Overclocking RAM on Sabertooth X99

[xkm1948]

It's a 2400 kit

As I said, DDR4 memory IC chips are heavily binned into different speed/timing tiers. Chances are you will never be able to get your kit to be stable at 2800/3000/3200. From what I have gathered from you I would say 2666 with tight timing will be your best bet. RAM vendors charge a premium for high speed high density RAM for a reason. @cadaveca has tested tons of RAM kits he may have better knowledge on memory IC chip quality.

Also this means potentially your 5820K's IMC may be stronger than expected. If you pair your 5820K with some good quality DDR4-3200 14-14-14-34 kits it may just work right out of the box.

 

[basco]

one of the easier ways to get timings down is Dram clock period:
its like setting the chipset to run at lower timings for another speed bin(ex. 2133(=normally level 11) + level9=timings for 1866)
and twcl as low as possible.

on my sabertooth min+max cache ratio is inverted

 

[RejZoR]

The problem with that setting is that they tell me setting one is DRAM 800 and then I have ZERO idea how it progresses forward to level 19. So stupid.

 

[basco]

i dont know exactly too but mine is running 2200mhz c12 with dram clock period of 9.
2400 needs 10 but i have no good imc.
that should give you an idea

 

[TheHunter]

Cpu system agent VCCSA is mostly the key, not sure about VCCIO atm, think I don't have that on my mobo or is that VTT DDR? In my case VTT DDR seems to auto adjust according to higher ram OC.


The most sensitive timings are 3rd timings., but also yield best ram bandwidth. In my case @ 2600MHZ - auto it sets them too loose and I actually get negative bandwidth scaling then at lest say 2400. Also I need DIGI+ ram current at least 110-120% (max goes to 130%).

Ok DDR3 is a bit different, but adv timings are somewhat the same if Im not mistaken.

Some of my testing adventures with 3rd timings
https://www.techpowerup.com/forums/threads/intel-haswell-overclocking-clubhouse.185344/page-59

 

[RejZoR]

Then how can you explain this:

2400 MHz - Bootable and stable (advertised RAM speed)
2666 MHz - Bootable and stable
2800 MHz - Unbootable under any voltage or timing configuration
2933 MHz - Unbootable under any voltage or timing configuration
3000 MHz - Unbootable under any voltage or timing configuration
3200 MHz - Bootable, but unstable even under elevated voltages. Not crashing per se, but ASUS RealBench keeps reporting that it's files got corrupted during 10 loops of the Heavy Multitasking test.

By what logic are those 3 frequencies entirely unbootable no matter what, but 3200 MHz is almost stable. It just goes against all logic.

 

[eidairaman1]

Then how can you explain this:

2400 MHz - Bootable and stable (advertised RAM speed)
2666 MHz - Bootable and stable
2800 MHz - Unbootable under any voltage or timing configuration
2933 MHz - Unbootable under any voltage or timing configuration
3000 MHz - Unbootable under any voltage or timing configuration
3200 MHz - Bootable, but unstable even under elevated voltages. Not crashing per se, but ASUS RealBench keeps reporting that it's files got corrupted during 10 loops of the Heavy Multitasking test.

By what logic are those 3 frequencies entirely unbootable no matter what, but 3200 MHz is almost stable. It just goes against all logic.

Spd tables/xmp profiles are confused.

 

[cadaveca]

Then how can you explain this:

2400 MHz - Bootable and stable (advertised RAM speed)
2666 MHz - Bootable and stable
2800 MHz - Unbootable under any voltage or timing configuration
2933 MHz - Unbootable under any voltage or timing configuration
3000 MHz - Unbootable under any voltage or timing configuration
3200 MHz - Bootable, but unstable even under elevated voltages. Not crashing per se, but ASUS RealBench keeps reporting that it's files got corrupted during 10 loops of the Heavy Multitasking test.

By what logic are those 3 frequencies entirely unbootable no matter what, but 3200 MHz is almost stable. It just goes against all logic.

Yep, this is normal. It's a memory hole and those speeds require BCLK OC on some board/mem/CPU combos. Most memory kits of those speeds (other than 2800) will automagically use BCLK OC when you enable XMP profile. I tested quite a few 3000 MHz kits, all of them set 125 BCLK.

 

[xkm1948]

Then how can you explain this:

2400 MHz - Bootable and stable (advertised RAM speed)
2666 MHz - Bootable and stable
2800 MHz - Unbootable under any voltage or timing configuration
2933 MHz - Unbootable under any voltage or timing configuration
3000 MHz - Unbootable under any voltage or timing configuration
3200 MHz - Bootable, but unstable even under elevated voltages. Not crashing per se, but ASUS RealBench keeps reporting that it's files got corrupted during 10 loops of the Heavy Multitasking test.

