I9 9900k

Distracted · 18 Nov 2018 at 14:45

moorhen2 said:
SA and IO normally set too high on auto, they are generally used when RAM overclocking, when I am oveclocking my 3200mhz kit to 4133mhz they like an SA of 1.23v and IO of 1.22v, way lower than the auto rules set by the bios.

That's lower than auto was setting mine for the XMP 3200 / 14

I've lowered both to 1.1 and it seems fine. Any idea what sort of voltage I should be aiming for? I'm not taking my RAM any higher for the moment so I can raise the VCCIO and VCCSA if I need to later.

moorhen2 · 18 Nov 2018 at 16:22

Like I said, your SA and IO voltages are normally adjusted to compensate when your RAM and Uncore/cache overclocking, so if you were to run HCI or similar memory testing software your 1.1v may not be enough. As you said the 1.1v you are using is ok, but you will only know that's the case if you were to run any memory testing.

I don't tend to use XMP, I like to set values and tersiary settings manually.

Ian.

Distracted · 18 Nov 2018 at 16:25

I'll look into some memory testing software when I fall down that hole. Thanks for the tip.

moorhen2 · 18 Nov 2018 at 16:40

Distracted said:
I'll look into some memory testing software when I fall down that hole. Thanks for the tip.

Glad to help if I can.

What memory kit do you have, my G-Skill Trident Z 3200cl14 are animals. Max stable frequency 4234mhz.

Distracted · 18 Nov 2018 at 16:52

moorhen2 said:
Glad to help if I can.

What memory kit do you have, my G-Skill Trident Z 3200cl14 are animals. Max stable frequency 4234mhz.

Trident Z RGB 3200 cl14 8GB x 4 kit.

G.SKILL F4-3200C14Q-32GTZR

I know it is samsung B die and must be well binned given the lower cas vs the cl16 kit you can get, but I've not tested it beyond using XMP yet.

Distracted · 18 Nov 2018 at 16:55

I don't know what the IMC on my CPU is like either, so if I run into issues I wont know if it is the RAM or the CPU.

I do have access to another 9900k so I'm going to be testing that shortly as well.

moorhen2 · 18 Nov 2018 at 17:07

Distracted said:
I don't know what the IMC on my CPU is like either, so if I run into issues I wont know if it is the RAM or the CPU.

I do have access to another 9900k so I'm going to be testing that shortly as well.

You will only know how good your IMC is when you start to overclock the Cache and memory, at stock it shouldn't have any effect on the core overclocking, just concentrate on doing the core, then move on to the rest.

Sorry if it sounds like i am being patronising.

moorhen2 · 18 Nov 2018 at 17:11

Distracted said:
Trident Z RGB 3200 cl14 8GB x 4 kit.

G.SKILL F4-3200C14Q-32GTZR

I know it is samsung B die and must be well binned given the lower cas vs the cl16 kit you can get, but I've not tested it beyond using XMP yet.

Have read that the RGB versions are not quite as good as the standards for some reason.

Distracted · 18 Nov 2018 at 17:55

moorhen2 said:
You will only know how good your IMC is when you start to overclock the Cache and memory, at stock it shouldn't have any effect on the core overclocking, just concentrate on doing the core, then move on to the rest.

Sorry if it sounds like i am being patronising.

So if i raise the cache multiplier, the quality of the imc determines how far i can go and remain stable? Does that mean oc the cache will also depend on vccio or vccsa voltage?

Distracted · 18 Nov 2018 at 17:56

moorhen2 said:
Have read that the RGB versions are not quite as good as the standards for some reason.

That's a real shame. I'll be sure to let you know how I get on once i start fiddling with timings.

moorhen2 · 18 Nov 2018 at 18:08

Distracted said:
So if i raise the cache multiplier, the quality of the imc determines how far i can go and remain stable? Does that mean oc the cache will also depend on vccio or vccsa voltage?

No not necessarily, as you know the cpu and cache share the same voltage rail, so your set vcore is also the cache voltage, IO and SA are more for the memory side of things, once you start overclocking ram, these voltages will help.

moorhen2 · 18 Nov 2018 at 18:14

Well much better temps when stressing at 5.2ghz without any AVX instructions taking effect, 10 minutes of AIDA64 stress test without FPU ticked. Low 70's. 1.332v under load.

Now with Cache as well and the temps go up quite a bit, but still not as high as with AVX.

0121danwilliams84 · 18 Nov 2018 at 19:01

To put the above results in perspective. I have been trying to stabilise my 4.9GHz overclock. Its basically stable. HW info was giving WHEA errors every now and then. With vcore now at 1.340v had no errors recently. With this setting, the highest voltage shown in Windows is 1.312v.

moorhen2 · 18 Nov 2018 at 19:17

Dropped the LLC to 7 from 8, full load voltage at 1.279v-1.288v. Nice bit of vdroop, which I like. 1.340v in bios.

This is still without FPU.

moorhen2 · 18 Nov 2018 at 19:20

0121danwilliams84 said:
To put the above results in perspective. I have been trying to stabilise my 4.9GHz overclock. Its basically stable. HW info was giving WHEA errors every now and then. With vcore now at 1.340v had no errors recently. With this setting, the highest voltage shown in Windows is 1.312v.

Can you not use less voltage for that frequency by any chance ?

Distracted · 18 Nov 2018 at 19:24

So i didnt get to play much this weekend, I'm still waiting on parts for my loop which i expect to arrive next week.

Right now I'm running 5.2 avx -1 1.3v cache 43

So i raised cache to 50 for ***** and giggles and in aida i could run the stress test for 10 mins with no whea.

So i tried to see if my cinebench score had improved at all and it wouldnt complete. It took an additional .3v before it would run. So 1.33 and there was zero difference in the score with the cache at 43.

I know this test is quick and dirty and doesn't show the system as stable.

I'm thinking I might not bother with oc the cache too much since there seems to be little to gain by doing so.

I'll test more next week if the parts finally arrive.

Edit 5.2, not 5.1

moorhen2 · 18 Nov 2018 at 19:30

Add some FPU/AVX and the temps as expected jump right up.

moorhen2 · 18 Nov 2018 at 19:35

Distracted said:
So i didnt get to play much this weekend, I'm still waiting on parts for my loop which i expect to arrive next week.

Right now I'm running 5.2 avx -1 1.3v cache 43

So i raised cache to 50 for ***** and giggles and in aida i could run the stress test for 10 mins with no whea.

So i tried to see if my cinebench score had improved at all and it wouldnt complete. It took an additional .3v before it would run. So 1.33 and there was zero difference in the score with the cache at 43.

I know this test is quick and dirty and doesn't show the system as stable.

I'm thinking I might not bother with oc the cache too much since there seems to be little to gain by doing so.

I'll test more next week if the parts finally arrive.

Edit 5.2, not 5.1

Cache overclocking in general doesn't really give much in the way of gains overall, maybe in extreme overclocking. Might give you better times in AIDA64 Cache and Memory benchmark.

0121danwilliams84 · 18 Nov 2018 at 19:41

moorhen2 said:
Can you not use less voltage for that frequency by any chance ?

I really wish I could! It's just not stable with anything less. I mean it boots at 1.25 but is not stable at all. Testing thoroughly, I am at 1.340v 4.9GHz with no WHEA errors...yet.

Distracted · 18 Nov 2018 at 19:41

moorhen2 said:
Cache overclocking in general doesn't really give much in the way of gains overall, maybe in extreme overclocking. Might give you better times in AIDA64 Cache and Memory benchmark.

Seems that way. If i start using some software where the Cache speed is going to have an impact then I'll up it, but it is looking like I wont bother tbh.