3866 MHz-Too much is enough.

[Alaric]

I quit at 1.500v, even though I saw no change in temps at that voltage. I settled at 1.400v (1.392v actual) with the speed/timings below. For my uses the stock DDR4 3000 CL14 was overkill. Current settings seem like the perfect amount of too much. LOL

 

[UltraTaco]

Yeah, you're definitely up there, man.

 

[JLK03F150]

Impressive!

 

[Alaric]

I got the tRFC down to 535 clocks after I posted, because leaving well enough alone apparently wasn't an option for me. LOL It's my first (successful) attempt at OC memory, and all the credit goes to G.Skill and Woomack. His OC thread with the Trident Z is a how-to even I couldn't screw up. http://www.overclockers.com/forums/...nt-Z-2x8GB-DDR4-3200-CL14-F4-3200C14D-16GTZKW

 

[JLK03F150]

Have you played with the Aida64 memory benchmark (does it even work w/ trial ver.) to see the difference in bandwidth? That's a really nice increase in clock speed and you still have a low CL.

My primary PCs are still DDR3 and have very mild overclocks. One day I'll get to play with DDR4, but everyone will have forgotten what they did "back in the day" when they had DDR4. lol

 

[Mandrake4565]

My primary PCs are still DDR3 and have very mild overclocks. One day I'll get to play with DDR4, but everyone will have forgotten what they did "back in the day" when they had DDR4. lol

Funny you say that JLK, I was messing with my 4770k the other day for the first time in at least a year and a half. Had my 2200 G.Skill Pi's in it and my brain was going into overload dropping the primary timings into single digits.

And nice work Alaric

 

[Alaric]

You should check out some of Woomack's numbers in the link. How about 4000 MHz at 12-12-12-28-220! And 4133 MHz at 17-17-17-38 on 1.4v! Anyone with a set of the 3000+ C14 or C15 Trident Z should go for it. Now all I have4 to do is figure out why the Ripjaws in the Bulldozer rig won't go over 1600 MHz no matter where I set the voltage or timings. Grrr!

 

[Mandrake4565]

Now all I have4 to do is figure out why the Ripjaws in the Bulldozer rig won't go over 1600 MHz no matter where I set the voltage or timings. Grrr!

Cpu Nb voltage is your friend bud! I need 1.3875 v on mine to run 1866 Mhz. Been running it that way since I bought it in 2012.

 

[Alaric]

Cpu Nb voltage is your friend bud! I need 1.3875 v on mine to run 1866 Mhz. Been running it that way since I bought it in 2012.

Ah ha! I will definitely do that. Many thanks. I know the FX chips have a weak IMC, but after a year+ with an intel rig it seems my brain threw out the AMD box in the corner. LOL

 

[Woomack]

Nice result for Z170 and non-ROG/OCF mobo
Above that you will need to bump voltage above 1.6 or relax timings. On the other hand I don't think you will make much more than 4000 while performance will be worse.
tRFC can go even lower but it won't help in performance out of something like SuperPi.

 

[Alaric]

I forgot to ask about cache speed. Would there be any benefit to running it at x43 with the RAM at 3866 MHz? Default is 4000 MHz on the uncore but it does 4300 MHz easily enough.

 

[Woomack]

Check it in some tests. On dual channel Intel chipsets usually cache is not helping much as it's already fast. On quad channel and AMD, cache is usually limiting memory performance. I doubt you will see any difference in daily work but I guess you will see a bit better performance in some benchmarks.

 

[Alaric]

That was my initial thought, that the cache is already running faster than my RAM, and Skylake has a large cache to start with. Would quantity of RAM, say 32 GB, show more benefit from raising cache speed? Or is the bandwidth just not sufficient in dual channel to make much difference?
Total RAM noob here, so if these are dumb questions I have an excuse.

 

[Woomack]

There are no benefits of larger RAM if you can't use that RAM. If you use 16GB but have 32GB in total then it's like 16GB is not doing anything. Marketing is pushing higher RAM capacity while it's good almost only for professional work.
Cache is only a link between CPU and RAM. Regardless how much RAM you have, it will help about the same. Usually it's also = speed of memory controller. On quad channel platforms overclocking cache let you to use higher RAM bandwidth. Actually when you check your RAM bandwidth and maximum theoretical ( AIDA64 is showing both ) then on 2 channel boards it won't be that far from each other. On quad channel it's much bigger difference which can be reduced by overclocking cache. It's still not possible to be close to maximum theoretical bandwidth.