By what logic are those 3 frequencies entirely unbootable no matter what, but 3200 MHz is almost stable. It just goes against all logic.

From what I have learned tweaking my Sabertooth X99, HWE IMC really hates 2800/3000. It is a lot easier on the IMC to work on 3200. It has something to do with 100BCLK. Since both 2800 and 3000 requires 125BCLK while 3200 and 2666 works on 100BCLK. It is definitely something to do with the HWE IMC.

If I were you I'd spend sometime reading the RVE RAM tweaking guide over ROG forum, it helped me a lot while I was still using ADATA RAM@2800. Higher VCCSA is not always good, there is a sweet spot for each and every combination of CPU and RAM. Just going up or down in 0.05V increments for VCCSA until you can find one fairly stable.

For primary timing, loosen it to 17-18-18-35 at least. Depends on the RAM IC of your kit. Samsung IC overclocks a lot better than Hynix IC.

For UEFI settings, go to External Digi power control. Set DRAM Phase Control to "Extreme", Set DRAM current capacity to 140%. For DRAM Eventual Voltage enter something like 1.35~1.37V. For the DRAM voltage in the main AITweaker tab enter 1.37~1.38V. The higher DRAM voltage help passing DRAM training during post while maintain a working voltage in Windows(ie, "eventual")

Also, adjusting CPU power phase to either Optimized or Extreme also helps stability for higher RAM speed. At least that was the case for me. Using "Standard" I can't get my DDR4-3000 to even post.


So far 3505 UEFI is the best for Sabertooth X99. Steady RAM/L3 Cache performance increase over 3402. Boot up time has decreased drastically(now takes less than 5 seconds! It used to take ~25 seconds cold booting) And now it seems I need less vcore to keep my 4.25GHz OC stable. It used to be ~1.2V. Now it is about 1.18V. Lower voltage cooler CPU. Looks like improvement to me.

 

[RejZoR]

The thing is, I could easier stabilize 2800 or 3000 RAM. But 3200 is already so high I can't stabilize it no matter what. Even when it looks great, RealBench starts whining in around 8th loop about some corruption...

 

[cadaveca]

The thing is, I could easier stabilize 2800 or 3000 RAM. But 3200 is already so high I can't stabilize it no matter what. Even when it looks great, RealBench starts whining in around 8th loop about some corruption...

The Sabertooth isn't exactly a hardcore memory-clocking board though, so you are limited by what the BOIS offers. If you have the RAmpage board, you might have better luck. I am using an MSI X99 GAMING PRO CARBON, and both my 5930K and my 6950X can run all memory dividers without any issues on 100 MHz BCLK. You can see that in some of the memory reviews; 3000 MHz sticks on 100 BCLK, no problem. X99 did not offer many memory multipliers at launch, but with successive BIOS updates, the other dividers started to work, and now are rock-stable for me. (keeping in mind I'm using sticks rated at those higher speeds or higher)

 

[jboydgolfer]

but somehow I managed to OC my 2400 MHz (15-15-15-35-2T) RAM to 2666 MHz like 2 days ago

did you happen to change the DRAM reference clock from 100 to 133? that would account for the 266Mhz increase? i know this is an older post, but i happened upon this one, and thought id see if you found the reason for the jump in Mhz.

 

[xkm1948]

The Sabertooth isn't exactly a hardcore memory-clocking board though, so you are limited by what the BOIS offers. If you have the RAmpage board, you might have better luck. I am using an MSI X99 GAMING PRO CARBON, and both my 5930K and my 6950X can run all memory dividers without any issues on 100 MHz BCLK. You can see that in some of the memory reviews; 3000 MHz sticks on 100 BCLK, no problem. X99 did not offer many memory multipliers at launch, but with successive BIOS updates, the other dividers started to work, and now are rock-stable for me. (keeping in mind I'm using sticks rated at those higher speeds or higher)

I concur that. Sabertooth X99 is not the champion overclocker. It is weaker comparing to both X99E-WS and Rampage5Edition10. Mostly it is just the 5 yrs warranty that is attractive. However on the other hand I never had any ASUS MoBo die on me within warranty.

 

[RejZoR]

So, 2800, 2933 or 3000 can only be done with BCLK 125 (just realized the XMP stuff you guys were mentioning). That kinda sucks since that overclocks everything. Might still give it a try for the lolz.

Wasn't expecting this to be an issue since Sabertooth X99 kinda falls into the upper range, but they apparently spent more time on ruggidnes than OC and all that. It's still a decent overclocker, just not extreme one. I guess this might also be the reason why CPU clocks to 4.5GHz easily with 1.125V and then needs 1.3V for 4.6GHz and even then stability is questionable. Might be some trickery behind the scenes causing this...

