- last edited on by
ROGBot
[I will include/update my system specs immediately within the next few hours after submitting this posting]
Hi folks,
Just a little over a month ago, I started overclocking my new hardware setup. CPU Overclocking went relatively fine so far (though the voltage necessary to achieve stability is a little too much...) ; right now, I'm running the CPU @ 4.8/4.8/4.5 GHz (Core/AVX/Cache) with 1.35V of manual fixed voltage with LLC Level 5 - and it seems, that there is still a little headroom left for going higher but that isn't my priority right now.
My actual priority is Memory Overclocking and that didn't go nice at all since the very beginning. Initially, I was optimistic - my memory kit seemed to readily pack DDR4-4000 @ CL16-16-16-32 (or 40), maybe even @ CR1, but I don't remember anymore what voltages I had to give or whether it was CR2 or indeed CR1. However, soon reality kicked in, and I realized that the memory is unstable at those speeds and/or latencies and especially at CR1. But much more demoralizing is the fact, that I need very high voltages to stabilize the memory @ DDR4-4000 CL16 even without CR1 (@ CR2). More on that later.
CR1 works fine and absolutely flawless at any CPU/Cache frequency at the default XMP speed (DDR4-3200) at the rated timings (CL14) but with tRAS lowered to 28 (14-14-14-28) and some secondary timings enhanced.
At DDR4-3600, I already start to run into problems with CR1 @ CL15-15-15-30: I need significantly more or much more voltages (VDIMM, VCCIO, VCCSA) just to get the memory to POST or boot into Windows! More importantly, it seems that no matter how high I crank up the voltages, the memory just isn't STABLE! Sometimes during booting Windows, sometimes within a few minutes after booting into Windows, sometimes after a couple of hours playing Arma 3 (which takes much advantage of higher RAM speeds, respectively a lower memory latency), sometimes after many hours of playing or using Windows, the system will CRASH! Sometimes (more often than occasional) the system won't even POST after a restart and needs a couple of loops (memory training???) before successfully POSTing or booting Windows again. It just isn't stable.
For CR2, I need a VDIMM of "only" 1.4V along with VCCIO 1.1V and VCCSA 1.15V for DDR4-3600 at the same latencies (CL15-15-15-30).
For DDR4-4000 @ CR2 @ CL16-16-16-32 (or 40), I need at least VDIMM 1.55V, VCCIO 1.175V and VCCSA 1.225V to get the system *barely* stable - and this isn't even CR1, this is CR2! Like with DDR4-3600 CR1, the system is never fully stable at those speeds, despite CR2 - it will likely crash on the next day or within a few hours.
However, at least for DDR4-4000 (CR2, 16-16-16-32), it seems that I'm very close to achieve full/permament stability. The situation definitely improved - I can now sometimes play for many hours (I tested Arma 3 only, though) or even use my system 1-2 days before it finally crashes. But when it crashes, the next crash very soon follows if I keep staying at that setting - that's why I often revert to default XMP setting (DDR4-3200 with above-mentioned timings and default voltages) in order to give the system time to "recover" before trying again...
Not even minding stability, the system won't even POST when I try to go higher - if I either increase frequency (I tried DDR4-4266 CL16 with a 100:133 strap, CR2) or try to lower latency (going from CL16 down to CL15 for DDR4-4000, CR2). No matter what insane voltages I give (I gave up to 1.675V VDIMM, 1.30V VCCIO and 1.30V VCCSA in both cases), the system endlessly loops trying to POST without success - after the 5th failed loop, ASUS Safe Mode kicks in and I try to adjust voltages and/or settings without leading to success...
My end goal is either DDR4-4000 CL16 @ CR1 or DDR4-4000 CL15 @ CR2 or alternatively DDR4-4266 CL16 @ CR2 - in the latter two variants, the absolute latency is the same... 7.5 nanoseconds. So, either 8 nanoseconds @ CR1 or 7.5 nanoseconds @ CR2. And 4000MHz frequency as minimum.
My impressions:
- I often have the feeling, that I'm running into a wall at certain settings - no matter how much voltage I give, it just doesn't work. Which leads to Point #2 (next point)...
