• Welcome to the new Internet Infidels Discussion Board, formerly Talk Freethought.

Apple's upcoming CPU-architecture change: ARM-based "Apple Silicon"

Processors down to the small comptrollers have long gone to ARM.
 

The Apple Neural Engine contains several neural-engine cores and a "planar engine".

The neural-engine cores do convolution. This is multiplication repeated for each data point, multiplication done for that point and its neighbors with the amounts specified as the kernel. Mathematically, for input x, kernel K, and output y,

\( \displaystyle{ y(i) = \sum_j K(j) x(i-j) } \)

Convolution output is then fed into a "planar engine". It can be run in pooling or size-reduction mode, breaking up the input into tiles then finding a single value for each tile, or it can be run in elementwise mode, doing operations on individual members.

This is a kind of tensor processing unit, named after a common term for multidimensional data, data indexed with multiple values.

This kind of unit is good for multimedia operations, where one needs to do the same thing at each individual location in the data.
  • 0-tensor - scalar - isolated value
  • 1-tensor - vector - list - mono audio data (time)
  • 2-tensor - stereo audio data (time, channel), grayscale image data (vertical, horizontal)
  • 3-tensor - color image data (vertical, horizontal, channel), grayscale video data (time, vertical, horizontal)
  • 4-tensor - color video data (time, vertical, horizontal, channel)
 

The M series of chips is a series of  System on a chip designs. They have multicore CPU's and built-in GPU's (Graphics Processing Units) - so one can do 3D game graphics. They also have image signal processors, encoders and decoders for common video-data formats,  PCI Express for persistent memory, and USB and Thunderbolt support. They also have a "Secure Enclave", a core intended for highly-secure operations like Face ID.

The GPU includes "dynamic caching", using only as much memory as it needs, "mesh shading", and  Ray tracing (graphics)

Ray tracing works by reversing the path of a ray of light from the source to the scene to the eye or camera.

What's a mesh shader? What are shaders? A shader is a mini-program or function that is run by a GPU to do various things. A fragment shader calculates pixel values. A vertex shader calculates polygon-vertex positions. A geometry shader can do things like tessellations, filling out a curve or a surface with points in between. A compute shader is generalized. Mesh shaders are the next step, incorporating vertex and geometry shaders.

What I have been describing is the M-series  Die (integrated circuit) or bare chip. To be usable, chips must be packaged, and the M-series package includes RAM chips along with an M-series chip. That means that these chips' RAM cannot be upgraded, though it may be possible to add external RAM that works like SSD memory.
 
Will the PeeCee world catch up to Apple?

2024 could be the year the PC finally dumps x86 for Arm, all thanks to Windows 12 and Qualcomm's new chip | PC Gamer - "It's been a very, very long time coming, but it looks like it's finally happening."
We've already reported on Qualcomm's new 12-core Arm uberchip, the Snapdragon X Elite, and its claims of x86-beating performance and efficiency. But it takes two to tango when it comes a major transition like moving from x86 CPUs to Arm chips. You don't just need hardware, you need software, too.

And that, dear PC fans, is where Windows 12 supposedly comes in. Reports indicate that Microsoft is planning to add specific support for Snapdragon X Elite in future builds of Windows.

...
A decade or so ago, the assumption was that Arm was really only suitable for low power applications. You needed x86 for high performance. But Apple's 'A' and 'M' chips have proved that wrong. Apple's Arm cores now have significantly higher performance per clock cycle than any traditional x86 PC processor.

For sure, the top x86 chips still give more outright CPU performance thanks to higher clocks and core counter. But Apple has proven that Arm can compete and then some for fundamental number crunching grunt.
Click to expand...
 
lpetrich said:
steve_bank said:
Processors down to the small comptrollers have long gone to ARM.
Click to expand...
I'm confused. :(
Click to expand...
ARM licenses processor architecture as intellectual property. Processors are designed in high level software like VHDL.

ihttps://en.wikipedia.org/wiki/VHDL#:~:text=The%20VHSIC%20Hardware%20Description%20Language,%2C%20documentation%2C%20and%20verification%20purposes.

A company like Microchip used to design their own processors, but it is cheaper to buy a common architecture and add peripheral sections.


ARM architecture family - Wikipedia

en.wikipedia.org en.wikipedia.org
www.ti.com

Arm-based processor design & development | TI.com

Simplify your design process and accelerate time to market with TI’s embedded hardware and software development resources for Arm®-based processors
www.ti.com

Free C compilers and an IDE. Connect with USB.If you want to play around with it Microchip has ceap demo board.
 
Fixed link:  VHDL
The VHSIC Hardware Description Language (VHDL) is a hardware description language (HDL) that can model the behavior and structure of digital systems at multiple levels of abstraction, ranging from the system level down to that of logic gates, for design entry, documentation, and verification purposes. Since 1987, VHDL has been standardized by the Institute of Electrical and Electronics Engineers (IEEE) as IEEE Std 1076; the latest version of which is IEEE Std 1076-2019. To model analog and mixed-signal systems, an IEEE-standardized HDL based on VHDL called VHDL-AMS (officially IEEE 1076.1) has been developed.

VHDL is named after the United States Department of Defense program that created it, the Very High Speed Integrated Circuits Program (VHSIC). In the early 1980s, the VHSIC Program sought a new HDL for use in the design of the integrated circuits it aimed to develop. The product of this effort was VHDL Version 7.2, released in 1985. The effort to standardize it as an IEEE standard began in the following year.
Click to expand...
 
There are sets of software tools that go from an HDL language to manufacturing the chip.

It is all very routine now.
 
You must log in or register to reply here.
Back
Top Bottom