[Ramad]

Quote:

Originally Posted by 1usmus

The higher the temperature of the RAM the more voltage is required. Cool your memory to 25 degrees and they will run at 1.28. The version of the bios has nothing to do with it

3466 on 1.32

I will try to publish more screenshots later

I don't think you understood my post to elmor, it's not about heat or RAM voltage.
The picture you posted is of an ASUS Prime running RAM at a higher frequency, what does it have to do with C6H and BIOS 3501?

 

[Gettz8488]

Any insight would be helpful hear I'm pstate overclocked with offset voltage -0.01875 everything has been fine until the new bios update looking at hwinfo i'm not seeing my SVI2 Voltage downvolt at all staying at 1.31 would see it downvolt. i no longer see it can it be happening so fast i don't notice it? is there another way to check if its downvolting?

 

[Ne01 OnnA]

Quote:

Originally Posted by wisepds

How???? Arrrg!!!! i want it too!!!

Here with Foobar Music on SB-Zx dedicated ASIOx64 Driver

Here download also Timer Resolution (for easy adjust 500uS for gaiming etc:
https://mega.nz/#!oBE00I4I!TqVFQ-to41-syHgUQ77m9mcH0FmkUYr90sbaxINBQNg

 

[harrysun]

Finally I got F4-3200C14D-32GTZ 2x16GB dual-rank (DR/2R) Samsung B-die @3333MT/s CL14 "stable" with BIOS 1701. As always, my current stable setup can be found in the signature!

How stable is this setup:

Spoiler

*) T_Sensor1 = Ambient temperature in the case

It's time to share my learnings:

1. Start with Ryzen DRAM Calculator (credits to @1usmus)
2. Use paper and pen for your results first, until you have found the right RAM timings, DRAM Voltage + VTTDDR Voltage, VDDSOC, ProcODT & MemCadBus settings.For all of you who want to see how it starts to find the first working OC settings (Click to show)Searching for the right settings is a quick and funny task, NOT. Here is my paper work before continue with Google Docs document:
3. IBT AVX Extream 30 loops is the quickest way to confirm your RAM timings, DRAM Voltage + VTTDDR Voltage, VDDSOC, ProcODT & MemCadBus settings. HciMemTest is the next fastest test to do so.
   • Use verified DRAM Voltage + VTTDDR Voltage combinations only (credits to @1usmus)Quote:Originally Posted by 1usmus
     
     I want to share with you the results of my testing
     
     
     In most cases, the memory works incorrectly because of incorrect voltage DRAM and VTT_DDR
     On this picture in green I selected the best options
     
     do not argue, they are not universal, to calculate their personal voltages, use my calculator, the tab VDRAM / VTT DDR Calculator
     
     
     
     Quote:Originally Posted by 1usmus
     
     
   • Care about the temperature for your RAM sticks:Quote:Originally Posted by Esenel
     
     Quote:Originally Posted by 1usmus
     
     I shot a jitter phenomenon on video. When the 52 degree memory module reaches the platform, errors appear. This video can be a huge scandal.
     
     
     
     For me it looks exactly the same.
     It is a reproducible phenomenon for me at 3466 Mhz CL14. At first I thought it was just my RAM not being capable of the OC, until I realized the errors just begin when the RAM is at 52.X°C.
     Then I forced my fans to 100% during HCI Memtest and I passed the 400% without any error at ~47°C.
4. BOINC or AIDA64 stress testing for 8h is a good starting point to continue, for a overall stability but "stability" is not confirmed yet either.
5. Continue with a digital test report document. Feel free to copy this Google Docs document as template.
6. With Prime95 Blend 2048k-2049k you will finalise the right settings for Rtt and CLDO_VDDP, if 24h+ can be passed without errors. Prime95 Blend 2048k-2049k is very Rtt / CLDO_VDDP sensitive. Small changes in Rtt / CLDO_VDDP result in quick errors. Typically it fails withing 1h, sometimes you need 2h-4h. Prime95 Blend 2048k-2049k tests may fail after 19h too, if the Rtt / CLDO_VDDP settings are wrong. The difference between a good CLDO_VDDP value and the right one is, that a good CLDO_VDDP let you pass Prime95 only for 19h but the right one for more then 24h+.
   