- When hitting "the wall", very often the possibility comes into my mind, that not voltage but something else is the bottleneck - most likely other memory-related BIOS settings like DRAM Ref Voltage Control, RTL-IOL Control, Skew Control, Memory Training, MRC Fast Boot & MCH Full Check, enabling or disabling unused DIMM slots, DRAM Clock Period, other more "exotic" voltages (f.e. "DRAM VTT Voltage" or "VPPDDR Voltage" and others - those two are listed in "Tweaker's Paradise" menu) or even raising Core/Cache voltage!
- I get the impression, that increasing main memory voltages (VDIMM, VCCIO, VCCSA) does *not* prevent a crash (hence, bring out stability) but only POSTPONES it - the higher the voltage, the longer it takes before the crash finally occurs!
- Can increasing Core/Cache voltage aid in achieving stability? I ask this because since Skylake, core voltage and cache voltage are linked. If the IMC of the CPU is even slightly affected by Cache/Uncore overclocking or the Cache is even slightly affected by memory overclocking (higher frequency and/or tighter timings), then increasing Core/Cache voltage could do the trick.
- Memory temperatures were never a subject of the problem. I repeatedly checked AIDA64 logs and even after hours of stressful runs (high memory frequencies, aggressive timings, very high memory voltages, Uncore overclocked @ 4.5 GHz), both DIMMs never even reached 45°C ! So, if AIDA64's measurements are indeed correct, temperature as a cause for stability problems can likely be excluded...
Here is an exerpt (not complete list!) of the BIOS settings I picked up in my BIOS menu, whom I think I could experiment with to fix stability issues or achieve higher frequencies/lower timings:
Voltages:
- "PLL Termination Voltage"
- "DRAM VTT Voltage" (Tweaker's Paradise)
- "VPPDDR Voltage" (Tweaker's Paradise)
- "DMI Voltage" (Tweaker's Paradise)
- "Core PLL Voltage" (Tweaker's Paradise)
- "Internal PLL Voltage" (Tweaker's Paradise)
- "Ring PLL Volltage" (Tweaker's Paradise)
- "System Agent PLL Voltage" (Tweaker's Paradise)
- "Memory Controller PLL Voltage" (Tweaker's Paradise)
- "PLL Bandwidth" (Tweaker's Paradise)
- "Eventual DRAM Voltage" (Tweaker's Paradise)
- "Eventual PLL Termination Voltage" (Tweaker's Paradise)
- "Eventual DMI Voltage" (Tweaker's Paradise)
Settings:
- "DRAM Write Recovery Time"
- "DRAM Write Latency"
- "DRAM CLK Period"
- "DLLBwEn"
- "Training Profile"
- "XTU Setting"
Sections:
- "Memory Training Algorithms"
- "DRAM Ref Voltage Control"
- "RTL-IOL Control"
- "Skew Control"
This is the current BIOS configuration I'm barely stable with @ DDR4-4000 CL16:
Looking forward for any walkthrough...
T.S.O.M.
Hi folks,
Just a little over a month ago, I started overclocking my new hardware setup. CPU Overclocking went relatively fine so far (though the voltage necessary to achieve stability is a little too much...) ; right now, I'm running the CPU @ 4.8/4.8/4.5 GHz (Core/AVX/Cache) with 1.35V of manual fixed voltage with LLC Level 5 - and it seems, that there is still a little headroom left for going higher but that isn't my priority right now.
My actual priority is Memory Overclocking and that didn't go nice at all since the very beginning. Initially, I was optimistic - my memory kit seemed to readily pack DDR4-4000 @ CL16-16-16-32 (or 40), maybe even @ CR1, but I don't remember anymore what voltages I had to give or whether it was CR2 or indeed CR1. However, soon reality kicked in, and I realized that the memory is unstable at those speeds and/or latencies and especially at CR1. But much more demoralizing is the fact, that I need very high voltages to stabilize the memory @ DDR4-4000 CL16 even without CR1 (@ CR2). More on that later.
CR1 works fine and absolutely flawless at any CPU/Cache frequency at the default XMP speed (DDR4-3200) at the rated timings (CL14) but with tRAS lowered to 28 (14-14-14-28) and some secondary timings enhanced.
