Chinese ACCEL electro-optical AI chip is thousands of times faster than NVIDIA computing accelerators

And it is also much more energy efficient than them

Tsinghua University in China described the principle of creating and operating a super-efficient and extremely energy-efficient artificial intelligence accelerator ACCEL, tailored for tasks related to computer vision

The all-analog chip architecture combines traditional electronic and optical computing to deliver enormously greater performance than the NVIDIA A100 while consuming 4 million times less power.
Experts believe that the creation of ACCEL should not be seen as increasing market fragmentation. Rather, it brings us closer to the future of heterogeneous computing, while simultaneously steering the direction of semiconductor development towards the creation of specialized devices to meet specific needs, as opposed to universal solutions.
ACCEL is capable of performing up to 4.6 trillion operations per second in computer vision tasks, outperforming the NVIDIA A100 by 30 thousand times. The so-called system energy efficiency is expressed in the ability to produce up to 74.8 quadrillion operations per second using 1 W of power. The flagship NVIDIA H100 is not even close to these figures with its 80 billion transistor budget.
Since resource-intensive operations are performed using photons, the accelerator can be built using a 180-nm process technology. And further reduction of standards opens the way to increasing productivity.
#etf #chip #BTC #xrp #sbf
$XRP $ETH $BTC