Single Core Boost query: 4790k

[MGediG]

SPECS:

4790k
Gigabyte z97mx Gaming 5

Noctua DH15S
5 other fans including industrial top fan



Currently have overclock stable at 4.8Ghz, Vcore 1.286, I can probably tweak it lower...

I have all C states off and hyper threading off. I have tweaked a few other voltages: ring, vrin, memory, and uncore is at 40.

Would like to boost the first core to 50x and so I've turned on turbo boost, set all cores to 48 and only the single threaded turbo to 50, I haven't changed wattage or amps from auto.

In all of my monitoring and benching so far I haven't seen the single core or any figure hit above 4.8Ghz.


What am I doing wrong? Do I need to managed the power in a different way?

...and how does single core usage work anyway? I assume windows is always doing lots of things in the background, so when would the computer actually switch to one core only? Or does it do it all the time but very quickly off and on? So it might not be noticable?

 

[mcconkeymike]

I don't have an answer for you, but I do have a question. Why are you trying to hit 5ghz and what made you think that turning off hyperthreading was a good thing? Yes high clocks do help get work done quicker, but you've seriously handicapped your 4790k by shutting off hyperthreading. What is the point of what you are trying to accomplish?

 

[grimfox]

I think HT at 4.8G > 5GHz no HT.

Single core boost is primarily for heavy single threaded applications. Usually when most of the cores are idle but there is a single core that's very busy it will boost that core to work a little faster since it doesn't have the power of the other cores to also manage. Best analogy I could think of is a dam with 4 gates. You can open one gate really far and it wont flood the river down stream, but if you open all 4 that wide it would cause a flood. It's not a perfect analogy. Particularly in your case because your boost is only a little higher than your all core frequency.

edit(word choice)

 

[TJ Hooker]

What application(s) are you using to bench and try to hit 5 GHz?

 

[MGediG]

Apologies, I should have been more specific.

I am trying to run flight simulators (il2 Sturmovik BoX, and also il2 CloD). They are both heavy on the single thread use, monitoring shows about 50% usage on Core 1 and others half or less. It's evident I'm getting a bottleneck somewhere for a campaign mod I'm using, called PWCG. You can set various "densities" of other aircraft, vehicles on ground, all sorts of objects basically, and I'm trying to keep the map as populated as possible. Also the AI for all other planes takes some CPU power. There is an effect that users are experiencing to varying degrees - "slow motion". The game slows down even though fps counters are not, and it becomes impossible to run fast forward (speed up time mode for autopilot boring times). My system is doing fairly well compared to some, some get in in vanilla campaign mode and at low density. I'm trying to optimise as much as possible so I can run a realistic campaign experience with this mod, which allows much more "density" than vanilla. RAM may also be a bottleneck, but less likely, and I have tweaked it to the edge of what it can do and there is no upgrade worth buying for this setup. So, single core boost is all I have left to do.

There are also lots of stutter isses with the game, most of which I have smoothed out by turning off C states and lots of latency optimisation in Windows etc

That's also the reason I turned off hyper threading, for this game. It's a fix for stutters/micro stutters that are evident with it on. There's a slight drop in FPS but it's a worthy sacrifice to kill the stutters. I'll be turning it on for pretty much any other use.

I intend to have more than one profile in the BIOS, one for this game, one for most other games, and one for work. I have a hotswap setup with multiple installs.

I am using p95 and aida64 as my primary stress testers. I have also used a benchmark called Single Precision Julia within aida64, I'm not entirely sure what it does, and I haven't googled it yet! but I assumed single core test. I have also run games and GPU benchmarking, with the turbo settings above, with one core on 50. None of them seem to output a peak in single core frequency with my setup as is. Also have used Novabench, though it's not really very detailed or intensive.

I'm using CPUID HWMonitor to look at stats, I was using RealTemp but can see single core data now.

Will be more than happy for some other options to specifically bench single core use, and ideas to achieve what I'm hoping for. Thanks


edited some words - saw the percentage use of Core 1.

 

[MGediG]

OK so apparently I need one or some C states enabled for turbo to even work. I'll try this out and report back, any other input is appreciated