At DDR4-3600, I already start to run into problems with CR1 @ CL15-15-15-30: I need significantly more or much more voltages (VDIMM, VCCIO, VCCSA) just to get the memory to POST or boot into Windows! More importantly, it seems that no matter how high I crank up the voltages, the memory just isn't STABLE! Sometimes during booting Windows, sometimes within a few minutes after booting into Windows, sometimes after a couple of hours playing Arma 3 (which takes much advantage of higher RAM speeds, respectively a lower memory latency), sometimes after many hours of playing or using Windows, the system will CRASH! Sometimes (more often than occasional) the system won't even POST after a restart and needs a couple of loops (memory training???) before successfully POSTing or booting Windows again. It just isn't stable.
For CR2, I need a VDIMM of "only" 1.4V along with VCCIO 1.1V and VCCSA 1.15V for DDR4-3600 at the same latencies (CL15-15-15-30).
For DDR4-4000 @ CR2 @ CL16-16-16-32 (or 40), I need at least VDIMM 1.55V, VCCIO 1.175V and VCCSA 1.225V to get the system *barely* stable - and this isn't even CR1, this is CR2! Like with DDR4-3600 CR1, the system is never fully stable at those speeds, despite CR2 - it will likely crash on the next day or within a few hours.
However, at least for DDR4-4000 (CR2, 16-16-16-32), it seems that I'm very close to achieve full/permament stability. The situation definitely improved - I can now sometimes play for many hours (I tested Arma 3 only, though) or even use my system 1-2 days before it finally crashes. But when it crashes, the next crash very soon follows if I keep staying at that setting - that's why I often revert to default XMP setting (DDR4-3200 with above-mentioned timings and default voltages) in order to give the system time to "recover" before trying again...
Not even minding stability, the system won't even POST when I try to go higher - if I either increase frequency (I tried DDR4-4266 CL16 with a 100:133 strap, CR2) or try to lower latency (going from CL16 down to CL15 for DDR4-4000, CR2). No matter what insane voltages I give (I gave up to 1.675V VDIMM, 1.30V VCCIO and 1.30V VCCSA in both cases), the system endlessly loops trying to POST without success - after the 5th failed loop, ASUS Safe Mode kicks in and I try to adjust voltages and/or settings without leading to success...
My end goal is either DDR4-4000 CL16 @ CR1 or DDR4-4000 CL15 @ CR2 or alternatively DDR4-4266 CL16 @ CR2 - in the latter two variants, the absolute latency is the same... 7.5 nanoseconds. So, either 8 nanoseconds @ CR1 or 7.5 nanoseconds @ CR2. And 4000MHz frequency as minimum.
My impressions:
- I often have the feeling, that I'm running into a wall at certain settings - no matter how much voltage I give, it just doesn't work. Which leads to Point #2 (next point)...
- When hitting "the wall", very often the possibility comes into my mind, that not voltage but something else is the bottleneck - most likely other memory-related BIOS settings like DRAM Ref Voltage Control, RTL-IOL Control, Skew Control, Memory Training, MRC Fast Boot & MCH Full Check, enabling or disabling unused DIMM slots, DRAM Clock Period, other more "exotic" voltages (f.e. "DRAM VTT Voltage" or "VPPDDR Voltage" and others - those two are listed in "Tweaker's Paradise" menu) or even raising Core/Cache voltage!
- I get the impression, that increasing main memory voltages (VDIMM, VCCIO, VCCSA) does *not* prevent a crash (hence, bring out stability) but only POSTPONES it - the higher the voltage, the longer it takes before the crash finally occurs!
- Can increasing Core/Cache voltage aid in achieving stability? I ask this because since Skylake, core voltage and cache voltage are linked. If the IMC of the CPU is even slightly affected by Cache/Uncore overclocking or the Cache is even slightly affected by memory overclocking (higher frequency and/or tighter timings), then increasing Core/Cache voltage could do the trick.
- Memory temperatures were never a subject of the problem. I repeatedly checked AIDA64 logs and even after hours of stressful runs (high memory frequencies, aggressive timings, very high memory voltages, Uncore overclocked @ 4.5 GHz), both DIMMs never even reached 45°C ! So, if AIDA64's measurements are indeed correct, temperature as a cause for stability problems can likely be excluded...
