I think you are on safe ground with both claims, based on some cursory research I did to support similar language in the New History of Modern Computing.

ENIAC had a writable electronic memory of 200 decimal digits and could do around 300 multiplications a second – caveats apply pre/post conversion to the modern code paradigm and which version of the setup was being used. (Multiplication time is a better measure of performance than addition time, and also avoids making assumptions about parallelism under ENIAC’s original control mode). That’s why ENIAC quickly became a yardstick to which any other system could be flatteringly compared.

The faster cellphone IIRC is an iPhone 11 based on the Apple A13 Bionic system on chip. According to context-less numbers fetched with a quick Google that does 154.9 GigaFLOPS which is around a thousand times more powerful than the 160 MegaFLOPS of the Cray 1. Now there are probably a lot of footnotes about sustaining peak performance, assumptions about multithreading or vectorization, etc. but when there are three zeros to play with I can’t imagine them overcoming the gap.

The interesting question is how many year you would have to go back in time for the iPhone to be the world’s most powerful supercomputer. I don’t think we did that, but we did try something similar to make the point that graphics cards are where processing power is concentrated these days, to the extent that bitcoin mining and supercomputers are both being built from them. So if anyone who knows more about these things would like to fact-check the following paragraph I’d appreciate it:

A desktop computer with eighteen process cores on a single giant chip brings an impressive degree of parallel processing. Compared to the latest graphics hardware, though, that is barely parallel at all. As of early-2019, Nvidia lists its flagship Titan V graphics card for a price might give pause to even the most dedicated gamer: $3,000. Yet given its 5,750 processor cores it could also be viewed as an extremely affordable alternative to a large building stuffed with mainframes. Nvidia markets the Titan for “deep learning” applications, claiming a throughput that would have made it the world’s most powerful supercomputer just fifteen years earlier. Similar chips drive the world’s fastest supercomputers, including the Titan system that Cray built for Oak Ridge National Laboratory in 2012 and its successor, the Summit computer delivered by IBM in 2018. The later holds 9,216 IBM Power processors and 27,648 Nvidia chips each similar to the one on the Titan V card. 

I’m trying to hedge bets a little with “claiming a throughput.” And of course this is out of date already, and will be further out of date next year when the book appears. So we should at least switch to the past tense.

Best wishes,

Tom

From: Members <members-bounces@lists.sigcis.org> On Behalf Of PETER ECKSTEIN
Sent: Wednesday, August 26, 2020 12:44 PM
To: SIG Computer <members@sigcis.org>
Subject: [SIGCIS-Members] Smart phones vs. classic computers

After many years, I am completing work on the manuscript of book on the early lives of America’s computer hardware pioneers. To put things in a bit of historical perspective, I have included in the conclusions a reference to today’s smart phones as having a “processing capacity that dwarfs that of ENIAC or even the fastest of Seymour Cray’s supercomputers.” (Mauchly, Eckert, and Cray are three of my subjects)

Before I try to pass that along as fact, I would appreciate some expert opinion. Is it true? Is “processing capacity” the best measure of the contrast? Wouldn’t it also be true of storage capacity?

Peter Eckstein

Ann Arbor Michigan