8086k - Single/dual core vs all core overclock?

[ESzczesniak 128]

My 8086k CPU is supposed arrive tomorrow to complete my new build.  It is almost exclusively used for P3D v4 and DCS World.  P3D is well known as not a huge multicore user, and DCS I believe is better, but not much.  I will overclock, but also don't want to overly stress the CPU or have sky high temps. 

Is there more value in overclocking only one or two cores to a 5.2-5.3 GHz (that would be trivially easy probably) and use an affinity mask to assign P3D to an overclocked core?  Or would it be better to overclock all cores, but possibly accept a lower overclock in the 5.0-5.1 GHz range? I am running a water cooler and should ya e fairly good temp control  

I appreciate any thoughts. 

Edited November 12, 2018 by ESzczesniak

[martin-w 3,235]

I don't believe anyone here has experimented with per core overclocking. So in essence you will be a pioneer and we will be the ones awaiting your results and guidance.

All Cores at 5.1 GHz isn't a trivial overclock, it's a pretty high overclock. Yes, the 8086K is a binned 8700K, so the voltage and thus temps stand a good chance of being lower, but worth remembering that the max Turbo of 5 GHz only applies to one core and only if one core is active, so 5.1 on all cores is significant, especially with HT on too.

If you want a moderate, easy, all core overclock, then MCE might be a way to go. MCE will set all cores to max turbo. The voltage would be generic, in accordance with your board manufacturers auto rules, but could be tweaked lower if you desired. 

Also worth remembering the difference between 5 GHz and 5.2 GHz in terms of frame rate is very small.

[Bob Scott 7,719]

43 minutes ago, ESzczesniak said:

My 8086k CPU is supposed arrive tomorrow to complete my new build.  It is almost exclusively used for P3D v4 and DCS World.  P3D is well known as not a huge multicore user, and DCS I believe is better, but not much.  I will overclock, but also don't want to overly stress the CPU or have sky high temps. 

Is there more value in overclocking only one or two cores to a 5.2-5.3 GHz (that would be trivially easy probably) and use an affinity mask to assign P3D to an overclocked core?  Or would it be better to overclock all cores, but possibly accept a lower overclock in the 5.1 GHz range?

I appreciate any thoughts. 

It's a huge mischaracterization of P3Dv4 to classify it as not being a multi-processor program.  Yes, its performance still depends heavily on having a fast single-core capability, because everything revolves around the throughput of the main thread running (usually) on Core 0, but it *heavily* tasks the other five processors, but on a more intermittent basis.  Being able to power through those periods of high workload at a fairly steady frame rate is key to a smooth simulation.

The 8086K is a very good CPU...it should be able to hit 5.0+ on all cores with any good cooling solution, whether that be an AIO sealed liquid loop, a high-quality air cooler like the Noctua NH-D15, or something more aggressive.  5.0 GHz on all six cores will rock P3D for you.

The only way I know of to O/C just a select few cores is to use affinity masking to keep processes completely off the others, which would be a big mistake.  I think a lot of folks misinterpret the "Core Ratio Limit" settings in the BIOS as allowing separate per-core clocking, but my read of the docs and experience with it is that it sets different clock multipliers (for all active cores) depending on how many cores are active.  So you can set one speed for just one active core, another if two are active, another for three, etc.  P3D will keep all six cores gainfully employed, so what you get is the lowest setting, the one for all 6 cores active, most of the time.  In other words, setting separate per-core clocks is not an option.  It might be different with very high-end motherboards like the ASUS Maximus Apex, but there is no per-core clocking with my Maximus Hero.

Regards

[ESzczesniak 128]

I wasn’t saying 5.1 GHz was trivial, trust me. Given reports I’ve seen, >50% of 8086k will hit 5.0, but it’s arpund 50% that will do better (wish I could find that site again). 

The 5.2-5.3 on a single core I was expecting to be easier since that’s only a couple hundred MHz over the single core turbo. 

[Bob Scott 7,719]

Silicon Lottery overclocks and bins a lot of chips, so they have some pretty good depth in their statistical data.  They say 100% of the 8086Ks are hitting 5.0:

https://fuse.wikichip.org/news/1412/core-i7-8086k-overclockability-silicon-lottery-stats/

Regards

[martin-w 3,235]

 

Quote

So you can set one speed for just one active core, another if two are active, another for three, etc.