   CLDO_VDDP has a big impact on stability:Quote:Originally Posted by neur0cide
   
   CLDO_VDDP does have a major impact on stability. But this is mostly true for DR modules.
   I have tested ~25 kits in the last 8 months (SR and DR B-die and DR D-/E-die exclusively) and on about half of them I experimented with CLDO_VDDP. I found that almost every DR kit needs adjustments on CLDO_VDDP to show its true potential, whereas SR kits rarely need any - except of course in case of a memory hole .
   Quad kits with SR modules (4 Ranks) also benefit from alternate values sometimes, but not nearly to the extent of true DR modules and usually I don't have to stray far away from the default 950mv.
   
   For example the kit in my signature (F4-3200C15D-32GTZ) was running fine with 3200-14-13-13-13 @1.375v. CLDO_VDDP=866mv allowed me to move up to 3333-14-14-14-14 @1.395v. With CLDO_VDDP=855mv I was able to lower DRAM voltage to 1.375v.
   
   
   Use proposed CLDO_VDDP (credits to @1usmus):Quote:Originally Posted by 1usmus
   
   I decided to write a small article explaining why some systems work fine at 3200+ and others do not at all
   
   Influence of СLDO_VDDP on MEMCLK "holes"
   
   CLDO_VDDP is a voltage regulator for the module (physical interface) of encoding and decoding of the transmitted and received data stream. The purpose of coding is to simplify the process of restoring the data stream of the receiver. It determines the signals, signal ratios and time parameters necessary for transferring control information, reading and writing data to DRAM devices. In plain language, CLDO_VDDP is the voltage that regulates the memory access at a certain frequency. "Hole" in turn - the frequency gap on which the memory controller can operate with our RAM.
   
   Consider this simple picture:
   
   
   
   
   It shows 3 identical systems (motherboard + RAM + processor). All 3 systems were overclocked and received the following results:
   
   1) The system was perfectly dispersed to a frequency of 3333 MHz
   2) The system was accelerated to 2933
   3) The system did not start at all
   
   If all systems are the same, why such results? Let's understand. The bottom line is that each memory controller (IMC) has its own technical characteristics ("voltage" and time) and at the same voltage / frequency it will behave differently, namely it will have different access to memory. Red marked our MEMCLK holes, these are the very hole-mediators through which our memory controller communicates with RAM, and if there is no hole in the frequency range chosen by us - the system does not start or start, but the memory runs with errors. At you I think there was a question as these holes to move and expand - all is very simple, voltage CLDO_VDDP allows to spend the given manipulations. The only difficulty is that these holes can not be mathematically calculated. A vivid example of CLDO_VDDP 866 which is magical for many. The hole of this voltage is in the region of 3300-3500 MHz, but again not for all systems. As shown by our internal tests, not all of it works, I repeat all the IMC are different and require a different voltage CLDO_VDDP to achieve the same frequency.
   
   In view of the fact that the shape of the voltage CLDO_VDDP is wave, the minimum voltage change can drastically change the stability of the system. The voltage step is 1 mv. Borders from 700 to 975.
   
   I also want to publish a list of CLDO_VDDP, which can help stabilize your memory
   
   CLDO_VDDP list (volts) (Click to show)0.562
   0.568
   0.573
   0.579
   0.585
   0.590
   0.596
   0.601
   0.607
   0.613
   0.618
   0.624
   0.630
   0.635
   0.641
   0.646
   0.652
   0.658
   0.663
   0.669
   0.675
   0.680
   0.686
   0.691
   0.697
   0.703
   0.708
   0.714
   0.720
   0.725
   0.731
   0.736
   0.742
   0.748
   0.753
   0.759
   0.765
   0.770
   0.776
   0.781
   0.787
   0.793
   0.798
   0.804
   0.810
   0.815
   0.821
   0.826
   0.832
   0.838
   0.843
   0.849
   0.855
   0.860
   0.866
   0.871
   0.877
   0.883
   0.888
   0.894
   0.900
   0.905
   0.911
   0.916
   0.922
   0.928
   0.933
   0.939
   0.945
   0.950
   0.956
   0.961
   0.967
   0.973
   
   upd 1 : small addition-explanation
   
   The curve represente your dram voltage signal, changing cldo_vddp move back and forth this signal compared to your available frequency setting.
   The purpose to changing it is to make the curve crossed your setting line on the appropriate frequency setting.
   Whereas the zone where there is no curve present is a memory hole .
   Depending of your IMC or RAM the curve can vary in lenght or thickness.
   
