[fedora-arm] 1ghz ARM Laptop (12in 1280x800 LCD)

[Gordan Bobic]

Matt Sealey wrote:
>>>> Unfortunately, I'm not a hardware hacker. But, as a consumer, I'd say that a "1gb NAND Flash" is quite a bit below the level. I also wouldn't care much about a 1280x720 screen if the hardware wouldn't be capable of playing the video flawlessly. Or, if there was an HDMI port to connect to TV, which is, in my taste, better suited for watching.
>>>  yep - all the designs i've worked on, and all the CPUs found (so far)
>>> were selected precisely because of the HDMI output capability.  i'd
>>> kinda ruled out the spea1310 because you need an extra IC converting
>>> LCD to DVI/HDMI or even *shudder* a PCI-e graphics IC (volari Z11 god
>>> help us is about the only free-software-compatible option)
>> I was just thinking about that, actually. If you're making a custom
>> mobo, then as long as you can find an ARM SoC tht has PCI-e, you could
>> just apply an MXM module and plug in an ATI or Nvidia GPU for which we
>> already have passably working OSS drivers.
>>
>> Of course, this defeats the purpose of the exercise - who in their right
>> mind would use a 2W CPU with a 30W+ GPU in a laptop?
> 
> Not to mention cost. Unless it is in significant (>100,000 unit)
> quantities, MXM modules are $100 a pop. Even then, it is not a great
> discount.
> 
> Who in their right mind would pair an $18 SoC with a $100 GPU? It is
> design insanity to quadruple the BOM like this.

Indeed, it's not sensible even if does gain the "OSS drivers" feature.

>>>  gordon is right about the SD/MMC card thing, but the "level 10" ones
>>> can at least guarantee above 10mbytes/sec *read* capability.   so
>>> _yes_ to the SATA interface.
>> The 10MB/s is _supposed_ to be for worst-case sequential writes. There
>> is, however, no defined benchmark
> 
> (there is, but you need to be a paid up member of the SD Association to get it)

Whatever it is, it appears to have been designed specifically to be 
completely meaningless.

>> I'm glad we agree on anything other than SATA being unworkable. :)
> 
> I will also point out that a SATA NAND controller that is capable of
> more than 20MB/s is going to cost you about $2.50 per gigabyte,
> minimum $35. 2.5" SATA SSDs disks are also rather large for the
> purpose. Sandisk make some nice MicroSATA SSDs as do, now, Toshiba if
> you have a December 2010 MacBook Air, but they are either not as fast
> as you want (20-30MB/s), or more expensive than the rest of the board
> put together. There is a very good reason Apple's hardware costs so
> much money and it is absolutely not price gouging.

It's not about bandwidth, it's about latencies. Embedded NAND just 
doesn't seem to deliver and non-SSD modular flash (e.g. SD/CF) is 
appalling. A SATA controller would fix the problem because you could 
plug in a $100 X25-V and be done with it (instead of a $200 SD card 3/4 
of the size and 1/10 of the performance). If I can get 1200 IOPS out of 
the SD card on reads, the interface can handle it - I'll be more than 
happy with 1200 IOPS on writes that the SD interface can deliver, but 
the suitably performing media doesn't exist. SATA media handles it just 
fine, OTOH, and given the most recent X25 series shrink, it looks like a 
no-brainer when it comes to TCO. I'd rather pay an extra $100 for the 
base machine + another $100 for a decent SSD than pay $100 less for the 
machine and pay $200 for the flash media that is 10x worse. The TCO ends 
up the same but the performance in the SATA case is 100x+ better. Not 
really a difficult decision. SD and CF media is vastly overpriced and it 
sounds crazy to use it when there are much better and cheaper alternatives.

> If you want cheap, plentiful and fast NAND on the board, you would be
> better off using the internal flash controllers and soldering the
> chips to the board. However you are limited by size and of course the
> performance is not as you would expect (NAND disk does not necessarily
> equate to 215MB/s Intel X series performance. Intel have a very
> specific configuration to get that speed. Embedded NAND controllers
> won't do it. Embedded SATA controllers won't even get the full
> performance of the Intel devices).

It doesn't need to hit the performance that the Intel X25 can deliver, 
it just has to enable usage of a device that can saturate it. SD cards 
just don't cut it - period.

> This is exactly where you will get if you're trying to massage the
> Tegra2, MX53, OMAP4 into a high end laptop; it will not meet your
> expectations.

My expectations in this regard are:
1200 IOPS on random 4KB reads and writes.
SD interface can deliver this read performance easily.
_NO_ SD card can deliver this write performance, but a SATA SSD half the 
price and double the capacity can. Sure, the disk may not reach it's 
potential limit but so what - you'll still end up with a laptop that is 
100x faster for less cost or at least comparable cost.

> They are not designed for those environments. Tegra and
> OMAP4 are phone/tablet chips almost entirely focused on Android. MX53
> is for in-flight entertainment, Ford Sync, handheld media tablets,
> that kind of thing. Remember when you think about speeds on devices,
> they are always listed as maximums. Yes, SATA-II is 3Gbit/s (actually
> about 2.4Gbit after you get past the 10-8 encoding) but does a device
> need to run at that? Not at all. USB 480Mbit/s? Nope.
> 
> You can't design a system, keep it cheap and make it a volume seller
> by just picking the chips with the biggest numbers.

I'm not talking about chip-picking I'm talking about price/performance 
of components. This wouldn't be an issue of the port on the back of the 
Efika MX was suitable for a 1.8in SATA SSD. Unfortunately, it's uSD, and 
that means it is limited by media availability to 1 write IOPS with more 
expensive media. For anybody intending to use this non-removable 
expansion slot (e.g. for a different OS), that is a seriously false economy.

Gordan