Future of the CPU board?


28nm process part. Perf/Watt isn't going to be much better. When they go to 20 or 14, then we're talking.
 
Where do you see that? I don’t see anything indicating that to be the case.

My bad, just double checked, I got confused by a picture of some FPGA expansion board on the bottom of the page. It does seem that they have a real SoC...
 
Broadcom's off-the-shelf SoCs are all about network related solutions, though - everything else they do is custom.
 
SiFive has a board that I believe currently has people working on Linux support for.

https://www.sifive.com/boards/hifive-unleashed

Ah, that's second or maybe later silicon. I've found the video I saw, which apparently uses a Kendryte K210 CPU, while this newer board is based on a U540 SoC. That initial board apparently ran micropython inside WSL inside Win 10 embedded, which sounds like a lot of effort to go to to me, but this new one supports Linux natively. Still, they're charging a dollar less than $1k to buy one, which is beyond my budget for shits and giggles at present.
 
I have always wondered if someone took the old 80386 CPU 1.5µm to 1µm lithography masks and shrank them - and the rest of the system platform of a 'maxed out' 1987 80386 DX with on-die 4GB of RAM, an on-die 16bit Soundblaster equivalent and an on-die SVGA card equivalent, IDE drives connection - pile everything that would have been in a 'maxed out' desktop machine form 1987 into a single chip - would it be any good?

It would lack any control for WiFi, USB, SATA, etc... So, that part would suck...

Actually, I would totally love to see something like this, but based on a 486dx4/100. Reason being is that a TON of things written around the time period of the 286,386 and 486 lines were best run on a 486dx4/100. It would be the definitive time period for which to build a 'retro' computer for the time. Windows 3.1, or even Windows 95 would be completely functional on something along those lines. Instead of software emulation. Just my $.02.
 
That would not be available (?), and hard to get to be quick enough and compatible enough to run all the peripherals.
For the ultimate benefit of catering to a specific emulation, a little bit better.
Then again those are a limited amount of people, but in a demographic of people with money to spend.
However, it can't take off, other in a hipster way, as it will remain niche for the sake or tradeoff of being so.
If you want new games or opportunities, and especially capture new markets, it is either faster, or more secure, or more battery life.
Marketing the current offering to new markets will perhaps open up opportunities. Audio, high security and camera people all have money.
Beyond the current offering, which is covered, I see risc-v being the most "more pyra than the pyra" in terms of openness and dealing with
agreeable companies. That or though now with the relatively new NXP wrt freescale, the imx8m, to share efforts and possible purchasing power with the librem 5.
But actually both. I think maybe the risc-v people would want and are so inclined as to help offer their product a platform, to mutual benefit.
Right at the beginning of either, there are unserved and very willing customers wanting to buy, same with a metal case (openpandora), and slightly bigger or e-ink screen.
Once the basics work, adding more offerings is a case of seeing what the future brings, and who will line up for each option.
 
In the end, though, for the Pyra's application there are constraints. Having the right peripheral connections, availability, and performance per watt - being able to be clocked -down- far enough to stay within the thermal envelope come to mind as the top of that list.
 
Bringing this back.

28nm process part. Perf/Watt isn't going to be much better. When they go to 20 or 14, then we're talking.

I agree that better process will give better performance. However, accessibility is a big issue here. Powerful (enough) chips being built on these improved processes are pretty expensive and difficult to source. You typically need to be a big name purchasing large quantities to get access to the latest and greatest or even get close.

That’s what’s nice about the iMX series and why I think the iMX platform is perfect for this project. This platform provides good access to new technologies in low quantities for a solid price. I could go on digikey right now and purchase a single iMX8M for around $50. I could also get access to the TRM and relevant documentation.

Even though the iMX8 series is built using the same 28nm process, it will still provide improvements over the A15s, which are notoriously horrible when it comes to thermal performance. I recommend giving the following a read if you’re interested and feel able to digest. https://www.tomshardware.com/reviews/arm-cortex-a72-architecture,4424.html#p1


upload_2019-8-20_16-26-34.jpeg


I definitely think the iMX8QuadMax, which is set to launch by the end of the year, is worth checking out as the successor. It features A72x2, A53x4, M4x2, DSP, and Vivante GC7000x2 making it a real beast of a machine, especially compared to the OMAP5432.

Freescale (now NXP) software support is notoriously good as well so I expect a lot of software support for this SOC regarding power states and on-demand core control for maximum power efficiency per the big.LITTLE architecture. I have worked a lot with NXP SOCs and the NXP software support has always been complete and well maintained years after release.
 
Last edited:
I love ARMs graphs. Forget misleading axes, just omit them entirely!

It’s much easier to just make up numbers :)
[doublepost=1566348243,1566346572][/doublepost]Unfortunately there’s no final metrics in here for the 28nm process, but the article I linked does give a really interesting overview of what was done to make improvements to the A72 core from the A57. The A57 itself hosts improvements over the A15.

I think the main thing gained from reading this article is to show people that there is more that can be done to improve power performance than switch manufacturing process. That definitely makes a big difference, but at a big cost, which is that projects like the Pyra will struggle greatly to access this hardware. I don’t think we should write off potential hardware simply because it is not manufactured with <28nm. However, we’ll see how the thermal envelope is of the iMX8 once the documents are released ;)
 
It’s much easier to just make up numbers :)

Especially when you can cherry pick a specific set of conditions for which the 'new' has a special trick to be faster - and not give any detail as to what was actually running for this, likely highly selective to their favor, "same workloads" 'test'. Trust nothing presented as marketing spin. Until someone gets one in a dev board and runs it side by side to a Pyra or OMAP5 dev board with a direct measure of consumption and heat generated, claims like that are pretty much worthless.
 
Agreed. There is also so much more than raw core metrics as it also heavily depends on vendor implementation. Furthermore SOC parameters and available bandwidth have a huge effect as well.

The image posted is merely eye candy. I highly recommend reading the article.

Speaking of which. Do we currently have any measurements for the current Pyra SOC and peripherals?
 
Last edited:
There will be Compute Module 4 at some point.
So it is a matter of time before someone makes a Pyra like device powered by compute module 4.
 
Back
Top