PostgreSQL Benchmarks: Apple ARM M1 MacBook Pro 2020 | Hacker News Reader

PostgreSQL Benchmarks: Apple ARM M1 MacBook Pro 2020

info.crunchydata.com

32 pointsStreamBright9 comments

Seeing how AMD processors from this year like the 5950X compares would be interesting.

Yeah, my thoughts exactly. We already know that Intel's offerings suck, and have sucked for a while. I get that Apple has exclusively put Intel into their MacBooks and hence that's used for comparison almost everywhere, but Intel currently isn't the real competition when it comes to ARM-vs-x86 - AMD is. And hadn't it been Apple's plan to take laptop CPU development into their own hands, they would have probably fixed their Intel problem by sourcing from AMD, so this Intel fixation in comparisons doesn't make all that much sense.

Thus it would make much more sense comparing the M1 against AMDs Zen 3 CPUs, or, with regard to a better match in terms of TDP envelope, the 4800U. Anandtech's early M1 benchmarks, which included the 4800U, hinted at much closer competition there, but IIRC they only benchmarked CineBench or something, so it was just one use case. I would love to see a broader comparison between the 4800U and the M1 in various application-oriented benchmark situations, including PostgreSQL.

If you're making the case to Apple users to upgrade to M1, the other Apple products are the relevant comparison. If you're hoping to entice Win/Lin users then AMD has more relevance.

The problem for me is that I need either macOS or Linux.

There don’t seem to me many laptops suppled with payed for Linux out of the box and even fewer with AMD processors to compare.

I’m looking for a small, lightweight option with good battery life.

I’m not interested in taking a crapshoot on whether replacing Windows will work “ok”. I want vendor support from the start.

A System76 Lemur, Dell XPS13 or Thinkpad basically. They don’t seem to come with AMD. To be honest,I don’t think I’d consider upgrading to an M1 mac right now either until they add more than 16Gb RAM.

Would this be an option you would consider?

https://www.tuxedocomputers.com/en/Linux-Hardware/Linux-Note...

The shape of the first chart is really interesting to me. The M1 has a positive linear slope from 1 to 4 clients. Then from 4 to 8 clients the slope is essentially zero, maybe slightly positive. And then beyond 8 the slope goes negative.

This seems to imply the four efficiency cores really are “efficiency” cores and don’t have much computing horsepower at all.

Or there is a bug, and it's not even running on those cores in the benchmark.

In SPECint2017, performance goes from 21.60 to 28.85 when going from 4->8 threads.

https://www.anandtech.com/show/16252/mac-mini-apple-m1-teste...

(https://images.anandtech.com/graphs/graph16252/119365.png)

There is a bit more background on the approach and toolchain here:

https://info.crunchydata.com/blog/postgresql-benchmarks-appl...

https://github.com/gregs1104/pgbench-tools

Now I am convinced there are at least some incentive for Apple to revive the XServe. The usage in their own Datacenter should be enough as a baseline for production along with Mac Pro.

Reply on news.ycombinator.com