Interesting hardware factiod about the Raspberry Pi

[jimrh]

Wandering around on the web, I found an interesting Wikipedia article on superscaler processors, which lead me to an article on vector processors.

Within the article on vector processors there are a couple of interesting mentions of the VideoCore architecture of the Raspberry Pi:

Interestingly, though, Broadcom included space in all Vector operations of the Videocore IV ISA for a REP field, but unlike the STAR-100 which uses memory for its repeats, the Videocore IV repeats are on all operations including arithmetic vector operations. The repeat length can be a small range of power of two or sourced from one of the scalar registers.

and

In addition, GPUs such as the Broadcom Videocore IV and other external vector processors like the NEC SX-Aurora TSUBASA may use fewer vector units than the width implies: instead of having 64 units for a 64-number-wide register, the hardware might instead do a pipelined loop over 16 units for a hybrid approach. The Broadcom Videocore IV is also capable of this hybrid approach: nominally stating that its SIMD QPU Engine supports 16-long FP array operations in its instructions, it actually does them 4 at a time, as (another) form of “threads”.

The Videocore architecture is described more fully here.

VideoCore

Based on these readings, and the little I know about the Raspberry Pi’s architecture, the current crop of Raspberry Pi operating systems woefully under-use the video capabilities - primarily because Broadcomm won’t give the Raspberry Pi Foundation the necessary information.

So, not only is the Raspberry Pi capable of virtualization, but it also has a number of complex vector graphics pipelines capable of doing interesting things!

Interesting!

[jimrh]

P.S.

If we had better information about the video and vector capabilities of the Pi, it could be a BUTT KICKING object recognition and AI platform.

Sigh. . . .

[cyclicalobsessive]

@jimrh

Raspberry Pi capable of virtualization

I’m not sure I understand what you are saying with that. Docker runs on the RPi, and aren’t there RPi VMs that run on Docker on other architectures?

https://linuxconfig.org/how-to-run-the-raspberry-pi-os-in-a-virtual-machine-with-qemu-and-kvm

[cyclicalobsessive]

@jimrh

If we had better information about the video and vector capabilities of the Pi,

And Broadcom wouldn’t agree to the ridiculously low price the Raspberry Pi Foundation wanted. Same concept of Tesla vehicle capability varying with a what you are willing to pay?

[jimrh]

@cyclicalobsessive

@jimrh

Raspberry Pi capable of virtualization

I’m not sure I understand what you are saying with that. Docker runs on the RPi, and aren’t there RPi VMs that run on Docker on other architectures?

I was referencing an earlier post where I saw something like “virtual instructions available on [4] cores” in the boot log and I mentioned that this means the Pi can run a virtual environment.

Others mentioned that they had heard of things like QEMM running on the Pi.

This article mentioned additional, (potentially hidden), capabilities involving vector mathematics, vector processing, (can you say “Cray Supercomputers”?), and other seriously advanced stuff on a board smaller than a lot of smartphones. Of course, nobody is comparing the Pi to a honkin’ beast like a Cray, (except this article), and the vector capabilities are significantly scaled down, but it’s interesting that they’re there.

@cyclicalobsessive

@jimrh

If we had better information about the video and vector capabilities of the Pi,

And Broadcom wouldn’t agree to the ridiculously low price the Raspberry Pi Foundation wanted.

Interesting!

The story I read by one of the original designers was that he worked for Broadcomm at the time, knew their SOC’s kicked butt, and wanted them for the Pi, because it would add tremendous capabilities and make the design so much easier.

Broadcomm, on the other hand, had serious questions about a group who thought they could “out Arduino” the Arduino, and make something the same size that was a “real computer” and not just a maker’s tinker-toy.

They, (begrudgingly), allowed the Raspberry Pi Foundation to begin development with their chips, but were not holding their breath.

When the RPF asked for what amounted to “the keys to the castle”, (detailed VPU information), that’s where they drew the line, though they did provide binary blobs that implemented a small subset of the chip’s video capabilities.

Now that the Raspberry Pi is officially “a thing” Broadcomm is being a tad more forthcoming, but really doesn’t want to give out too much information.

IMHO information about the API’s and such, even without detailed architectural information, would be valuable.

As a point of fact, NVIDIA has been making money hand-over-fist since they decided to allow third parties access to API and architectural details of their CUDA processing and rendering pipelines. People are buying horrendously expensive video cards, (plural), just to exploit the advanced math and array handling NVIDIA makes available.

There is even a whole series of “video cards without video” that only implement the GPU - in some cases multiple GPUs on one card that people are using in local systems to replace expensive processor time on supercomputers.

Broadcomm would do well to pay attention to NVIDIA’s business model.

[cyclicalobsessive]

@jimrh

Pi can run a virtual environment

In fact, I believe I am headed there to access the PiCamera from the 64-bit Ubuntu.

The PiCamera MMAL library only runs in 32-bit raspbian (as of this date), so some guy created a Docker VM with 32bit Raspbian that loops reading the PiCamera device and reposts the frames on a socket for the 64-bit Ubuntu to use. Crazy that the RPi needs to virtualize itself, but “where there is a will, there is a way” (for folks smarter than me).

[jimrh]

@cyclicalobsessive

“where there is a will, there is a way”

And I always say “where there’s a will, there’s usually a probate lawyer making a ton of money!”