[Arm-netbook] Tegra2 OC-ing on Toshiba AC100

[Gordan Bobic]

On 12/31/2011 07:24 PM, lkcl luke wrote:
> On Sat, Dec 31, 2011 at 7:09 PM, Gordan Bobic<gordan at bobich.net>  wrote:
>> On 12/31/2011 06:54 PM, lkcl luke wrote:
>>> On Sat, Dec 31, 2011 at 6:36 PM, Gordan Bobic<gordan at bobich.net>    wrote:
>>>> I expect this may be of interest to some people on the lists:
>>>>
>>>> http://www.altechnative.net/?p=332
>>>>
>>>> It covers the hardware mod to improve the cooling, and includes kernel
>>>> patches for both unmodified 1200MHz and modified 1456MHz operation.
>>>
>>>    wild!
>>
>> They OC really well. Every single 1000MHz CPU I have hits 1200MHz with
>> no problems. I only have one (out of 5) that won't do 1248MHz reliably
>> without the cooling mod (it errored out twice in the 24-hour full load
>> testing period, unlike the others), which is the only reason the patch
>> for unmodified laptops is limited to 1200MHz. I have only modified 2 so
>> far, but they both hit 1456 without any stability issues. The thermals
>> will cause a problem when the GPU starts pumping out heat, too, but
>> that's what nITRO is for - it throttles the CPU back to standard maximum
>> clock speed of 1000MHz (or whatever you configure it to) once the
>> temperature exceeds 90C (also configurable). The top clock speed is also
>> configurable (OC speeds are 1200, 1248, 1352, 1404, 1456).
>
>   very cool.  well, it's 40nm so hmmm where are those power stats on
> ARM CPUs?   oo ouch, the site on arm.com says 1.5 watts @ 2ghz but 0.5
> watts @800mz if power-optimised.  so yeah the GPU must be pushing out
> quite a bit.

Unfortunately, Nvidia publish neither the operationsl temperature limits 
nor the TDP wattage for Tegras. :(

>>>    btw turns out that the guy who did the nvidia desktop GPU
>>> reverse-engineering took a quick cursory look at the tegra and found
>>> it was "odd" but not seriously seriously different.
>>
>> Does that mean he might be prepared to look into reverse engineering the
>> Tegra GPU in exchange for donations to the project?
>
>   worth asking him.

Can you do it? Or email me his email address off list and I'll ask myself.

>> It would be nice to at least have OSS supportable GPU options other than
>> the Volari Z11 - not that there are any GPU-less ARM boards that have a
>> fully wired PCIe/m-PCIe slot... The only ARM board I have seen with a
>> fully wired PCIe slot (alas, the AC100 only has USB lines wired up to
>> the slot!) is the Compulab SBC-A510, but that has a Marvell Armada 510
>> on it which already has a GPU of it's own.
>
>   achh... *sigh* the other problem is that the tegra 2 doesn't have
> SATA - the compulab machine managed to add SATA through a USB-to-SATA
> converter.

Tegra2 has PCIe. They couldn't get a SATA chip working on PCIe, granted, 
but the fact they couldn't do it doesn't mean it's not doable. SATA over 
USB is going to significantly impact both the latency and the transfer 
speeds. Having said that putting the upper limit of IOPS at 1000 isn't 
actually all that crippling if you're not building a system that 
requires ultra-low-latency storage access.

>   the tegra 3 on the other hand wooooow... :)  SATA-II, 4-lane PCI-e,
> and more besides.  utterly cool.  oh, and developer boards were just
> announced on the 18th dec a few days ago.

Asus Transformer Prime is due out in UK in 3 weeks' time, and that will 
have Tegra3 in it. But I won't be acquiring one until full fat Linux 
kernel is bootable with support for all the major components (what are 
the chances that Asus might comply with the GPL and provide the full 
kernel sources?).

In the meantime, however, I'm hoping that the arrival of Prime will 
hammer the prices of the old model down to a more sane range close to 
the AC100, even if only because the Transformer has 1GB of RAM, which 
would be a nice improvement on 512MB in the AC100 given the current 
generation of bloatware.

OTOH, being forced to work in a more modest amount of RAM might actually 
lead to people re-learning to program properly (or better, leave the 
software industry and go and work in an industry more suitable for their 
talents such as the fast food industry). Firefox has no right to require 
168MB of RAM when Midori does the same job in 71MB. Not that I don't 
think 71MB is excessive for a web browser, too, mind you (by a factor of 
10 or so) - but it shows that uselessness certainly has degrees.

Gordan