----
"...So when, exactly, do you anticipate the processing power to deliver 24fps of a 200MP image in a body small enough to wield in the field?..."
----
===
I should preface the following statement in that
I should disclose I am one of the few people on
Earth who have access to 7 nanometre electron beam
etching machines used for prototype CPU production
...AND... that our company creates 64-bit, 128-bit
and ultra-wide-word 256-bit CPU/GPU designs
from the ground up for aerospace, supercomputer
and "other" uses!
So I can ABSOLUTELY SAY the designs for a
combined RISC (Reduced Instruction Set Computer)
and GPU (Graphics Processing Unit) bonded
to a large CMOS sensor of 16-bits-per-channel
and more are HERE AT THIS VERY MOMENT if
Canon wants them! We'll even throw in the in-between
liquid microchannel-based cooling system layers to
ensure adequate cooling.
The die is usually 2/3", one-inch, 35mm or even 65mm
because we bond the gpu/risc cpu processing system
right to the back substrate of the CMOS sensor along
with a multi-layer active cooling system. Our usual
temperature ranges are space-rated in an almost total vacuum
environment from way-below-zero C up to +350 Celcius
so we KNOW it can handle harsh environments!
I know it kinda not fair to bring all this high-tech
up here on this board this up ...BUT....since we
do it ourselves, then it means YOU CAN BET
that Canon is doing it too!
In terms of processing power, the math is as follows
( 8192 by 6306 pixels ) x 8 bytes per pixel (16-bits
per RGB colour channel plus 16 bits for alpha channel
or a distance/depth-map channel = 51,658,752 bytes per frame
x 60 fps = 3,099,525,120 bytes per second RAW transfer rate.
Our chips can do 16 gigabytes per second RAW
so 3.1 gigs per second is PEANUTS for our gear!
So at 8 milliseconds per frame for ADC sampling
and another 8 milliseconds for compression of a video frame
to almost any video CODEC algorithm, a single 256 megabyte
L3 cache is more than enough to handle compression
to Motion JPEG or Wavelet using a four-frame video buffer
for 60 fps video output!
You do need to change the CMOS photo-site current drain
from line-by-line to a 16x16 pixel 2D-XY block which is
WHY we bond the processor to the BACK of the sensor
which allows us to divide the CPU/GPU into precisely
clocked stream processors that ONLY handle the
incoming signals from a single 16x16 photosite
block for ADC (Analog-to-Digital Conversion at
24-bits per channel downsampled to 16 bits)
and further image processing! We 3D LAYER
the stream processors since the ADC and
compression/stream processor circuitry
is quite a bit larger than any single
16x16 block of photosites.
The massively parallel architecture is typical
of what NVIDIA and AMD do on the GPU cards
of desktop PC's and laptops with the only
difference being our on-board VRAM caches,
which are HUGE!
Since we are dealing with LIMITED NUMBER of
graphics and video compression processes that
only need to be timed down to a latency of
8 milliseconds per frame, it's a TRIVIAL
engineering task!
So I am more than just sure Canon
is doing what we do simply because
WE are doing it RIGHT NOW!
Now to get the cost to below 16 000 euros
just might be a bit of an elevated effort!
"...So when, exactly, do you anticipate the processing power to deliver 24fps of a 200MP image in a body small enough to wield in the field?..."
----
===
I should preface the following statement in that
I should disclose I am one of the few people on
Earth who have access to 7 nanometre electron beam
etching machines used for prototype CPU production
...AND... that our company creates 64-bit, 128-bit
and ultra-wide-word 256-bit CPU/GPU designs
from the ground up for aerospace, supercomputer
and "other" uses!
So I can ABSOLUTELY SAY the designs for a
combined RISC (Reduced Instruction Set Computer)
and GPU (Graphics Processing Unit) bonded
to a large CMOS sensor of 16-bits-per-channel
and more are HERE AT THIS VERY MOMENT if
Canon wants them! We'll even throw in the in-between
liquid microchannel-based cooling system layers to
ensure adequate cooling.
The die is usually 2/3", one-inch, 35mm or even 65mm
because we bond the gpu/risc cpu processing system
right to the back substrate of the CMOS sensor along
with a multi-layer active cooling system. Our usual
temperature ranges are space-rated in an almost total vacuum
environment from way-below-zero C up to +350 Celcius
so we KNOW it can handle harsh environments!
I know it kinda not fair to bring all this high-tech
up here on this board this up ...BUT....since we
do it ourselves, then it means YOU CAN BET
that Canon is doing it too!
In terms of processing power, the math is as follows
( 8192 by 6306 pixels ) x 8 bytes per pixel (16-bits
per RGB colour channel plus 16 bits for alpha channel
or a distance/depth-map channel = 51,658,752 bytes per frame
x 60 fps = 3,099,525,120 bytes per second RAW transfer rate.
Our chips can do 16 gigabytes per second RAW
so 3.1 gigs per second is PEANUTS for our gear!
So at 8 milliseconds per frame for ADC sampling
and another 8 milliseconds for compression of a video frame
to almost any video CODEC algorithm, a single 256 megabyte
L3 cache is more than enough to handle compression
to Motion JPEG or Wavelet using a four-frame video buffer
for 60 fps video output!
You do need to change the CMOS photo-site current drain
from line-by-line to a 16x16 pixel 2D-XY block which is
WHY we bond the processor to the BACK of the sensor
which allows us to divide the CPU/GPU into precisely
clocked stream processors that ONLY handle the
incoming signals from a single 16x16 photosite
block for ADC (Analog-to-Digital Conversion at
24-bits per channel downsampled to 16 bits)
and further image processing! We 3D LAYER
the stream processors since the ADC and
compression/stream processor circuitry
is quite a bit larger than any single
16x16 block of photosites.
The massively parallel architecture is typical
of what NVIDIA and AMD do on the GPU cards
of desktop PC's and laptops with the only
difference being our on-board VRAM caches,
which are HUGE!
Since we are dealing with LIMITED NUMBER of
graphics and video compression processes that
only need to be timed down to a latency of
8 milliseconds per frame, it's a TRIVIAL
engineering task!
So I am more than just sure Canon
is doing what we do simply because
WE are doing it RIGHT NOW!
Now to get the cost to below 16 000 euros
just might be a bit of an elevated effort!
Upvote
0