Future of the CPU board?

[gpsqueeek]

We do? When's the release date?

In two months, as usual ^^

 

[Klumpen]

In two months, as usual ^^

I wish I could "like" this...

Well, hopefully it's possible to stream Kanopy with it via HDMI, since my PC's cooling is too loud and that would be one of the purposes I'd have for it that doesn't have been taken by other devices over the years since my pre-order.

 

[ToastBucket]

The problem with the iMX8 is that it uses A53 cores, which are actually slower than the A15 ones. As it has more cores it can compensate for it in multi-threaded workloads, but it will remain slower in single threaded ones.

However i would like to know what the general thoughts are on the Amlogic S922X, it has 4 A73 + 2 A53 cores and a mali G52 MP4 gpu.

The iMX8 actually uses A72s and A53s. Looks like it’s ARMs new big.LITTLE architecture which utilizes high performance cores in times of need and high efficiency cores in other scenarios.

That Amlogic seems interesting. Any idea what availability is like and how the BSP support is? I’m a pretty strong supporter of Freescale/NXP SoCs since their board support is so good (I do BSPs for a living and just love when a customer is working with an iMX).

Also curious what thoughts are on the Kirin 960. I highly doubt ED could get any but still a pretty interesting part. It’s used on Androids official reference platform, so it’s what they’re using along with Pixel HW for AOSP development.

https://www.96boards.org/product/hikey960/

 

[spud42]

https://androidpctv.com/comparative-amlogic-s922x/
bit of info here on the SOC... TDP might be too much for the Pyra application. seems to hit 7.7W

 

[Kuro]

Looks like i was wrong, For some reason i thought only the i.MX8 M and X existed. Anyway i don't know much about the details of these SoCs so i have no idea about they compare in TDP, BSP support or graphics performance, but as the Amlogic wasn't being discussed here i was wondering why.

From my point of view it had a good cpu, could be found in some SBCs (like the odroid n2) which means it might not be that hard to get, and had a gpu that has free drivers being developed for.

 

[ToastBucket]

Looks like i was wrong, For some reason i thought only the i.MX8 M and X existed. Anyway i don't know much about the details of these SoCs so i have no idea about they compare in TDP, BSP support or graphics performance, but as the Amlogic wasn't being discussed here i was wondering why.

I was originally talking about the iMX8M in my first post. I should have been talking about the iMX8, which is currently in preproduction which is why you may not know that it existed. I haven’t been able to find much info yet on the TDP but I did get to play with it a bit at an NXP conference a few weeks back and was pretty impressed. Availability is also pretty good for this family of SoCs.

 

[Grench]

https://androidpctv.com/comparative-amlogic-s922x/
bit of info here on the SOC... TDP might be too much for the Pyra application. seems to hit 7.7W

28nm process based CPU - same as the OMAP5432. No 'real' gain over the current CPU.

 

[Kuro]

28nm process based CPU - same as the OMAP5432. No 'real' gain over the current CPU.

The links says the CPU is in 12nm and the GPU in 16nm, where did you see it was 28nm?

 

[hns]

There is another factor to consider: I/O voltages.
Usually, mobile processors have 1.8V for general I/O interfaces (HDMI, MIPI, USB, SDIO have their own voltage level definitions) while industrial processors typically use 3.3V.
The reasons behind are radio emission (1.8V has ca. 70% less switching noise compared to 3.3V), power (charging/discharging wires and inputs to 3.3V needs more energy than 1.8V). 3.3V is more robust in industrial environments against external interference.
The Pyra main board and display board are designed for 1.8V I/O, i.e. mobile processors or industrials with 1.8V capability.
While it is possible to add level shifter chips, they need a lot of space on the processor board.
Some obtainable candidates in mass production that fulfill this requirement (but only this) are Z8550, i.MX8M, R3399.Another consideration is that a good audio chip must also be placed here.
And the final one is that the SoC should be able to handle the right mix and number of interfaces (I2C, SPI, MMC/SD, USB3, GPIOs, HDMI, MIPI).

 

[matzesu]

Maybe whe could use the CPU from this new Mac Pro Apple Anounced on Monday,
Ok just a joke, this CPU is maybe to power hungry, whe would need a Mobile Termo Nuklear Generator for the Pyra then ..

