Talk:Computer performance by orders of magnitude

Discussion

I know this page needs help in coming together. Please donate your time in improving citations and filling in information!

10²⁴ and above

The statements here, like "Strong AI in this computing era will probably drive scientific innovation" and "It will be at least of the level of human intelligence" seem overly bombastic given the ammount of uncertainty and debate around strong AI, not to mention the inherent uncertainty in speculating on things so far into the future. I have moderated the wording a bit and added a "dubious" tag until someone can come up with more sources to back up the statements or rewrite it.

Knuthove 02:57, November 2, 2010 (UTC) —Preceding undated comment added 02:57, 2 November 2010 (UTC).

References

A lot of the stuff on here contains no references, this is vital especially for the dubious assertions about human brain power. Some of this article reads like Kurzweil fanboy Copy and Pastes. Veritasvoswiki (talk) 16:29, 9 March 2011 (UTC)

Alot of this was derived from his work, but its also easily deducible. I'm actually a pretty good study on reaction times, I've won a number of trophies in drag racing. The usaul reaction time is no better then .001s and +/-.003. So you can expect serial processing in the brain can't be over 1000hz. Further visual perception is less then 60hz, because the light-bulbs in the U.S. don't flicker.

The addition and multiplication examples were just looked up. The overall parallel throughput of the brain is theoretical.

The last level of computing represents the ability to realistically simulate all the chemical, cellular, molecular, protein, and other actions of the brain to an extent that human intelligence evolves out of simulation of the real brain instead of programming. So while we may never be able to design our own strong AI, we can mimic a human brain in every way imaginable. —Preceding unsigned comment added by 74.129.199.207 (talk) 04:48, 7 May 2011 (UTC)

“light bulbs in the US do not flicker to the human observer” —article

“Visual perception is less then 60hz, because the light-bulbs in the U.S. don't flicker” —talk

Oops, that's not a good source for the claim. First, 60 Hz AC electrical power drives two peaks of power per cycle, one in each direction, which would make a fast-reacting bulb flicker at 120 Hz, not 60 Hz. Second, even at 120 Hz the sinusoidal power waveform doesn't abruptly cut off power after each peak, but maintains some lesser power during almost the entire cycle, making the variation gradual and continuous, and much more difficult to perceive. Third, traditional incandescent lightbulbs are not fast-reacting, and the phosphor coatings inside florescent tubes are selected so that they, too, are not fast-reacting, so that even the gradual variation in driving power is further smoothed towards a continuous level of light. So the fact that people don't notice flickering lightbulbs really doesn't say much about the speed of visual perception. 76.100.23.153 (talk) 04:40, 19 November 2012 (UTC)

The reaction time of the fastest nerve propagation in the brain is recorded at about 268 MPH. The slowest occurs at about 1 MPH. That gives an average of roughly 134 mph. Assuming it goes to mid brain and back as in this type of activity is hardwired (a bright flash?), and the average skull is 20 cm, the fastest observable objects possible is .0032s which is 312 hz. The fastest time to reach the brain and be processed was actually 220hz per government test, and that is a one way delivery. In the airforce test they didn't have to react just see it. This points very strongly that reaction times are unlikely to exceed 100hz.

TheTrueChAoS (talk) 17:52, 19 February 2013 (UTC)

References

Symbolic Processing in the Brain

Talking about the processing power of the brain is stepping in quicksand. The "graphics processing subsystem" that presents us with a clear image based on raw input is extremely powerful and hasn't yet been fully simulated in a computer, so we can't even use that as a yardstick. The aural processing subsystem is probably almost as complex. But a key difference between human consciousness and the way digital computers work is that the brain thinks in symbols - layers of abstraction above raw data. For example: catching a baseball involves real time processing from the eyes, calculation of the trajectory from that data and motor neuron commands to place the hand in the right place at the right time. Forcing the brain to do simple arithmetic (the forte of a digital computer) is not a simple thing - the brain must abstract the concepts of number, build a virtual symbolic computing engine that simulates the activities of an Arithmetic Processing Unit with stacks, buffers, lookup tables, etc. and then force data through it. The result in real time is as though this virtual engine has a processing speed of something on the order of 100 bits/second - which is not the point (though some fool might claim that this proves how stupid people are). We may manipulate symbolic abstractions slowly but give real time outputs to data that is directly processed. Which is more important? In a survival sense real time processing is paramount. In the appreciation of a theoretical sense symbolic processing comes out ahead.

Mccainre (talk) — Preceding undated comment added 16:40, 18 August 2014 (UTC)

What’s the unit??

It’s not FLOPS, from what I can tell, is it…?

This list is pointless, without a unit! — Preceding unsigned comment added by 85.197.40.14 (talk) 16:49, 19 March 2014 (UTC)

74.98.216.253 (talk) 17:20, 14 January 2017 (UTC) The unit is inverse seconds (hertz) because the quantity being measured is frequency.

Discussion of the scope of the subject matter in the pre-amble

I feel there are too many oversimplified choices made here as to theoreticaly calculations of human-brain capacity. Specifically ref. Ray Kurzweil's speculative calculation dated 1999 for "roughly the hardware-equivalent of the human brain" in the "Exascale" section, and also ref. the focus on the 1997-dated theoretical Matrioshka brain called out in the "Post zettascale" section.

Modern supercomputers exist, but they are massively-parallel. Lots of these epic (not EPYC!) chips working together.

Please consider bolstering this page with two graphs - one graph showing single-chip performance over time (FLOPs works I guess.. - but memory retrieval time, cache levels and new algorithms, core counts, fast I/O channels - these and I'm sure many more measures, that don't directly pertain to the FLOPs count, should be captured too, maybe in a separate graph).
And please add the latest graphics cards then in yet another graph... Graphics cards that pair well with modern AMD/Intel chips and implement vastly-parallel, e.g. n-dimensional matrix multiplication compute capability, over a fast bus.
Hope this makes sense. Paid into the fund for this site a few times, I feel this article could and should be strengthened... YaWha (talk) 01:29, 23 August 2023 (UTC)