 

[UltraTaco]

Got a question if I may...
I used to run triple channel ram on my x58(i7 920), but recently switched to dual channel. Would I benefit from running faster cache than 2x the speed of ram?(2x is minimum allowed on that platform)

Running
CPU 3.67ghz(175x21),
ram 1400(175x8),
cache 3150(175x18). <--- more than 2x ram

 

[Woomack]

On X58 it won't help much and you have to keep it close to x2 for best stability. However, you can try something like 190-200 x18 with memory at ~1600 and then all will go up.

 

[UltraTaco]

Well, I keep running into all sorts of stability crasher issues when I go into 193+ bclk, so I'll have to stay away from that. Even tried playing with ich, oih, pll, voltages to no avail. Probably needs expert tweaking.

I decided to drop out if turbo and up bclk to 184. It gives me this. I just decided to run with HT enabled and any higher could speed, it would throttle because of temps, so I'm trying to optimize stuff around this frequency. I'll go into heavier iverclocks when Im ready to upgrade, but for now this will have to do. Here's what I have now. This is as far as I can go on cache speed without sharp climb in vtt voltage

View attachment 195951

Sorry for derailments. I stop now.

Ps-temps below 70 during stress, but only because cold in the room. Summer it'll probably get higher.

Ps#2-i remember someone mentioning, I think it was Alaric, that increasing ram speed didn't do much performance, but increasing cache speed gave some improvements, hence this new quest I'm on. But it was on his plaform, so mine probably wouldn't benefit much. Although I do get a small bump like 2-3fps in for in fSX, which is great how, considering it runs in 30's horribly optimized

 

[Alaric]

Nope, I was just asking if increasing cache speed would be a plus with increasing RAM speed on the Intel DDR4 platform, but it doesn't seem to do anything worthwhile. AMD platforms are more grateful for cache tweaking and tighter timings than the Intel DDR4 platforms, though.


I was poking around G.Skill's site and found that the fastest kit they have on their QVL for a Giga Z170 mobo is 3466 MHz@16-18-18-38. Their faster 8GBx2 Trident Z kits are only on Z270, Z370, and X299 boards (for Gigabyte boards). My board is on their list for all of them up to the 3466 MHz kit, though.
So, anything I get over 3466 MHz is punching above my board's weight class and that makes me a happy camper.
My 3866 16-16-16-34 has proven to be not quite stable shock. So I'm going to try creeping up on it and see if moar volts and looser timings gets me there. If I get stable that way, I'll tighten the timings as much as I can and work the volts down after that. VCCIO is 1.15v and System Agent voltage at 1.1v, so I'll take them to 1.2v each. I figure I'll work towards the best stable speed I can get under the ceiling and check my temps, and work backwards if needed for thermals.
Anybody know how many volts is required to let the Magic Smoke out of my (now very expensive) DDR4s?

First I need to clean up pages of notebook paper spread from Hell to breakfast and get my notes organized. Having the info in my rig doesn't help when it won't boot. LOL

 

[Alaric]

Did a short (5 min.) memory stress test in XTU, and I'm a little surprised. No problems with the RAM, with 1.440v the max temp was only 39C and 38C for the pair. What I didn't expect was 100% across all 8 threads on the CPU for 5 minutes. On the upside, the highest temp there was only 75C, so my AIO is good for the OC I'm running. LOL.



3733 MHz with fairly loose timings is good so far. I'll bump it up one strap and loosen the timings a little more, and keep the volts at 1.440v.

 

[Alaric]

And an update! I'm using the Intel XTU stress tester until a setting fails, then I'll back off one or two steps and memtest it for a few hours to check for stability. So far I got these numbers through the XTU test. With the case fans turned up higher the RAM temps never went over 36C! The overvolt doesn't seem to be doing any thermal damage so far. AIO is still in the game, too. Think I'll try 4000 MHz next.