Hacker News Reader

Comment by throw0101c | Hacker News Reader

> While a new CPU alone isn’t a thrilling update, it’s important that they do these regularly so consumers looking to buy aren’t forced to buy a 3 year old product with no idea when a refresh will come.

Also: incremental updates add up.

A (e.g.) 7% increase from one year to the next isn't a big deal, but +7%, +7%, +7%, …, adds up when you finally come up for a tech refresh after 3-5 years.

View on HN

Its 2025, the fact that Apple is delivering CPUs with actual, noticeable annual performance improvements is pretty astounding in itself. Sure its not 1990s levels, but its still pretty great.

M silicon/SoC is the best thing to happen to computing, for me.

I have 64GBs of RAM in my Macbook Pro. I load a 48GB DuckDB to RAM and run real-time, split-second, complex, unique analysis using Polars and Superset. Nothing like this was possible before unless I had a supercomputer.

Is it really that much better than some small form AMD Ryzen with 2x32 SODIMM thrown in? I get that the M series is amazing in terms of efficiency and some people love Apple hardware but you could likely have had that performance with a $700 setup.

The only server that actually matched the performance of a Mac Studio was XEON Max series (formerly codenamed Sapphire Rapids HBM) with 64GB of integrated memory into the CPU package. the latency between the CPU and RAM is simply too big in a regular PC.

for DB's bandwith to RAM and Storage is just as important.

that's the thing. latency to RAM is everything. I would take 10x lower bandwidth in RAM vs 100x better latency from CPU to RAM.

The only x86 CPU that does this is the Xeon Max: https://www.intel.com/content/www/us/en/products/details/pro...

There are other possible solutions but they are expensive.

the problem with Macs, among other things, is lack of ECC RAM.

Have you tried other PC with 64 GB of RAM?

yes. we have PCs. AFAIK, the cheapest PC that compares for my workflow is an EPYC/NUMA or another very expensive CPU/latency optimized server. We have a complex stack, with clients running unique queries that we can't predict and gigabytes loaded into RAM, L3 cache doesn't always save us. I haven't found another solution, I wish we could drop the Macs cause the OS is pretty awful.

We're using Macs as servers. But it's a small operation.

i'm guessing you're using macs newer than M2, so they can't run linux; but i wonder if fedora server (asahi remix) would suit your operation well

we also tried Asahi on M2 Ultras but we had big performance issues on the DBs compared to Mac OS.

Also: we shouldn't make a big deal out of every update then. Celebrating M1: alright, but then M2-M500 are boring and not even worth noting, because you know there's a new one every year.

adds up to 22.5%

> adds up to 22.5%

after 3 years

and 40% after 5 years.

yep! I was only throwing in the final number for those folks who were like me and saw math, and wanted an answer.

Sorry for the nit, but it's compounding improvement, not additive. Its 25% after 3 years and 45% after 5

That’s what he’s saying.

No. He said that the 7% 'add up', when the proper term would be 'compound'.

You are both right. What is compounding? It is when you add the gains, year by year.

The gains are multiplied though, no?

In the vocabulary of finance you don't multiply in gains, you add them. It probably historically derives from dividends being added. At the transactional level you never actually multiply money, after all (unless you're a bank).

What is multiplication if not adding over time?

Everyone is right! It’s a multiply-accumulate (accumulate of course a synonym for add)

Reply on news.ycombinator.com