13600K build and Crucial Ballistix Max DDR4-4400 tweaking

[Voodoo Rufus]

New build tweaking has commenced.

Build:
Asrock Z690 Extreme
13600K
2x16GB Crucial Ballistix Max DDR4-4400. Bios says it's single rank, also.

See screenshot for current specs. I was able to get 4266 running on XMP and Gear 1 with stock volts all around. 4400 will not play nice with VCCSA up to 1.35V, so I'm probably sticking with 4266. Looks like good benchmark numbers though. I cannot figure out how to manually configure Command Rate 1T in the bios though. The more detailed settings in this Asrock bios elude me.

Next, I want to start tweaking the primary timings and see if I can snug them up. Then on to wringing out this 13600K if I can figure out how to abolish those stupid PL1/2 limits. I just want a temperature limit.

Update 1: 4266 was stable with memtest64, but not in P95. 4200 is stable in P95, so rolling with that for now. Still trying to figure out how to unlock the "safeties" on the CPU.

 

[ShrimpBrime]

If it's A2 pcb, keep the 2T.

Should scale with vdimm. Especially if you want tighter than C19 at 4200 and up.

Vccia and Vccio should be fine 1.20v with just the xmp profile...

Not sure with AsRock bios layout though. Haven't used one in a long time.

 

[Woomack]

I have no idea how it's overclocking on ASRock, but their latest motherboards were pretty average for RAM OC.

Below are some examples on Ryzen 5900X and 4650G APU, so maybe you get an idea about some settings.It should run at least at 5000-5200 CL18 on Intel, but I remember that some motherboards required CL19. VDIMM up to 1.6V. Other things I think I told you in the "conversations". Sub-timings won't go much tighter than in XMP, but you can play with tREFI and one by one with other timings (I don't remember what I was setting).

On Intel, you are not losing as much at Gear 2 as on AMD, so try to push it higher.
If I find other results, then will post them. They were in many threads on the forums, and I'm sure I had results on Intel too.










I ran this kit at 5400 CL18 with Ryzen 5700G APU for maybe 2 months in my gaming PC. For tests up to 5700 CL20. The same as I already mentioned, 1.60-1.65V VDIMM was about max, above some point if you want higher frequency then you only relax timings as this IC doesn't like too high voltages - it won't damage/degrade, but will show errors or won't boot at all when the voltage is too high. 1.60V is what some brands were setting in XMP at 5000+, so should be still safe for longer.

I'm not sure how high is max SA voltage for the new Intel series. For the 11th gen it was 1.5V+. 1.40-1.45V is the max that will set motherboards when you use memory kits with 4800+ XMP, so I'm sure you can use some more than 1.35V.

 

[Voodoo Rufus]

Thanks, Woomack. That will take a little bit to absorb.

I switched gears and started playing with the iGPU and the core clocks. I managed to get to 5.4GHz all P-core and 4.2GHz E-core so far, but then it went unstable. Max temps around 95C and 205W, which is about the realistic max for my DRP4 heatsink. I did notice that the VRMs run cold on this board, only 45-50C or so. Tons of thermal headroom there. I'll be going to a custom loop when I add my 2080Ti to the machine.

I did some iGPU benchmarks but decided that overclocking it would not be a good use of time, though I think I might be able to squeeze some decent speed out of it. I may return to it for fun.

Tonight I decided to give my Thermal Grizzly lapping tool and contact frame a try. Judging from reflections and flatness, it looks like I still have the knack for lapping. I don't have my notes with me, but I will see if I narrowed the core temperature gap doing this, and see if I drop some degrees.

 

[Woomack]

I went up to 5.7GHz P / 4.5GHz E on my 13700K, but it was at the edge of throttling with the 360 AIO cooler (like hwinfo64 already marked one core as throttling, but the results were not really worse). The same clock was on ASRock and Gigabyte. I'm not really sure what voltages, as hwinfo64 shows VID (1.45-1.55V) and BIOS settings are different from anything else. When I set 1.3V in BIOS, then everywhere is ~1.45 +/- 0.05V. On Gigabyte it's closer to ~1.40V.

 

[Voodoo Rufus]

A few hiccups trying to figure out how to mount my pump, but the best spot ended up being the original spot. I just had to hard mount everything and it fit with no room to spare.

The Phanteks Enthoo Pro M is an older chassis design, best used with air but can do water. AIOs would be easier since you would throw the radiator in the top exhaust area. Doing a front mount is more difficult, and the front bezel is a pain to remove for cleaning.

Currently bubble flushing/leak testing. Loop took about 650mL/3 cups of coolant. Flow rates are good.

Components:
Heatkiller IV CPU and GPU blocks
GTX240 radiator with push/pull Gentle Typhoons
10W DDC pump

I think this should deliver some decent temps and noise levels. Testing will tell. Definitely one of my cleanest loops yet. Finally got to use my fan mounted pump bracket. And I love these eVGA GPU cables. They're so flexible compared to flat ribbon power cables. It's also a bummer that the Z690 Extreme has no water temperature headers to tweak the radiator fan curves with. Being 1150rpm fans, I'll probably just let everything run full speed.

It was never my intent with this build, but I might have to steal this cheap MB/CPU for my main rig. Be a shame to relegate my 9900KS to "backup rig" duty.

