MSI Z490 Tomahawk + 10600k wont hit 4.8 at default

[DaveB]

My early experience with MSI was meh. I built a dual Tualatin system based on on about 20 years ago, and had some scattered Z77 ATX and ITX mobos that were just OK. But recently, I was pleasantly surprised with the MSI B360 ITX and B450 mATX motherboards I picked up on the cheap as open boxes from Microcenter. Each was under $50, and both ran flawlessly. I sold the mITX system the MSI B360I Gaming Pro AC was in with an i5-9400F and GTX 1660 Ti, but still have the MSI B450M Pro-VDH running an i9-3900X with an RX 5700 XT. Most of my AMD and Intel builds over the past 22 years were based on Gigabyte, Asus and ASRock motherboards, with scattered MSI, Soyo, Supermicro, Tyan, BioStar, Epox, Shuttle, FIC boards and even one Albatron Socket 754 motherboard. My experience with most all of them was good, although I've soured on Asus and ASRock over the past few years. But if I found a good deal on one, I'd still pull the trigger.

 

[connta]

anyone have any comment on this?

"my voltages on Auto for VCCIO and VCCSA seem a bit excessive as 1.2 and 1.26 respectively. This is without any OC and from what I read those values are pretty high, even the motherboard itself labels anything above 1.15 and 1.2 as "red" when setting it manually so excess 0.05 on both. I tried lowering these to 1.14 and 1.19 and everything seems to be stable and temp also seems to drop a bit, what would you say these should be set on for default? "

basically, what should those voltages be on default for no OC, since I already proven they are 0.06 excessive anyone got a pointer how much lower should i go?

 

[EarthDog]

1.0 and 0.9 are defaults, iirc. Those values aren't remotely high as in cause a problem, just perhaps too high for the speed memory you're running. Im at 1.2 and 1.1 for ddr4 4000 for my cpu.

You can go as low as its stable.

 

[connta]

Thank you very much. Only asked to not waste time lowering by 0.01, from what you say i should be fine at 1.1 and 1.0 for the start of test and ill take it from there.

 

[connta]

After some experimentation I settled on this, no anomalies day to day, IBT 10 passes maximum, prime95 avx off 2-3h stable.

Per core: 48/48/48/47/46/46 (my proc default is 48/48/47/47/45/45)
VCore: 1.18
VCSA: 1.05
VCIO: 0.95
Cash: 44

Basically this is a slight OC compared to default but it doesnt need too much voltage and the heat is also great at 75C (Prime95, AVX off), considering its the middle of the summer. If up to 4 cores are loaded (gaming) most what I see is 48/48/47/47/47/47. At this point 48 all core just looks like too much hassle and most likely too much temperature without much clear benefit, all core 50 probably wouldnt be cooled properly by Megahalems anyway so for now sufficient, I still dont have a 120Hz display (but will soon). I'll further look at OC'ing the memory rather than the CPU.

 

[trents]

Is this stable when running stress tests with AVX instructions? AVX has found it's way into a lot of different kinds of software in the last couple of years.

 

[connta]

Well, thats a good question, so far it looks good.

When I turn on P95 with AVX ON the processor reverts back 100Mhz even tho AVX offset is set manually to 0, probably to try and stay in the 125W TDP spec. Also the voltage drops to 1.15 but even with that voltage drop the consumption goes up to 150-160W and Megahalems is not up to that challenge pushing temps to 87-88C which is a bit more toasty than I would like. Dont have any real world software that I know for sure uses AVX to test it out, would it stand its ground when it boosts to 4.8/7 and runs AVX. I assume Cinebench 20 uses AVX and it runs totally fine at 4.6 all core and 1T single core boosting to 4.8/7 also normally so theres that.

What I found out further is that when vcore is set to manual it wont exceed 1.18 but it also wont drop the voltage when the proc is in idle state which is expected. When i switch voltage to adaptive and also set it to 1.18 it will hold that voltage when boosting all core 4.6 but when boosting single/dual core (when we see the highest clocks) it will go over that and spike up to 1.24, is this normal or again MSI specific behaviour?

And lastly, about Cinebench 20 scores... all over the reviews I find that they vary widely even tho Cinebench shouldnt be too affected by the memory speeds as that could be the only variable i see, anyone have a comment on that? For example most reviews say that 10600k multicore C20 will get you >3600 yet I only see 3500 on default and 3630 on these "OC" values of 4.6 all core. Could this be to most motherboards using MCE by default and reviewers are just not turning it off or is it in fact memory difference?

 

[trents]

Yes, CB R20 uses AVX.

 

[connta]

Ok tested out new voltage mode, "Adaptive + advanced offset mode" and the results are seriously good. Basically this mode will let you make offsets adjustments at specific points of multiplier. I was like lets try it out see what happens and set all the points to "Auto" but the offset on "-" and just wanted to see where that would get me across points. I was surprised that under this mode it runs 4.6 all core stable at 1.11-1.13 (down from 1.15-1.18) and when i run C20 1T to verify boost it is doing from 1.13-1.18 and basically staying at 4.8 all the time, there are also no more spikes to 1.24.

It ran C20 1T and multi core (with best scores so far too, lol), Intel burn test 10 passes on High and its currently running P95 AVX off in the background for 40 minutes while i do day to day stuff and also play some Heroes of Newerth. All that at cool 72C, in the summer.

It is really a bad state of affairs if Intel is allowing MB vendors to blast CPU with so much unnecessary power just to see a slight uptick in benchmark numbers. This MB so far "overvolted" VCCIO and VCSA by 0.2 and CPU from 0.04 to 0.1 as it turns out, so quite a lot, around 10C difference with a decent cooler.

 

[EarthDog]

You can do that... but why so much granuliarty? Seems like a lot of effort for the returns. If you want to mitigate spikes, adjust LLC. One of the settings should be really close to no increase/little droop.

 

[connta]

He he, it really aint so much effort/gain thing as its a hobby, trying things out, tinkering... but as it turned out, no "granularity" needed, when set on Auto and just offset to "-" it does actually great, it was suppose to be a baseline setting from which to go but it turned out to be a "hole in one" without the need to touch much else, it ran all the benchmarks with good results, prime for 3h and some gaming all under superb thermals, for sure looks stable and with minimal effort. I mean i did a couple days of intermittent testing and only came close to results it gives now, wish I tried this first as a baseline.

Tried playing with LLC before but it didnt seem to cut the spikes actually.