[ddtaylor]

Can we just go back to numbers where higher is always better?

[rincebrain]

Well, if we have 32 cores at 2 GHz, and 48 cores at 1.5 GHz, from the same chip architecture, which is better?

For some workloads, it's one; for others, it's the other.

A one-dimensional performance axis is hard in practice.

[atmavatar]

For some reason, that comparison reminds me of:

"If you were plowing a field, which would you rather use? Two strong oxen or 1024 chickens?"

-Seymour Cray

[MBCook]

You could use a synthetic benchmark. Create your own Geekbench like thing and use those numbers.

Intel-$singlecore-$multicore

So an Intel-6-12 had twice the single core performance of the Intel-3-48 but is much worse in multicore.

Yeah, things are WAY more complicated. But at least it gives you a chance without memorizing spec sheets first. Higher numbers are always better. No redefining. This year’s 5 can’t be better than last year’s 20 in performance.

I don’t follow Intel/AMD chips anymore. If I wanted to buy a computer I’d have to learn it all from scratch. This would give me a nice leg up.

[ddtaylor]

For the most part the transistor count (or something akin to it) is usually increasing, even if the die space is larger.

[bubblethink]

>Intel-$singlecore-$multicore

There will be power specs, and igpu specs, and ecc, and so on. It's hard to capture succinctly.

[Dylan16807]

Gpu can be a third number. Power doesn't need to be in the model number. Neither does ecc, also all cpus should have such an easy to implement feature.

[opportune]

You can’t preserve ordering mapping permutations of a high-dimensional space (specs: clock rate, power, cores, cache, pipelining/vectorization, etc) with another high dimensional space (benchmarks, workloads, and scenarios) to a single number. It’s possible even for a chip with superior numerical specs across the board from another chip to perform worse due to lack of support for specialized instructions or differences in thermals.

[ddtaylor]

Almost every other industry makes some kind of attempt to do exactly that because consumers want simple choices.