Early Linux 4.14 Kernel Benchmarks Are Looking Promising

[davidbepo]

like i said an interesting release with performance improvements
too bad i can't shut down on it : https://forum.manjaro.org/t/cant-shu...-4-13/30574/14

[starshipeleven]

These are not minor improvements. Makes a little suspicious - is there a good reason there's so much of an improvement?

No. 640k ram ought to be enough for everyone, progress is a lie.

[sdack]

Well, maybe…

Interesting. Seems to be a feature only available with newer CPUs however.

> fgrep -w pcid /proc/cpuinfo || echo ""

> _

[schmidtbag]

No. 640k ram ought to be enough for everyone, progress is a lie.

How is that quote remotely relevant to my statement? I didn't question if there should be an improvement, I'm suspicious as to why there is so much of one. Last time I checked, this article doesn't specify what was changed to cause this boost in performance. So, I was wondering what was the cause of this. We get wary when a kernel update results in a loss of performance without a known reason (where sometimes that may just the side effect of improving upon something else). The opposite can be true too - a performance improvement this big could be the result of something else that isn't [fully] working properly anymore, hence my question.

HyperDrive seemed to understand what I was saying, why can't you?

[starshipeleven]

How is that quote remotely relevant to my statement?

Your idea of "maxiumum attainable performance before something is fishy" is wrong.
Software isn't anywhere near full hardware utilization in many places.

HyperDrive seemed to understand what I was saying, why can't you?

What told you I didn't understand?

I just reacted differently.

[schmidtbag]

Your idea of "maxiumum attainable performance before something is fishy" is wrong.
Software isn't anywhere near full hardware utilization in many places.

No, it isn't. You don't get up to a 65% performance increase out of the blue because "hardware isn't fully utilized in many places". To give you the benefit of the doubt - some of the tests were as little as a 6% increase, something that could easily be a result of hardware running more efficiently. But, that still doesn't change my point and my question: why? What caused this? I'm not complaining, I'm just curious. Sorry curiosity is so offensive to you.

What told you I didn't understand?

The fact you [somehow] thought I was implying that there shouldn't be an improvement, or, that an improvement isn't needed.

[sdack]

HyperDrive seemed to understand what I was saying, why can't you?

Stop throwing around one good question after the next! You're making the trolls dizzy.

[starshipeleven]

No, it isn't.

It's not the first time that some synthetic benchmarks got massive performance spikes after some patch, so I don't see why you got so worked up by this case.
What synthetic benchmarks show is that there is an overall improvement, and it's not so large to ring alarms (only some cases were spiking a bit, and it's synthetic), that's all.

The fact you [somehow] thought I was implying that there shouldn't be an improvement, or, that an improvement isn't needed.

Well, you did put it down a bit like that, that's why I reacted with that well-known sentence about 640k memory.

[starshipeleven]

Stop throwing around one good question after the next! You're making the trolls dizzy.

ah come on, quit hiding behind shmidtbag's skirt and show us you are a real man, for once.

[HyperDrive]

Interesting. Seems to be a feature only available with newer CPUs however.

That's odd. How old is your CPU? I have a laptop with a Core i3-380M (mid 2010) which does support PCID (not INVPCID, though, which was introduced with the Haswell family). Oh, and this seems to be one of those "features" Intel likes to fuse off at the lower end, for some unfathomable reason.