 

[basco]

i had not taken the sabertooth but i got it cheap-and really the 3505 bios changes a lot of i was not satisfied.
with samsung mem i have a hole too in 2800 2933 but 3200 boots but i did not try to get it stable
now even 2400mhz with dram clock 9 is possible before 10 was a must.
lower oc voltages like xkm1948 is seeing
this is the bios i was waiting for from beginning and for that price this mobo was sold it should have been from start(and asus was\is really good at bios\sabertooth was my 1st not satisfied\my cents).

and i think that you should really try: haswell sfr adjust and internal pll overvoltage if you go over 4,5ghz and can´t get it stable because i think both help for higher oc´s on some cpu´s and some not.
i am running 4,0 so no need

 

[xkm1948]

On a completely different note. I find it pretty funny that it took ASUS almost 2 years to iron out most bugs and improve performance on X99 platform. Yet we see people crying over how X370 is not a mature enough system. Double standards much huh?

 

[Tatty_Two]

On a completely different note. I find it pretty funny that it took ASUS almost 2 years to iron out most bugs and improve performance on X99 platform. Yet we see people crying over how X370 is not a mature enough system. Double standards much huh?

Good point, however you are comparing one board manufacturer on the chipset unlike X370.

 

[erixx]

Not really imo. I've manually increased those before and none of the clocks between 2800 and 3000 worked. I was using really elevated voltages on all subsystems that mater for RAM stability and it just didn't even post. After switching to 3200MHz directly, it booted up fine. Gonna check the voltages now...

EDIT:
CPU Agent: 1.16V
VCCIO CPU: 1.4V (on Auto), have tuned it down to 1.15V because 1.4V is red colored if inserted manually and that's just WAY too high. I only use yellow marked voltages. System boots, but have no idea about stability yet...

strange. My CPU SA is 1.216 and VCCIO is 1.048 (both on AUTO). RAM at 3200 and CPU at 4200Mhz (RV10 board)

 

[RejZoR]

On a completely different note. I find it pretty funny that it took ASUS almost 2 years to iron out most bugs and improve performance on X99 platform. Yet we see people crying over how X370 is not a mature enough system. Double standards much huh?

If they'll provide steady support for X370, I don't see any problems. And ASUS is known to support high end stuff for very long. I was getting tons of updates for Rampage II Gene, I get tons of updates for their router and Sabertooth X99 is getting tons of updates as well.

 

[xkm1948]

At this point I feel we should just use our HWE as it is. Once the AMD HEDT platform drops this summer, I bet a lot of 5960X/6950X users will be going for RyZen HEDT with 16 core 32 threads. Then we @RejZoR can snatch up those 5960X/6950X fairly cheap.

 

[RejZoR]

Turns out 125 MHz strap doesn't work well. Half the time it just freezes on POST screen or cycles so I can't even recover it easily. I guess I'll have to wait for more BIOS updates... But I'm kinda happy with what I have at the moment. My system is overclocked as fa as it can go 100% stable and yet not a single voltage parameter is even colored yellow in BIOS. The CPU at 4.5 GHz actually runs at lower voltage than stock lol XD RAM, based on tests has the highest gain at 2666MHz so that's actually also fine. The DRAM CLK Clocks setting is at 8 with timings 13-13-13-32-1T using just 1.28V. Any lower and it doesn't even boot no matter the voltages I use on anything, so it's very tightly set as it is.

If I'll ever be upgrading, it'll be Broadwell-E class for tiny IPC gain over Hasswell-E and 14nm process. Intel kinda screwed us over with Skylake-X being on X299...

 

[basco]

wow dram clock at 8 with 2666mhz c13 is not reachable with my imc- nice work

 

[xkm1948]

Tried DDR4-3200 again today on 3505 BIOS. Major fail. I forgot to change DRAM Command Rate back to 2T from 1T. Booting up and the system showing DRAM taking 1.5V!! Holy cow that scared the crap out of me. Power down right away.

So I guess I simply can't have all 8 slots of RAM filled and a cozy 100BCLK OC.

 

[RejZoR]

In my case Auto doesn't kick in the voltage. It'll just fail to boot.

 

[RejZoR]

Just one quick question about RAM. Again.

The main timings have 5 values, standard 4 and the command rate.

Currently at 13-13-13-32 with command rate of 1T.

I'm wondering about the 4th number 13-13-13-32. In general, it's a sum of first three numbers, but sometimes it's lower. Is there any general rule of how low it's good to go? I've tried it and it goes all the way down to 13, making it 13-13-13-13 which just doesn't make sense. But it worked lol