 

[Askarus]

I still vote for the Rockchip 3399

 

[Grench]

The links says the CPU is in 12nm and the GPU in 16nm, where did you see it was 28nm?

From this: https://en.wikipedia.org/wiki/Amlogic
The rest of the Amlogic 900 series is listed as 28nm. Apparently I was incorrect and Wikipedia is lacking information around Amlogic's 12nm process?
[doublepost=1559747586,1559747370][/doublepost]

I still vote for the Rockchip 3399

Which definitely IS built on a 28nm process - which means it's performance per watt is in the same class as the OMAP 5432.

 

[Letalis Sonus]

Apparently I was incorrect and Wikipedia is lacking information around Amlogic's 12nm process?

To quote the linked article:

As we can see the maximum consumption remains in only 7.7W as a result of the jump of manufacturing process of the CPU from 28nm to 12nm, although the GPU remains at 16nm.

 

[levi]

Which definitely IS built on a 28nm process - which means it's performance per watt is in the same class as the OMAP 5432.

Well, to borrow from Intel's terminology, there's still scope for a tock level change. In their (now outdated terminology) a tick was a process shrink, and a tock was a microarchitecture swap. In other words, the A15 is an old chip spec, I'd expect something newer to be normally more efficient at doing the same work even implemented in the same transistor size, although there's always the possibility they're targetting peak performance, and sod the thermals. Compared to the old swap from Netburst to Core at 65nm anything since then and pretty much always in the ARM domain will be less revolutionary, but I'd still expect some measurable change.

 

[Grench]

Well, to borrow from Intel's terminology, there's still scope for a tock level change. In their (now outdated terminology) a tick was a process shrink, and a tock was a microarchitecture swap. In other words, the A15 is an old chip spec, I'd expect something newer to be normally more efficient at doing the same work even implemented in the same transistor size, although there's always the possibility they're targetting peak performance, and sod the thermals. Compared to the old swap from Netburst to Core at 65nm anything since then and pretty much always in the ARM domain will be less revolutionary, but I'd still expect some measurable change.

Yes, one processor might be some margin more efficient than another on the same 28nm process. But, shrinking the process from 28nm to 14 or less is going to be measured in an order of magnitude change. There isn't much point in doing the development work for a 'new' SoC at 28nm - the gains simply are not there to match the effort.

 

[Letalis Sonus]

But, shrinking the process from 28nm to 14 or less is going to be measured in an order of magnitude change.

Depends on what exactly those numbers refer to. The transistor size itself hardly went down after all, there is only so much gain to be had when increasing the density of the circuity. The manufacturing node was not the only new thing, though, new types of transistors were also invented - maybe it was actually worth it to backport these to larger nodes?

There isn't much point in doing the development work for a 'new' SoC at 28nm - the gains simply are not there to match the effort.

Well... Infineon just switched to 40nm with their newest hexacore Aurix chips

 

[asimov-solensan]

I know it may sound tedious but we should bring again to the table the old "using rpi SOC in Pyra" question.

Basically being in the rpi ecosystem would just amazing, but it's SOCs had notably worse performance than an OMAP5. This is why I have been against using a RPI SOC.

But the SOC on RPI4 seems way better the the one in Pyra (please correct me here), so it would be a great improvement to get this SOC as a future upgrade.

Of course there are questions like thermal envelopment, CPU size, available ports ... That I can't answer but would love to hear from someone with more knowledge how feasible would be this BCM2711 on the Pyra.

 

[TrashyMG]

But the SOC on RPI4 seems way better the the one in Pyra (please correct me here), so it would be a great improvement to get this SOC as a future upgrade.

It would be neat to do a collaboration for us to build Pyra module RPi 4 module with some licensing involved. On quick glance it looks promising lot of I/Os likely line up, but I don't know all of the technical limitations. Nikolaus likely would need to chime in.

 

[levi]

Can you buy the RPi4 SOC without going to the faff of buying an RPi4 and unsoldering it's BGA SoC and associated RAM?

 

[TrashyMG]

Can you buy the RPi4 SOC without going to the faff of buying an RPi4 and unsoldering it's BGA SoC and associated RAM?

I think they're custom made for the RPi group...