Hacker News Reader

Comment by procodr | Hacker News Reader

For "Linux kernel compilation speed" they have compared a 16 GB model with 64 GB model for achieving 102.60% improvement! It's so funny =)

View on HN

Would increasing the RAM really make that big a difference in terms of kernel compilation time? It feels to me that 16GB should be plenty for any sort of caching that will speed things up.

The only way to know would have been to compare the models :)

I don't know about Linux compilation but... We build software written in cpp on macOS/Windows (clang/msvc).

As there are compilation units, translation of addresses, linking, etc.

Memory AND storage takes important part. While CPU might be "idle" in some parts.

For big projects especially, memory and storage speed would be important for proper benchmarking.

Yeah if you think about it in lower levels. Many optimizations use memoization to save clock speed. Therefore, you almost always need more memory to speed things up.

Yes, we intended to do an apples to apples comparison, but didn’t find the necessary units on hand when writing the post.

Saying that you intended to do something, and in the same breath that you were subsequently unable to do said thing, because you couldn't find what was required is very discouraging for a prospective customer. Also, "benchmarking" dissimilar systems comes across as marketing scam, scammy marketing, appealing to the deceivable, which some might view as reprehensible :(

Reprehensible...really? There's lots of that in the world, pretty sure this isn't it. They put the memory difference right there in the stats, it's not like they are hiding it.

Lighten up a bit. :)

Why didn't you just move the 16GB of RAM from the 11th Gen model to the 12th Gen model? Both use DDR4 after all, don't they?

Two different people on the team running the benchmark at two different locations (the challenges of being remote-first). To be clear, had we prioritized it, we could have built two equivalent systems for this benchmark at one site. However, we are mostly focused on building products, so most of our blog posts end up being someone having a interesting topic to discuss and pulling together data and assets we have on hand or can create quickly for it.

could you perhaps rerun it with one process mlockall()-ing 48GB of RAM, so both end up with 16G usable RAM?

I think you can use `brd` or `ramdisk` (NOT tmpfs) to eat up 48GB of ram on the newer machine. Just make sure you actually write data to the disks so that the memory is allocated.

That should be an easy way to do an apples to apples comparison and not require shipping any hardware.

GB, not Gb

yes, tanx

Reply on news.ycombinator.com