Here is an exerpt (not complete list!) of the BIOS settings I picked up in my BIOS menu, whom I think I could experiment with to fix stability issues or achieve higher frequencies/lower timings:
Voltages:
- "PLL Termination Voltage"
- "DRAM VTT Voltage" (Tweaker's Paradise)
- "VPPDDR Voltage" (Tweaker's Paradise)
- "DMI Voltage" (Tweaker's Paradise)
- "Core PLL Voltage" (Tweaker's Paradise)
- "Internal PLL Voltage" (Tweaker's Paradise)
- "Ring PLL Volltage" (Tweaker's Paradise)
- "System Agent PLL Voltage" (Tweaker's Paradise)
- "Memory Controller PLL Voltage" (Tweaker's Paradise)
- "PLL Bandwidth" (Tweaker's Paradise)
- "Eventual DRAM Voltage" (Tweaker's Paradise)
- "Eventual PLL Termination Voltage" (Tweaker's Paradise)
- "Eventual DMI Voltage" (Tweaker's Paradise)
Settings:
- "DRAM Write Recovery Time"
- "DRAM Write Latency"
- "DRAM CLK Period"
- "DLLBwEn"
- "Training Profile"
- "XTU Setting"
Sections:
- "Memory Training Algorithms"
- "DRAM Ref Voltage Control"
- "RTL-IOL Control"
- "Skew Control"
This is the current BIOS configuration I'm barely stable with @ DDR4-4000 CL16:
Looking forward for any walkthrough...
T.S.O.M.
Hi and welcome to the forum. I hope the following a) gets access to post and b) helps.
It would be useful to know the model of motherboard, CPU and the DDR4 RAM - at least the XMP specs of the RAM. Without those, I can only speak in generalities. The tag says b-die, which helps a little with the guessing.
Vccsa and Vccio are normally in the 1.25 to 1.35 volt range for overclocking RAM in the 4000s. Try to have Vccsa around 50mv higher than Vccio.
DRAM VTT is a sensing and terminating voltage on the bus lines. It should be at half of Vdimm, which is what the AUTO setting provides.
Vppddr is nominally 2.50 volts, which seems strange at first but it's real.
As you have seen, Vdimm is the problem. Too low won't let the RAM clock fast. Too high Vdimm is a cause of instability. I think a voltage between 1.50 and 1.60 is the max for reasonable stability, depending on the particular RAM kit, motherboard and CPU. Here is the order I would try for tuning toward your desired profile:
Vdimm 1.60, Vccsa 1.30, Vccio 1.25, RAM clock 4000, timings: CR 1, RCD/RP 19, RAS 36 and CL in steps 14, 15, 16, 17. My bet is on 15, but 14 might make it.
If a particular CL setting posts, boots and is almost stable in windows, try adjusting Vdimm up or down in steps of 20mv. Small voltage change can make that last bit of difference.
When you get a stable CL, try RCD/RP in steps: 14, 15, 16, 17... Before adjusting RAS, try RRD_S and _L at 4, 5 or 6. RAS will probably work at 28 after that.
It would be useful to know the model of motherboard, CPU and the DDR4 RAM - at least the XMP specs of the RAM. Without those, I can only speak in generalities. The tag says b-die, which helps a little with the guessing.
Vccsa and Vccio are normally in the 1.25 to 1.35 volt range for overclocking RAM in the 4000s. Try to have Vccsa around 50mv higher than Vccio.
DRAM VTT is a sensing and terminating voltage on the bus lines. It should be at half of Vdimm, which is what the AUTO setting provides.
Vppddr is nominally 2.50 volts, which seems strange at first but it's real.
As you have seen, Vdimm is the problem. Too low won't let the RAM clock fast. Too high Vdimm is a cause of instability. I think a voltage between 1.50 and 1.60 is the max for reasonable stability, depending on the particular RAM kit, motherboard and CPU. Here is the order I would try for tuning toward your desired profile:
Vdimm 1.60, Vccsa 1.30, Vccio 1.25, RAM clock 4000, timings: CR 1, RCD/RP 19, RAS 36 and CL in steps 14, 15, 16, 17. My bet is on 15, but 14 might make it.
If a particular CL setting posts, boots and is almost stable in windows, try adjusting Vdimm up or down in steps of 20mv. Small voltage change can make that last bit of difference.
When you get a stable CL, try RCD/RP in steps: 14, 15, 16, 17... Before adjusting RAS, try RRD_S and _L at 4, 5 or 6. RAS will probably work at 28 after that.
