Latest Linux Code Smashes 14M IOPS Per-Core With Intel Core i9 12900K + Optane

[phoronix]

Phoronix: Latest Linux Code Smashes 14M IOPS Per-Core With Intel Core i9 12900K + Optane

Back in the day, 2.5 million IOPS per core was an impressive feat... That day was little more than one year ago. With faster hardware and relentless optimizations by Linux kernel developers, 14 million IOPS per core is the new record now achieved...

https://www.phoronix.com/scan.php?page=news_item&px=Linux-14M-IOPS-Per-Core

[Danny3]

Wonderful!
Great time to be alive and great time to be a Linux user!

[fransdb]

If indeed the same hardware has been used, it's indeed impressive. At the same time I am wondering at what cost?.
I mean, if beside code optimization all kinds of safeguards have been removed, how trustworthy/stable is the resulting code?
That said, how about portability across various architectures?

[ermo]

A few months back, I opined (trolled?) that it couldn't be a coincidence that many Linux kernel devs used AMD hardware.

I'm here to eat crow now. Looks like the 12900K P-cores and the overall arch + platform is strong (and stronger than AMD's older 5xxx series cores + arch + platform).

GJ intel. Now, if you could just enable ECC support on the K-series...

[kozman]

I have mixed feeling about these perf gains and I'll use an analogy to encompass my thoughts on it. Firstly, cheers to super smart guys like Jens for realizing these gains. I bow low and humbly to their coding and mental acumen for getting perf to this new level. Now comes the analogy. Probably 15 years ago I switched from incandescent bulbs to CFL. My electric bills did go down a lot. For a few years. Then they sloooooooowly started creeping up year over year. Not noticeably at first but they did. Until one day I was getting bills that were as much or more than my incandescent bulb bills used to be. I know because I've kept a spreadsheet of all my home utility bills since 2005. Then LED bulbs came along. But I only switched to LED when the good (current gen) LED bulbs came out and now my whole house is LED bulbs. But my bills remain the same or more. So much for efficiency purporting to reduce operational costs when the industry just slides to cost scale to match what you were paying before you ever switched to CFL or LED.

So for the same reason that when disk space and memory got cheap and allowed app developers to create gigantic memory-consuming and disk-hogging programs, no one cared because "fuck it, disk space and RAM are cheap" so let's just bloat up everything. Why bother making efficient programs? This too will happen with the gains being afforded the Linux kernel through Jens' efforts and, in short order, they will cease being gains like my analogy from above. The industry will opt instead for the path of Leeroy Jenkins, Cloudbros or, as the gaming crowd likes to shout, "LET'S GOOOOOOOOOOOOOO!" Tell me I'm wrong.

[Volta]

kozman yeah, let's not optimize the Kernel, so apps developers will optimize their apps or - what's more probable, they'll switch to faster one.

[dragorth]

kozman. This is what happens, it is market dynamics. When you decrease the cost of bad programs running well, you get more bad programs. When you decrease the amount of energy used in every home, the bill doesn't go down, because the costs are the same. To save money on the electric bill, you have to chase the future, not the present, or get off the ride entirely.
I predict the electric company will think of a way to charge those that don't use the grid, probably through subsidies and the government increasing taxation on those off the grid.
You need competition to combat this. Linux optimization is a great example. If Linux was the dominate OS, this optimization might still happen, but maybe it wouldn't. It is already the leader, and thus the best, so why bother, there are less incentives to spur them on.

[cb88]

A few months back, I opined (trolled?) that it couldn't be a coincidence that many Linux kernel devs used AMD hardware.

I'm here to eat crow now. Looks like the 12900K P-cores and the overall arch + platform is strong (and stronger than AMD's older 5xxx series cores + arch + platform).

GJ intel. Now, if you could just enable ECC support on the K-series...

Not sure what your point is... 12900k launched in oct 21... AMD's last major update launched Nov 20... so that platform was available to use for 2 years almost already. And AMD AMD launches their DDR5 platform at the end of the year... if anything the 12900k is too little too late.

Its not like either of these platforms has earth shatteringly different performance from each other... the Zen 3 cores can hit at least 10M IOPS from what I remember.

[kozman]

kozman yeah, let's not optimize the Kernel, so apps developers will optimize their apps or - what's more probable, they'll switch to faster one.

Don't misunderstand what I mean. This benefits the Linux community. It's a good thing. I guess I see it like I see the works of famous painters. They all pretty much died poor and broke. It's afterwards that their works exchange for millions of unrealized dollars for them. It's the Amazon's and other hosting giants who will, IMHO, undeservedly reap the benefits of this by charging more, rather than less, for a benefit they didn't create but should be passing along to customers. That isn't market dynamics. That's just greed, plain and simple. This improvement will make what runs sufficiently well for them now---on the exact same hardware next year---run far faster. If this has the potential to quadruple performance on the same gear for not just a narrow group of tasks, only an idiot would upgrade knowing this is coming down the pipe. This will be a gigantic cost reduction for them whilst they tout huge performance gains for their services and the cost-to-consumer increase therein. Sure would be a great time for the little guy to begin setting up shop.

[sdack]

I have mixed feeling about these perf gains ... no one cared because "fuck it, disk space and RAM are cheap" so let's just bloat up everything. ... Tell me I'm wrong.

Not everyone is a Putin. There are still good people in this world. Many volunteers in open source do not program for the money, but simply because they can and because it serves the greater good. Contributing code to make open source better is a wonderful thing in this crazy world. You should not have mixed feelings about it, but only good ones.