 

[Voodoo Rufus]

Got it fired up and playing.

CPU: Stock volts at 5.6GHz was glitchy and would instalock on P95. I went full "you WILL do what I say" and manually set Vcore to 1.35V and P/Ecore PLL to 5 bins. Now P95 is stable, but now I am thermal throttling at 95C at 250W, dropping my core from 5.6GHz down to 5.4GHz and fluctuating. Hmm.....Not that I want it to run any hotter than 95C, but I don't want it choking my clocks. More tweaking to do....

The bad news: lapping the CPU did not narrow the core temperature spread. The good news: I dropped 10C from the DRP4 to the custom loop at the same settings. 85C package on the loop vs. 95C on air. Can't argue with that.

GPU: Running Heaven at 1080P Ultra. 280W GPU and 60W CPU. Top of radiator is 33C, and GPU temp is 44C. 1965Mhz and just cruisin'!

This has me thinking that most people are over-spec'ing their radiators. The GTX240 is a double row, and my fans are doing push/pull at 1200rpm, but it's handling the wattage totally fine.

 

[Voodoo Rufus]

Ah crap. GPU is glitching out, and not due to load. Getting desktop artifacts, driver failure and all sorts of weirdness. Maybe I got some water where it shouldn't have gone when I was flushing the card block.

Rig is sitting by itself with a box fan blowing on it to evap any possible water.

Now has a 500W spotlight to warm it up passively. If this doesn't work, I might have to remove the GPU, remove the block and alcohol bathe it, or cook it.

I'ma be if I bricked my 2080Ti due to negligence.

 

[Voodoo Rufus]

Morning update. I baked it with that 500W light for about 6 hours. The card metal components reached about 120F. This morning, the system seems to be functioning normally again.

 

[Voodoo Rufus]

The abuse/breakin continues.

Some more notes:

Morning test.
GPU running Heaven at 300W, 44-47C. 380W max furmark, 54C (430W total load). +100/1000 core/mem stable. 2040MHz sustained on Heaven. Good enough.

Dropped Vcore to 1.325 and returned PLL to auto. P95 230W. Slight throttling, hovering around 93-95C, dropping to 5.3GHz.
VID dropping to 1.30V.

Raised TJMax to 100C. Dropped Vcore to 1.32V? 1.328V in windows on load. VIDs still dropping to 1.29V on load. Not sure if an issue.
250W and 100C peak, no throttling. 32C on GPU for approximate water temps. Not fully stable.
Raised Vcore back to 1.325V and retesting. Still losing cores on P95.




I gotta say, these kind of temps on a single GTX240 radiator are quite pleasing. Now, I wouldn't want to run P95 and Furmark at the same time, and that's a totally unrealistic load, but I think a 360mm radiator would do just fine with room to spare for any high power gaming build. Unless we're talking those ludicrous RTX4090s on water.....

 

[EarthDog]

Both NV and AMD suggest not to run furmark and call it a 'power virus'. For something like max temps it's OK (incredibly unrealistic load), but if you notice, your gpu clocks are likely a lot lower than what it runs gaming (like closer to base clocks than boost clocks) to compensate. Loop 3dmark fire strike extreme or one of the newer ones. You still reach the gpu max, but do so while running clocks you'll actually run.

 

[Voodoo Rufus]

I know you hate it, but I run it specifically as a thermal load test to see what the loop as a whole can do, not stability.

The OC Scan only gave +50 but much flatter voltage/MHz curve towards the high end, still maxing out around 2075 or so. Manually adjusting I could get to +125 before it would simply crash Heaven. +150 and up was a no-go. All of this was with the GPU well south of 50C.

Fur and Kombustor will send the GPU straight to max power (385W), but looping Heaven it will cruise almost a full 100W lower, with essentially no change in clocks.

I ran the full gamut of 3Dmark tests yesterday to compare against the previous setup (2080Ti and 4770K) and the main rig (3080Ti and 9900KS). It's very interesting to see which tests are more CPU bound, which is Fire Strike and Time Spy lower than 4K. I did a custom run of Time Spy at 8K with max settings, and that did basically break it at <15fps. On the older 3DMark 11 and 06 benchmarks, this machine is faster than my main rig by a fair bit. CPU bound to the point that the 13600K outbenches the 9900KS. I think it's clear that if I want to bust 50K on the Coffee Lake rig, I need to tweak my CPU more (5.2-5.3GHz, some Conductonaut, and abuse the voltage).

I'm more concerned at this point to figure out what settings will stabilize the chip at 5.6GHz. I even lowered it to 5.4GHz and the E-cores to 4.2GHz and I still lose one or two P-cores in P95. But I can run R23 as much as I want without issue. Granted, R23 doesn't load the CPU as hard either. Just bumping the voltage from 1.30V to 1.32V seemed to make it happy in general windows use.

 

[Zerileous]

When you're referring to core temperature spread, are you looking just at P cores, or are you accounting both P and E cores. I believe E cores will run cooler no matter what (it's a feature not a bug). Regardless, that IHS do look purdy.

 

[Voodoo Rufus]

Yes, the spread between P-cores. The E-cores run a fair bit cooler.

 

[rainless]

What kind of temps are you looking at?

 

[Voodoo Rufus]

With 5.6ghz on the p cores, I'm getting low to mid 90s on R23 on a one hour run.