 

Yes that's absolutely correct Bob. That's how it works on my Maximus X Code and I believe my previous boards. You cant select "a specific individual core" to overclock, like core 1 or 2, just the frequency that equates to the number of cores active. So akin to how Intel Turbo works. It's specifically worded in the BIOS, in the per core mode, as number of cores active. 

Edited November 12, 2018 by martin-w

[martin-w 3,235]

15 minutes ago, ESzczesniak said:

The 5.2-5.3 on a single core I was expecting to be easier since that’s only a couple hundred MHz over the single core turbo. 

 

in theory yes, but as Bob pointed out you cant choose a specific core for that overclock, and it's also true that different cores have different max frequency capabilities. Silicone lottery also applies to individual cores. 

Edited November 12, 2018 by martin-w

[Bob Scott 7,719]

One other point worth mentioning is that the stats I linked to above are for chips tested with hyperthreading turned on.  All other things remaining equal, without HT you can generally get a 100-200 MHz higher stable clock, or keep the same clock at 20-50mV less core voltage, making it run a fair bit cooler.

 

[martin-w 3,235]

HT equates to 10 degrees usually. 

[ESzczesniak 128]

Thank you all for the clarification.  I understand what is being said now, but I didn't realize it was based on core utilization, not a per core process.  It makes it a no brainer at this point.  And I already plan on turning HT off. 

[vc10man 732]

1 hour ago, w6kd said:

It's a huge mischaracterization of P3Dv4 to classify it as not being a multi-processor program.  Yes, its performance still depends heavily on having a fast single-core capability, because everything revolves around the throughput of the main thread running (usually) on Core 0, but it *heavily* tasks the other five processors, but on a more intermittent basis.  Being able to power through those periods of high workload at a fairly steady frame rate is key to a smooth simulation.

The 8086K is a very good CPU...it should be able to hit 5.0+ on all cores with any good cooling solution, whether that be an AIO sealed liquid loop, a high-quality air cooler like the Noctua NH-D15, or something more aggressive.  5.0 GHz on all six cores will rock P3D for you.

The only way I know of to O/C just a select few cores is to use affinity masking to keep processes completely off the others, which would be a big mistake.  I think a lot of folks misinterpret the "Core Ratio Limit" settings in the BIOS as allowing separate per-core clocking, but my read of the docs and experience with it is that it sets different clock multipliers (for all active cores) depending on how many cores are active.  So you can set one speed for just one active core, another if two are active, another for three, etc.  P3D will keep all six cores gainfully employed, so what you get is the lowest setting, the one for all 6 cores active, most of the time.  In other words, setting separate per-core clocks is not an option.  It might be different with very high-end motherboards like the ASUS Maximus Apex, but there is no per-core clocking with my Maximus Hero.

Regards

Just what I wanted to read, Bob, as I already have the i7-8086k when it first came out. There was me thinking I'd goofed with only a singe core o/c of 5Ghz as compared to the latest offering, i.e. the i9-9900k.

[shivers9 1,174]

1 hour ago, vc10man said:

Just what I wanted to read, Bob, as I already have the i7-8086k when it first came out. There was me thinking I'd goofed with only a singe core o/c of 5Ghz as compared to the latest offering, i.e. the i9-9900k.

Only time will tell but all things considered, I think a good delided 8700K or the 8086K which seems to be the best spot to be in. All core at 5.0GHZ with not heat problems with any kind of decent cooling solution is mighty sweet. That's how I roll!! Now if we can just get this stupid GPU goat rope sorted out so I can order a new card. My 980TI is sweating like a one legged man doing the 100 yard dash.😁

Edited November 12, 2018 by shivers9

[ZLA Steve 370]

I did a few weeks of testing with per core OC versus total OC and found that my temps were not that much different. (Maybe 2-3º)

I run my CPU at a comfortable 5.0-5.1 (room ambient temp being the biggest driver of the top end) and top out at an avg of about 76º when using P3D & DCS in VR.