News Core i9-13900K Early Review Shows Big Gains Over Core i9-12900K

[Rdslw]

On a 35W system CPU compute becomes much less useful, how many people need cinebench or other heavily multithreaded apps on a 35W laptop?
True, we don't, but a lot of us uses them as entry point for other work related things, and after spending last 5 years with laptop, I can tell you, it's noticeable. That cinebench score should mean I will wait less for ansible or docker for local dev. You should see small speedups across the board in everyday work, regardless of the field. From my experience most demotivating this is when you have to sit on your ass and wait, not being able to continue, because your machine don't keep up.

A lot of corporations give laptop to workers, so if 35W and under become stronger, most business should see a faster pace, even if it's tiny.

 

[TerryLaze]

True, we don't, but a lot of us uses them as entry point for other work related things, and after spending last 5 years with laptop, I can tell you, it's noticeable. That cinebench score should mean I will wait less for ansible or docker for local dev. You should see small speedups across the board in everyday work, regardless of the field. From my experience most demotivating this is when you have to sit on your ass and wait, not being able to continue, because your machine don't keep up.

A lot of corporations give laptop to workers, so if 35W and under become stronger, most business should see a faster pace, even if it's tiny.

"I will wait less for ansible or docker for local dev"
Those are automation apps, do they really load up your CPU to 100% while you write scripts or during any part of working with them?

Because if they don't at least use 100% of your CPU then no cinebench is completely useless info, and even if it loads 100% if it has a much lower IPC usage than what cinebench uses than these numbers are still useless.

You have to investigate what really speeds up these tools, it might be memory, disk, cache or single threaded speed or whatever else, you can't just assume that everything gets faster because of multithreaded performance.

 

[JamesJones44]

This isn't just an SMT deficit. It's specifically an E-cores deficit. The example I mentioned was four P-cores and zero E-cores vs four P-cores and eight E-cores.

I didn't say it was an E-cores deficit, I wasn't even talking about E-cores at all! What I said is SMT affects max performance of a core as well. We are talking about max performance of P-Cores, depending on your workload SMT will affect P-Core performance. If you want absolute max performance from a CPU or P-Cores you need to examine your workload and adjust accordingly is what I said.

 

[Wolverine2349]

Some current games already show a performance benefit with more than 8 cores. Unless you want to be behind the curve, 8 cores, 8 P-cores, should be the absolute minimum to aim for, for max performance in a new system you plan to keep for 5+ years.

Which games actually can use more than 8 cores meaningfully assuming you are not streaming or running background intensive tasks??

 

[alceryes]

I didn't say it was an E-cores deficit,

Right. But I did.
In fact, I did in the post you replied to when you referred to SMT.

What I said is SMT affects max performance of a core as well. We are talking about max performance of P-Cores, depending on your workload SMT will affect P-Core performance. If you want absolute max performance from a CPU or P-Cores you need to examine your workload and adjust accordingly is what I said.

Ahhh, got it. You're just saying that any task utilizing more than one core will suffer a performance hit from SMT. Of course, any SMT function will introduce a performance penalty. This is universal. I'm referring specifically to the net performance loss of having E-cores active in many workloads.

When considering only P-cores (or AMD), the net performance gain from adding more P-cores to a process/task completely nullifies any SMT penalty and, in fact, provides a great performance boost. When considering Intel's 12th gen Alder Lake P-core plus E-core architecture, the performance penalty introduced by L3 contention and the lower bus speeds of the E-core interface are, in many cases, greater than any performance gain realized by the E-core's work performed. Note that you don't even need to be using the E-cores. As long as they are enabled, there will be a performance hit due to bus speed limitations.

 

[alceryes]

Which games actually can use more than 8 cores meaningfully assuming you are not streaming or running background intensive tasks??

Check out this vid -

View: https://www.youtube.com/watch?v=Cbyl4q3QFYA

 

[Rdslw]

"I will wait less for ansible or docker for local dev"
Those are automation apps, do they really load up your CPU to 100% while you write scripts or during any part of working with them?

Because if they don't at least use 100% of your CPU then no cinebench is completely useless info, and even if it loads 100% if it has a much lower IPC usage than what cinebench uses than these numbers are still useless.

You have to investigate what really speeds up these tools, it might be memory, disk, cache or single threaded speed or whatever else, you can't just assume that everything gets faster because of multithreaded performance.

I cannot say my cpu is fully loaded when I run local env, but I see a lot of bursts where 3 to 10 cores gets full throttle for like 15 seconds. (I have 6/12 currently)
I don't assume random cpu speed boost will elevate all my issues, 35W system with bursty loads like docker and ansible gain a lot from single threaded speed, even though they don't load your machine 100%, multithreaded headroom means you don't freeze your OS while they work.
I notice slowdowns in my everyday GUI work, when in background any tests/builds/automation is running, so for that extra multithread gain, have a high chance to elevate it, even when I cannot quantify the result. I assume new system will not slow down with my usual workload, while current one noticeably do.
If new generation will be insanely better than what I am currently riding, and I think ddr5 will do me a lot of good, I might be able to run multiple parallel builds/test cycles or similar thing that helps me not stare at automation, doing nothing.
before you say it, we run jenkins doing same thing, but we always run units locally first and if they pass, we let jenkins run full suite of tests.