Hacker News Reader

Comment by chaboud | Hacker News Reader

“ Lets start with the M1 Mac mini. It’s a nice baseline as it’s the Mac i was developing on up untill recently. The Mac mini completed the tests in 45.13 seconds across 8 cores.

…………

The AWS Graviton instance (c6g.metal to be specific) delivered a score of 14.63 seconds across its 64 cores. Around 68% faster than the Mac mini.”

“Faster” would be a matter of rate.

So if we have times and we know D = R*T, we’d have:

(D/T2)/(D/T1) as the speed up of R2, which is just (T1/T2), or 3.08. If we were to say how much “faster” c6g.metal is than a Mac Mini, wouldn’t we say 208% rather than 68% if the speed of the Mac Mini is our baseline? Am I missing something?

View on HN

We developers can always compensate for our lack of math skills with a very large test suite.

The time was reduced by 68%. Typical confusing factor, lots of people get a "reduction" mixed up with "rate of work". Happens all the time in games..

I would agree with you, 200% faster is more appropriate, or time reduced by 68%.

I'd actually go with "108% faster than the M1", since the "baseline" of the M1 is 100%. Alternatively, "208% of the rate of the M1" since that explicitly includes the baseline.

It’s 308% the rate of the M1. I already subtracted out the 100% for baseline to get to 208%.

Percentage increases will never not confuse people. Just like double negatives.

Just never use percentages to show ratio. Graviton instance is approximately 3 times faster.

I completely agree. X times the rate of Y is unambiguous.

I highly recommend this blog post, which makes the same point among many (great) others: https://sled.rs/perf.html

> When we speak about comparative metrics, it is also important to avoid saying commonly misunderstood things like “workload A is 15% slower than workload B”. Instead of saying “faster” it is helpful to speak in terms of latency or throughput, because both may be used to describe “speed” but they are in direct opposition to each other. Speaking in terms of relative percentages is often misleading. What does A (90) is 10% lower than B (100) mean if we don’t know their actual values? Many people would think that B is 1.1 * A, but in this case, 1.1 * 90 = 99. It is generally better to describe comparative measurements in terms of ratios rather than relative percentages.

> The phrase workload A is 20% slower than workload B can be more clearly stated as workload A was measured to have a throughput of 4:5 that of workload B. Even though many people will see that and immediately translate it to “80%” in their heads, the chances of improperly reasoning about the difference are lower.

I would say 2 times faster or 3 times the speed of the mac mini. Faster implies a delta already; I can say X is 0.5x faster than Y and it means its speed is 1.5x that if Y.

In common usage I believe 3 times faster is what most people understand for these figures.

In common usage people refer to their mass as weight. Nobody says "I mass 75 kilos".

I'm divided if that's an argument in favor or against your point, though.

Is there a language where people don't use weight for mass?

Terrific question!

https://linguistics.stackexchange.com/questions/44134/is-the...

Common usage doesn’t mean it’s correct or clear. Yes, a lot of people say 2 times faster but it isn’t. The whole point of this comment thread is just that.

You are definitely wrong. 1 times faster is not even a thing.

Very bold statement without a reference. I’d say 1 time faster, but I usually say it’s 2x the speed to prevent confusion.

In the mini each core takes 45.13 to complete the work of 8 (64/8) cores of the Graviton.

So if you devide the "sequential" time by 8, the mini core is actually faster as it does the work of 1 Graviton Core in 1/8 of the 45.13sec time:

5.64 sec for the Mini vs 14.63 sec of the Graviton for one unit of work (which is total/64)

You can't really deduce anything without knowing more about the test suite: How much network and file system does it do? How well can it use the full core count?

This guy benchmarks.

From the article:

> The test suite is embaressingly “parallel”. Each test can run on its own core so the more cores we have the faster the tests will run.

>Am I missing something?

You are not. It is very common mistakes.

Faster is used both in the context of time spend and rate. In sports for example we say someone was x seconds faster across the course all the time. There really is no definition that ties faster to only be associated with rate, so I would say this is just pointless nickpicking. It's like complaining that people use higher/lower when talking about price

Reply on news.ycombinator.com