2020 BRN Discussion, page-25251

I don't even link my own posts for later reference because as I have said before I am blessed or curse with a memory for narrative.

I disagree entirely with your assessment of how LDA Capital valued Brainchip and the valuation you state for Graphcore is what it states is the conclusion to be drawn from the amount invested but we are comparing eggs in England with neurons in Australia so we can agree to disagree.

However I found what I was actually looking for in the technical data on Graphcore's website as they do not put up front their power consumption figures so you have to dig deep:

"5.2. CONVOLUTION 75
This is a common occurrence among massively parallel computing platform
and is of little consequence in most cases. In our benchmark using optimized
tile mapping, performance saturates around 18 and 55 ... 58 TFlops/s in single
and mixed precision, respectively.
We also compare the devices in terms of energy efficiency by charting
throughput values divided by nominal power absorbed by each board in Figure 5.3. Results are expressed in units of TFlops/s per Watt (TFlops/s/W).
For the V100 GPU, we use nominal power, 250 W.
In both single and mixed precision cases, a single IPU delivers higher efficiency than two IPUs and higher efficiency than the V100 GPU."

Their most power efficient V100 GPU uses only nominal power of 250 Watts.

YES YOU READ THAT CORRECTLY ITS MOST POWER EFFICIENT V100 GPU USES ONLY A NOMINAL 250 WATTS.

So what you are doing in comparing Graphcore with Brainchip is illogical the only way Graphcore can win is if the competition was which one can drain a car battery in 30 minutes or less because AKD1000 using its milli watts to micro watts would never exhaust a car battery while Graphcore would drain it to a point where it could never accept another charge.

Graphcore seems like a good business with good leaders but its like comparing a telescope with a submarine they are entirely different species.

My opinion only DYOR.