/board-detail.php?product...

You may already know this, but the Ethernet controller on the Pi is actually a USB device, built into the LAN9512 [0] USB controller/hub.

The TS-7800 has a dedicated Ethernet controller, which presumably explains some of the difference.

[0] http://www.smsc.com/index.php?tid=300&pid=135

...it would appear that CPUs that include graphics and stuff are not that good at I/O.

I made the same observation on my blog[0]. It's a shame that decent I/O and fast CPUs/GPUs are rarely if ever combined in ARM systems. What good is a quad core CPU if you can't get any data to it?

[0] http://nitrogen.posterous.com/the-ideal-arm-platform

That page says "Quantity 100 starts at $229." Does that mean one board is $2.29?

I don't think so. Bulk pricing usually lists the per unit price. The quantity is denoted since you can get better pricing for larger quantity orders. For example, I may get a microcontroller for $4 when I order 10k, but you'd pay $25 if you just ordered one from DigiKey.

No, it means the price is $229 each if you order at least 100. Without the quantity discount, it's $269 each.

In case you want something hackable, check out Olimex.com.

In case you're afraid of real hacking ...

FTFY ;-)

Isn't SSD disk going to cost more than this board?

I think the final price of the Cubieboard is $50, the cheapest SSDs start at around $50 too [1][2], so yeah it's going to add to the price, but it's also going to make it useful for a lot more projects than an SD card.

1 - http://www.newegg.com/Product/Product.aspx?Item=N82E16820227...

2- http://www.crucial.com/store/partspecs.aspx?IMODULE=CT064V4S...

Yes, even crappy ones are going to be twice the price of the board itself.

Not sure why that matters?

Reply on news.ycombinator.com