   
   
   only the form is more oblate, in dozens of times
   
   Quote:Originally Posted by The Stilt
   
   Few suggestions regarding the controls allowed by the new (AGESA 1.0.0.6) bioses:
   
   - In case you run into a MEMCLK hole, adjust the CLDO_VDDP voltage. The VDDP adjustment window is rather narrow, usually < 100mV. Also the window is neither static or linear. Because of that the setting which is optimal for frequency x might not be optimal for frequency y. Also since the window is not linear, but more of a wave form e.g. VDDP at 975mV might work perfectly fine whereas 980mV won't be able to train the memory. The MEMCLK hole is both CPU and DRAM specific, but so far I haven't seen any evidence it being motherboard specimen specific. This means that swapping either the CPU or the memory (to another CPU or modules) might either introduce or the get rid of the MEMCLK hole. Personally I have 100% success rate in clearing the MEMCLK hole with CLDO_VDDP adjustment (1x R7 1700, 1x R7 1800X and 2x R7 1700X). All of the MEMCLK holes on these CPUs have been cleared using 937 - 1000mV setting. Do note that when you change the CLDO_VDDP voltage, saving the bios settings will not put the new CLDO_VDDP voltage into effect, since the CLDOs can only be programmed during a cold reset or a cold boot. Because of that I suggest that you save the new CLDO_VDDP value and press the reset button before the system has booted up again. Also CLDO_VDDP must be at least 100mV lower than the DRAM voltage at all times. Regardless it is not recommended to exceed 1050mV.
   
   - For Samsung B-die dual rank modules I suggest that 96Ohm ProcODT is used.

 

[gupsterg]

Quote:

Originally Posted by Tamalero

Disabled HPET = 1000ms tick rate.
Enabled HPET = erratically going from 1500 to a max 4100us. Mouse jerks around, programs are slower to boot.

Conclusion: HPET does not play well in my system lol.

Quote:

Originally Posted by Tamalero

Thanks!

Checking with that tool, I'm between 20us to 150us.
Max 1585us (Nvidia nvlddmkm driver lol)
highest interrpt: 748us.

*edit**

HPET ON:
Min : 345us
Max: 2500us
Highest Interrupt 3320us
Highest routine: 362us
Highest DCP: ntoskernel at 2941us.

Yep, HPET does not work well with my system.

You have ThreadRipper and aware you have Asus Prime X399-A. Probably come to see info that may help you with TR, but this thread has massive activity please don't create posts which may cause confusion from members reading them and not knowing what you have.

 

[CCoR]

Quote:

Originally Posted by 1usmus

HPET OFF
NX Mode Disable
Ai suite - off



78 us

Nice dude!

Hmm running same settings but it much higher (500) consistent. Did you run any command prompt tweaks to tune it?

 

[gupsterg]

Quote:

Originally Posted by harrysun


For me was the statement from @gupsterg not true. CLDO_VDDP has a big impact on stability:
Quote:

IMO changes to CLDO_VDDP wouldn't help based on what previous testing has passed. I reckon ProcODT / CAD Bus / SOC / VDIMM maybe what needs to be tweaked.

IMO is In my opinion, which was based on experience with my HW.

Plus you will also find a post in this thread by The Stilt where he states similar view, that unless CLDO_VDDP is vastly out of wack it shouldn't affect stability.

I'm glad you solved your issue and have shared your experience

.

 

[hughjazz44]

Quote:

Originally Posted by Gettz8488

does anyone know if you need to disable the skew in order to get correct temps?

Yes. However, in later BIOSes it's disabled by default, so just leave the setting at "Auto".

 

[dual109]

Quote:

Originally Posted by TheHawk88

I'am running my Corsair LPX sticks (CMK16GX4M2B3200C16) on 3200MHz based on The Stilts safe settings since a few Bios releases. Seems work just fine on 3501 too.
Yesterday i passed 2 hours of OCCT Linpack AVX.

