Particle Argon vs NASA IBM 7094

I recently watched the film “Hidden Figures” and can recommend it. One part of it was NASA getting their first electrical computer, rather than having banks of people working the figures by hand. This IBM 7094 was a big thing and give more raw calculation power to the US in the space race. It was quite a large unit - or more accurately series of units. This was instrumental in getting Americans in space and the build up to the first moon handing.

For fun, I was wondering how it compares to the Particle Argon that sits on my desk and can be hidden by two fingers.

There is a wikipedia page on it, which says the earlier 7090 could manage 100 kflops, and the 7094 which was installed could do almost double that, so lets assume 200 kflops. For memory it says it used 36-bit words and had 32,768 addressable memory locations. Assuming this was maxed out, I make that 147kb of memory (36*32,768 / 8).

The Argon 256Mb RAM, so it has almost double the amount of memory.

Any idea about the performance of an Argon in terms of flops? I have been trying to find data and struggling. There is the Cortex M4 data sheet:


I think this is suggesting 1-3 floating point operations per cycle. With a 64Mhz processor, say we assume 2 ops per cycle, is that 2*64 = 128 Mflops?

Cost wise the comparison is easy. Going back to the wikipedia article, the IBM was $2.9m, or about $20m with current inflation. The Argon is $30. I’ve found postage to the UK to be expensive, but probably not as expensive as a room sized 7094 would be with UPS!


It’s the other way around. Cycles per op, not ops per cycle. So more like 32 Mflops. In those bad old days, flops were benchmarked with the Whetstone benchmark:

Give it a try on the Argon. It would be interesting. In any case, we are seeing the effects of 60 years of Moore’s Law. My first professional job was developing a “super computer” that did 64 Mflops, cost $25 M, and filled a basketball court.

– Doug