Intel Cuts Prices of iGPU-devoid 9th Gen Core F and KF SKUs by up to 20 Percent

[LFaWolf]

It's not going to be very noticeable if you don't have some main process that is going to be affected by context switching time... 20ms of CPU time to switch threads to something else and back won't make excel feel much slower, but in a game 20ms can add enough to make a noticeable stutter.

If you have ever compared the 4c4t to 4c8t CPUs you would know exactly what it is like.

Most people don't complain much about stuttering (csgo of all games is one of the worst stutterfests on everything other than zen2 and most people don't complain) but if you are accustomed to 0 stutter gameplay you will definitely notice the difference. If you can deal with it keep it I guess, but I'm not going to recommend a second tier experienced if I know there is a better one for the same or less.

The people who are hit most are the cloud server guys, ask any of them and you will realise how big the hit actually is, it doesn't have as huge effect on games, but the effect is there.

You got the unit of measurement very wrong. There is no way context switching will take 20ms. That is forever. It is in microseconds -
https://eli.thegreenplace.net/2018/...ching-and-memory-overheads-for-linux-threads/

Even years ago it was still microseconds.
https://www.usenix.org/legacy/events/expcs07/papers/2-li.pdf

There is no way in gaming that you can notice the microseconds in context switching. You noticed some lag probably because your CPU is pegged, or you run of memory and it was doing some swap to the page file.

 

[GorbazTheDragon]

You got the unit of measurement very wrong. There is no way context switching will take 20ms. That is forever. It is in microseconds -
https://eli.thegreenplace.net/2018/...ching-and-memory-overheads-for-linux-threads/

Even years ago it was still microseconds.
https://www.usenix.org/legacy/events/expcs07/papers/2-li.pdf

There is no way in gaming that you can notice the microseconds in context switching. You noticed some lag probably because your CPU is pegged, or you run of memory and it was doing some swap to the page file.

As is hinted by both papers, context switching is heavily dependent on cache usage, in the case of cache side-channel attacks any context switch basically results in the worst case for the context switch.

I can't find the thread any more, but there was a guy on OCN who ran some (iirc cinebench) benchmark on a larger thread count than the CPU had (some haswell system back in 2015 or so) and resulted in significant (10%+) performance degradation past of the order of 64 threads per (SMT2) core. This was back when the big topic of discussion was whether the G3258 was at all useful or not... The windows scheduler, particularly on the consumer versions, is known to be much worse than the Linux ones in this regard, those numbers should give you a hint as to how much worse...

One context switch might not lead to a noticeable difference between the two, but when you have hundreds of threads combined with regular interrupts from user inputs you end up with a lot of context switches being necessary. If your scheduler then decides it wants to switch away from some thread only to have to go back to it again like windoze seems to do, those microseconds really add up a lot.

 

[LFaWolf]

Nah, I don’t know where you get this idea that the CPU is busy switching context back and forth all the time, but it does not work that way. The CPU will switch context when it suspends the process, and won’t switch back unless the process becomes active again. Why would it switch context for an active process while you are gaming? Also, how many processes are you running? The CPU is not switching context for all the running processes, and giving the overhead for context switching is only a few microseconds, it really isn’t much.

 

[GorbazTheDragon]

You can bet your bottom dollar most of them aren't truly staying suspended all the time...

 

[LFaWolf]

You can bet your bottom dollar most of them aren't truly staying suspended all the time...

What? What are you talking about? If the OS suspends your Excel process, it won't activate it until either you clicks on the Excel program to do something.

Looking at my own system, a heavily utilized machine for software dev, I only have around 300 processes. Context switching among them should still take only microseconds, and keep in mind that the OS will not switch context on all 300 processes at the same time, if at all. Saying the context switching causes Intel CPU to have stuttering is FUD, I am afraid.