Spoiler: And here are the settings i changed in my bios

Code:

Ai Overclock Tuner [Manual]
BCLK Frequency [100.0000]
BCLK_Divider [Auto]
Memory Frequency [DDR4-3200MHz]
Core Performance Boost [Disabled]
CPU Core Voltage [Offset mode]
CPU Offset Mode Sign [+]
- CPU Core Voltage Offset [0.02500]
CPU SOC Voltage [Manual mode]
- VDDSOC Voltage Override [1.02500]
DRAM Voltage [1.37500]
DRAM CAS# Latency [16]
DRAM RAS# to CAS# Read Delay [17]
DRAM RAS# to CAS# Write Delay [17]
DRAM RAS# PRE Time [17]
DRAM RAS# ACT Time [30]
Trc [60]
TrrdS [6]
TrrdL [9]
Tfaw [34]
TwtrS [4]
TwtrL [12]
Twr [10]
TrdrdScl [5]
TwrwrScl [5]
Trfc [480]
Trfc2 [300]
Trfc4 [132]
Tcwl [16]
Trtp [12]
Trdwr [6]
Twrrd [3]
TwrwrSc [1]
TwrwrSd [7]
TwrwrDd [7]
TrdrdSc [1]
TrdrdSd [5]
TrdrdDd [5]
Tcke [8]
ProcODT [60 ohm]
Cmd2T [1T]
Gear Down Mode [Enabled]
Power Down Enable [Disabled]
VDDP Voltage [0.85500]
VRM Spread Spectrum [Disabled]
Active Frequency Mode [Disabled]
SVM Mode [Enabled]
Super I/O Clock Skew [Disabled]
In sleep, hibernate and soft off states [Off]
Next Boot after AC Power Loss [Fast Boot]
POST Delay Time [0 sec]

Thanks for the info, the timings for stilts safe preset Hynix MFR SR did improve bandwidth greatly however after many hours of testing I could not quite get memory stable @3200Mhz even after increasing voltages considerably for SOC and DDR as it kept failing in AIDA stability test. My CPU was at stock clocks to ensure stabilty and will have another play tonight and maybe look at changing the ProD_OCt and few others..

Quote:

Originally Posted by TheHawk88

Hi, no there aren't any presets for Hynix memory in the bios. He posted the settings in the post i referenced. You have to enter them manually.

Thanks I worked that one out.

Quote:

Originally Posted by Targonis

What ProcODT, and geardown settings are you using? Have you tried geardown disabled with 2T?

I'm using the Stilt Safe setting for Hynix M-die MFR SR memory so ProcODT 60ohm. No I'll give those settings a try tonight.

 

[Tamalero]

Quote:

Originally Posted by gupsterg

You have ThreadRipper and aware you have Asus Prime X399-A. Probably come to see info that may help you with TR, but this thread has massive activity please don't create posts which may cause confusion from members reading them and not knowing what you have.

Any suggestion then?

The other thread of owners of Threadrippers are either semi active (noone in the PRIME motherboard, only active on the very expensive zenith and similars).
And the general threadripper thread as very few helpful posters (and a few very snotty ones)

 

[VPII]

Some advice please. My system appear to run fairly well without much problems. I do however get the get the odd blue screen error every now and then which when I googled it relates to memory. I followed the procedures stated by a microsoft technician, one of which meant using windows memory diagnostic tool mdsched but soon after starting it would find a problem. In the end I downloaded memtest86 and ran it last night which found 86 errors.

The memory in question is: G.Skill F4-3200C14D-16GTZR TridentZ RGB Series 16GB (2x 8GB)

Not sure if this link to the report will work, but maybe some of you can shed some light. Do I need to replace.

file:///F:/EFI/BOOT/MemTest86-Report-20180113-190049.html

 

[VPII]

Sorry link wont work.

