https://9to5google.com/2021/10/20/google-tensor-chip/
On 2021-10-29 00:22, RichA wrote:
https://9to5google.com/2021/10/20/google-tensor-chip/
It's actually optimizing general purpose ARM processors for particular machine applications while adding sophisticated modules for machine
learning, graphics, etc. Much as Apple has done with its "A" series
chips (iPhone, iPad) and "M" series (Mac, high end iPad) and other very specialized ARM derivatives.
Missing with the Google processor (as far as I can see) is the unified
memory architecture that Apple uses in its M series. This greatly
improves all functions as there is far, far less moving of data around between functional portions of the computer.
Pretty amazing stuff actually, and it will all be quaintly old in 10 years.
On Friday, 29 October 2021 at 15:54:57 UTC-4, Alan Browne wrote:
On 2021-10-29 00:22, RichA wrote:
https://9to5google.com/2021/10/20/google-tensor-chip/
It's actually optimizing general purpose ARM processors for particular
machine applications while adding sophisticated modules for machine
learning, graphics, etc. Much as Apple has done with its "A" series
chips (iPhone, iPad) and "M" series (Mac, high end iPad) and other very
specialized ARM derivatives.
Missing with the Google processor (as far as I can see) is the unified
memory architecture that Apple uses in its M series. This greatly
improves all functions as there is far, far less moving of data around
between functional portions of the computer.
Pretty amazing stuff actually, and it will all be quaintly old in 10 years. >>
Things haven't changed that much. Silicon chips, conventional storage, battery power, only just narrower trace widths. All the wonderous, long-predicted advances like optical or quantum processing have not come to pass.
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 285 |
Nodes: | 16 (2 / 14) |
Uptime: | 76:07:44 |
Calls: | 6,489 |
Calls today: | 2 |
Files: | 12,096 |
Messages: | 5,276,210 |