Benchmark rebasing!

I noticed Chris Hall’s latest benchmarks for the Pi 400 has figures in it like 42941% the numbers are getting silly.
How about rebasing the benchmarks on an Iyonix?
Whilst a RiscPC (the other possible base candidate) is/was more common there are many versions (610,710,StrongArm and 25MHz, 33MHz bus) an Iyonix was only one speed, also by being a bit faster it will give a bit more headroom for the future.

Unfortunately, the Iyonix suffers from horrible memory bandwidth. It’s not exactly a good all-rounder.

I agree, the Iyonix is quite a slow machine going by the numbers, it feels faster than it is thanks to the graphics acceleration and fairly good disc speed (at least with a 7200rpm drive).

The Raspberry Pi 1 probably makes more sense as a baseline.

Whilst the Iyonix may not be ideal, it is I think the best option, it is lot easier for most people to compare with as many/most people will have had experience of it and as a main machine. Whilst I suspect even more people will have used a Pi 1 very few will have used it as their main machine and long enough to remember what it felt like. Many/most Pi 1 users will have progressed to Pi2 or Pi3 after not long.

I’ve never even seen an Iyonix… I was using StrongARM Risc PCs until I got a Pi, but they hadn’t been my main machines for years by then, and the Pi still isn’t.

How many people really use a RISCOS beastie as their main machine??

If you are going by that argument, the Risc PC remains the machine most people (including those coming back to play with RISC OS after a while away) will be familiar with by an enormous margin. The question as you say is which one, I’d be inclined to go for the standard 233MHz StrongARM.

The problem is that the various machines have very different performance profiles depending on the details.

A plain StrongARM Risc PC with an original ADFS-connected Conner HD will be very different from one with a ViewFinder, a Kinetic, a fast SCSI podule, a fast IDE podule. Or all of those.

Benchmarking is hard. Real-world (i.e. real applications, real use cases) or synthetic? What is a “real” use case? I tried to do a collection of benchmarks back in the IYONX vs. Omega vs. Risc PC days, and it was very difficult. See here: http://legacy.huber-net.de/iyonixvsomega/bench_e.htm

Of course, a lot of people mailed me to argue for or against some of those tests I picked. It is an endless discussion. The only common ground was basically that Aemulor is really a terrific piece of software.

I also found out that it is very difficult to actually remember the “feel” of machines if you begin to mix different software and hardware from different times. I know that I was entirely happy to work with a Turbo-ARM2 A3000 for a long time, happily switching between various screen modes to overcome performance problems due to Video DMA. When I went back there a few years ago with a selection of software that I mainly used on a StrongARM Risc PC later on (like !Browse or Impression Style), I was shocked by the glacial speed of the A3000. Not at all the speed I seemed to remember.

…may be too much to ask, but as you can see e.g. on that german webpage on some CPU reviews, you can go with your mouse over a cpu of your choice, that one become 100% and all others are related to that one…so all wished from here would be satisfied…but I got no clue what kind of web programming voodoo you need for that.

Benchmarking is hard. Real-world (i.e. real applications, real use cases) or synthetic? What is a “real” use case?

Spot on.

very difficult to actually remember the “feel” of machines if you begin to mix different software and hardware from different times

Exactly.