What it states is as follow, only errors I'll describe

Test 4 [Moving inversions, 8-bit pattern] 3/4 (75%) 8 errors
Test 6 [Block move, 64-byte blocks] 1/4 (25%) 66 errors
Test 7 [Moving inversions, 32-bit pattern] 3/4 (75%) 2 errors

Last 10 Errors
[Data Error] Test: 6, CPU: 0, Address: 1D28D5678, Expected: FFFD00FF, Actual: FFFDFFFF
[Data Error] Test: 6, CPU: 0, Address: 1D28D5670, Expected: 0002FF00, Actual: 00020000
[Data Error] Test: 6, CPU: 0, Address: 1D28D5668, Expected: FFFD00FF, Actual: FFFDFFFF
[Data Error] Test: 6, CPU: 0, Address: 1D28D5658, Expected: 0002FF00, Actual: 00020000
[Data Error] Test: 6, CPU: 0, Address: 1D28D5650, Expected: FFFD00FF, Actual: FFFDFFFF
[Data Error] Test: 6, CPU: 0, Address: 1D28D5648, Expected: 0002FF00, Actual: 00020000
[Data Error] Test: 6, CPU: 0, Address: 1D08D5698, Expected: FFFD00FF, Actual: FFFDFFFF
[Data Error] Test: 6, CPU: 0, Address: 1D08D5690, Expected: 0002FF00, Actual: 00020000
[Data Error] Test: 6, CPU: 0, Address: 1D08D5688, Expected: FFFD00FF, Actual: FFFDFFFF
[Data Error] Test: 6, CPU: 0, Address: 1D08D5678, Expected: 0002FF00, Actual: 00020000

 

[seansplayin]

hi I'm using Corsair Vengeance LPx 3200 CMK16GX4M2B3200C16 on ASUS Crosshair 6 Hero bios 3008 directly off ASUS's website. I had stability troubles with the last beta bios 3101 so I used the official ASUS bios.
I now have my memory not only stable at default timings but stable at 3200 with lower timings.
At 1.35 memory voltage the default timings were stable at 3200 without increasing SOC voltage or any other voltages for that matter. the Default timings for 3200 are 16,18,18,18,36,75.
at 1.47 memory voltage my timings are 14,17,17,17,34,57. and have passed 12 instances of Memtest at 1GB per instance up to 2000% coverage which takes around 10 hours to complete.

A couple observations that can save you a lot of time-
-Setting the RAS to CAS delay read /write (tRCD) and RAS Precharge (tRP)below 17 I could benchmark but I could not pass stability tests regardless how much voltage I threw at it.
-Even though the memory SPD says that the Bank Cycle Time (tRC) should be 55 at 3200 I was unable to get it stable below 57 regardless of how much voltage I threw at it. the Bios defaults
to 75 and that works at 1.4v but to decrease this timing requires more memory voltage.
- I also increased the DRAM current up to 120% but not certain I needed to do this.
- I have Vboot at 1.5V but not certain I need to

 

[dual109]

Quote:

Originally Posted by seansplayin

hi I'm using Corsair Vengeance LPx 3200 CMK16GX4M2B3200C16 on ASUS Crosshair 6 Hero bios 3008 directly off ASUS's website. I had stability troubles with the last beta bios 3101 so I used the official ASUS bios.
I now have my memory not only stable at default timings but stable at 3200 with lower timings.
At 1.35 memory voltage the default timings were stable at 3200 without increasing SOC voltage or any other voltages for that matter. the Default timings for 3200 are 16,18,18,18,36,75.
at 1.47 memory voltage my timings are 14,17,17,17,34,57. and have passed 12 instances of Memtest at 1GB per instance up to 2000% coverage which takes around 10 hours to complete.

A couple observations that can save you a lot of time-
-Setting the RAS to CAS delay read /write (tRCD) and RAS Precharge (tRP)below 17 I could benchmark but I could not pass stability tests regardless how much voltage I threw at it.
-Even though the memory SPD says that the Bank Cycle Time (tRC) should be 55 at 3200 I was unable to get it stable below 57 regardless of how much voltage I threw at it. the Bios defaults
to 75 and that works at 1.4v but to decrease this timing requires more memory voltage.
- with these low timings I'm running the SOC voltage at 1.1v but I'm not completely certain I need it, and I also increased the DRAM current up to 120% but again not certain I needed to do this.

Thanks for pointers above. Got same kit and trying for 3200Mhz. Not quite stable using AIDA stability test but not far off. I'm using the Stilt safe preset for Hynix MFR M-Die ram. Any chance posting your bios settings. just txt formats fine.

