Improved over-clocking headroom

Will improved fabrication allow later pi’s to be over-clocked to a much higher level? Just wondered if a pi that defaults to full 100% compatibility with existing pi’s was possible but had overclocking potential to take the gpu to 1ghz and perhaps the cpu to 3ghz. It would be nice for the pi to remain 100% compatible with the original for a few years more but offer improved headroom for overclocking.

Actually I don’t even know the current fabrication process for pi, is it 40nm or better than that?

Still massively enjoying the pi, with risc os, raspbian and openelec xbmc. Just thought I’d throw that in.

I googled for “over-clocking headroom” to find out what it means, and found this splendid photo…

..with the caption “Ditch the puny* stock heatsink (left) for something beefier (right)”.

When your Rasberry Pi has been overclocked beyond your wildest dreams, please send us a photo of its tiny board surmounted by its gigantic heatsink :)

 
*puny = 3½ x 3½ inches across.

Martin:

but had overclocking potential to take the gpu to 1ghz and perhaps the cpu to 3ghz.

I suspect you misunderstand what the Pi (and its associated Broadcom SoC) is about. If you want raw speed, try an iMX6 or the like. Note, however, that they run at 1.5GHz. I’m not sure there is a commercial ARM that runs at speeds like those – ARM isn’t trying to compete with x86…
I think this is sort of state-of-the-art, a quad core 64 bit device running at (IIRC) 2.3GHz. But there’s a massive gulf between announcing a new processor, and making production silicon with stuff to run on it.

The Pi, certainly, would not be likely to turn up in a better/faster form in a hurry. Bells and whistles like that would bump up the price. As it is, we’re pretty lucky to have a SoC that runs at a sedate 700MHz but can handle being overclocked as far as 1GHz (depends on the chip). That’s a heck of a jump, and is a testament either to the capability of the Broadcom engineers to build in those sorts of margins, or maybe the paranoia of Broadcom engineers to build in those sorts of margins. :-)

Either way, please note well that the Broadcom SoC is best considered a powerful and capable GPU with a middle-of-the-road (slightly outdated) ARM bolted on for the boring operating-system stuff. It’s the GPU running the show and calling the shots, not the other way around. But, hey, who am I to complain? RISC OS runs okay, RaspBMC runs even better…

It would be nice for the pi to remain 100% compatible with the original for a few years more but offer improved headroom for overclocking.

Doubtful.

For certain, the popularity of the Pi may have made Broadcom revisit this and consider the future. I would be surprised if the RPi guys were not in talks with them regarding future silicon.
However there is a large problem. It’s an old ARM core. Nobody produces ARM11 tech these days, it’s pre-Cortex-era. So, do Pi&Broadcom revamp the older SoC and try to squeeze more performance from an old design (old enough that Canonical royally missed the boat by deciding not to support that architecture), or do they create a Pi-alike using newer core designs (and stick it to the OMAP!) but run the risk of numerous compatibility problems. I don’t think the GPU needs to change. If it can cope with Full HD, that is – frankly – a lot more than we have a right to expect from this price bracket. But if the ARM changes? There are a number of subtle little things that can cause potential problems.

Actually I don’t even know the current fabrication process for pi, is it 40nm or better than that?

40nm.

This is fun too: http://www.raspberrypi.org/x-rays-no-specs-required/

Still massively enjoying the pi, with risc os, raspbian and openelec xbmc. Just thought I’d throw that in.

Sure. We’d all like to see what our hardware is capable of. Just don’t lose sight of what the Pi was designed for and you’ll see that the ability to overclock as much as we can is an unexpected benefit, but not a design criteria.

Tennant: Is that a CPU heatsink or the front end of a motorbike? Whoa….!

Whoa is right, it’s a big beast!

And to think that article was written seven years ago…

Well I was thinking perhaps a pi at 22nm/28nm would firstly use even less power which would benefit many projects and also be reduced size. The use of certain components on the board would guarantee more effective over-clocking. Components that would cope with overclocking more easily. I understand that a revised pi may come along as a new generation, perhaps multi-core cpu etc but that of course needs new software and updates. Consoles like the ps3 and 360 go through many generations of fabrication, the gpu started at something like 90nm, then 65nm, then 40 or 45nm and perhaps even smaller in the future. Each time the gpu is capable of higher speed but of course is kept at a stock speed for compatibility. Each time the gpu has been revised slightly to suit the improved fabrication but functionality remains the same. That layout change may be to optimise cooling or other issues.

It may be the existing pi would have more overclocking headroom if certain components were replaced to aid that goal. Perhaps next year a pi that easily clocks at 1.2ghz cpu and 700mhz gpu and can be pushed to 1.5ghz quite easily with a 1ghz gpu.

It may be the existing pi would have more overclocking headroom if certain components were replaced to aid that goal. Perhaps next year a pi that easily clocks at 1.2ghz cpu and 700mhz gpu and can be pushed to 1.5ghz quite easily with a 1ghz gpu.

It might be that a smaller fab size, and the inclusion of a heat extracting metal slug, would improve speed. However the cost per unit would rise for the SoC and hence the board. Increased board cost might not be in the Pi Foundation roadmap.

40nm is the norm now but how long before a reduced fabrication size becomes equally as cheap. You probably couldn’t get 65nm fabricated now due to lack of fabrication facilities (pure guess) and as perhaps state of the art becomes 22nm, 32nm becomes the budget choice. Not sure when that would be though but surely within the next 12 months at the very latest.

You probably couldn’t get 65nm fabricated now due to lack of fabrication facilities (pure guess)

I expect some 65nm production facilities will have closed/will close but others will remain. There are many 30+ year old chip designs still in production and will continue for many years to come e.g. 74LS245 which I’m sure will be using a much wider process than 65nm.
Though some manufacturers may not bother moving production to another facility if the type/size of production ceases at their current fabricator facility if sales volume is too low to justify the effort.

What Chris said. There’s a lot of slower devices – AVR, PIC, a billion Chinese knockoff MP3 players, all the 74 and 4000 series logic… none of which requires cutting edge process.
If you’re running a smart washing machine on an 8501 clocking a megahertz, you don’t need tiny, you just need cheap.