 

[seansplayin]

just sent you pm dual109

 

[dual109]

Quote:

Originally Posted by seansplayin

just sent you pm dual109

thanks

 

[Zognarb]

Just an FYI regarding the timer/HPET associated snakeoil in this thread. DPC Latency Checker does not output valid data on Windows 8 and later, as noted on the creator's website:
Quote:

Windows 8 Compatibility: The DPC latency utility runs on Windows 8 but does not show correct values. The output suggests that the Windows 8 kernel performs badly and introduces a constant latency of one millisecond, which is not the case in practice. DPCs in the Windows 8 kernel behave identical to Windows 7. The utility produces incorrect results because the implementation of kernel timers has changed in Windows 8, which causes a side effect with the measuring algorithm used by the utility.

This is due to Windows 8 (NT 6.2) implementing a tickless kernel, so if you insist on obsessing over DPC latency, you need to use LatencyMon to get a valid reading.

By default Windows will automatically use the best hardware timer available, which today is likely the CPU TSC. Manually forcing a certain timer selection via useplatformclock is not a good idea, and is intended purely for debugging purposes as you can see in the MSDN documentation.

It's also worth noting that finer grained timers are not always a good thing.

 

[furkandeger]

Is there anything wrong with sense mi skew?

For me Auto and Disabled gives the same results, while Enabled with offset 272 seems to produce more realistic results. I shouldn't be having 45-50 degrees idle.

R7 1700 (non-X)

 

[hurricane28]

Spoiler: Warning: Spoiler!

Quote:

Originally Posted by harrysun

Finally I got F4-3200C14D-32GTZ 2x16GB dual-rank Samsung B-die @3333MT/s CL14 "stable" with BIOS 1701. As always, my current stable setup can be found in the signature!

How stable is this setup:

Spoiler: In a 30h run of Prime95 Blend 2048k-4096k one worker failed after 26h

*) T_Sensor1 = Ambient temperature in the case

It's time to share my learnings:

1. Start with Ryzen DRAM Calculator (credits to @1usmus)
2. Use paper and pen for your results first, until you have found the right RAM timings, DRAM Voltage + VTTDDR Voltage, VDDSOC, ProcODT & MemCadBus settings.
   
   Spoiler: For all of you who want to see how it starts to find the first working OC settings

   Searching for the right settings is a quick and funny task, NOT. Here is my paper work before continue with Google Docs document:
3. IBT AVX Extream 30 loops is the quickest way to confirm your RAM timings, DRAM Voltage + VTTDDR Voltage, VDDSOC, ProcODT & MemCadBus settings. HciMemTest is the next fastest test to do so.
   • Use verified DRAM Voltage + VTTDDR Voltage combinations only (credits to @1usmus)
     
   • Care about the temperature for your RAM sticks:
4. BOINC or AIDA64 stress testing for 8h is a good starting point to continue, for a overall stability but "stability" is not confirmed yet either.
5. Continue with a digital test report document. Feel free to copy this Google Docs document as template.
6. With Prime95 Blend 2048k-2049k you will finalise the right settings for Rtt and CLDO_VDDP, if 24h+ can be passed without errors. Prime95 Blend 2048k-2049k is very Rtt / CLDO_VDDP sensitive. Small changes in Rtt / CLDO_VDDP result in quick errors. Typically it fails withing 1h, sometimes you need 2h-4h. Prime95 Blend 2048k-2049k tests may fail after 19h too, if the Rtt / CLDO_VDDP settings are wrong. The difference between a good CLDO_VDDP value and the right one is, that a good CLDO_VDDP let you pass Prime95 only for 19h but the right one for more then 24hh+.
   
   For me was the statement from @gupsterg not true. CLDO_VDDP has a big impact on stability:
   Use proposed CLDO_VDDP (credits to @1usmus):
   

Good info man! Rep+

My CLDO VDDP is at 700 and i am pretty stable. I doubt increasing this voltage will cause me more stability but maybe i give it a try.

Also, where did The Stilt mentioned this? Was it in this thread or somewhere else?

 

[gupsterg]

Quote:

Originally Posted by Tamalero

Any suggestion then?

The other thread of owners of Threadrippers are either semi active (noone in the PRIME motherboard, only active on the very expensive zenith and similars).
And the general threadripper thread as very few helpful posters (and a few very snotty ones)

Signed up to relevant thread

.