https://youtu.be/xSAFs-t0DCg?t=1000
Did you check your IF is stable. Sometimes higher voltage makes things worse. IF appears to need just the right voltage to be stable, too much is as bad as too little.
Last edited:
Ballistix Sport LT Overclocking
- Thread starter Troezar
- Start date
https://youtu.be/xSAFs-t0DCg?t=1000
Did you check your IF is stable. Sometimes higher voltage makes things worse. IF appears to need just the right voltage to be stable, too much is as bad as too little.
TRFC I had read to start at around 600 (Edit: depending on module IC's of course) then reduce by 20 until unstable.then increase slightly to find stability. To get Trfc2 you divide trfc by 1.346. To get Trfc4 you divide trfc2 by 1.625 so the settings above trfc=560, trfc2=416 trfc4=256.
It looks like it. I belive all with AES in the part number is Micron rev E
Currently I have SOC = 1.1V, CLDO_VDDP = 0.9V, CLDO_VDDG = 0.95V. Just discovered that 3666 MT/s isn't stable either, just takes longer to crash. It could be the IF as you say but so far I haven't found how to fix it. I've tried CLDO_VDDG = 1.0V before but that didn't help. I know CLDO_VDDG = 0.9V will make it worse because that was the default setting with the early BIOS for this board and I couldn't even run XMP stable.
Think I might go with IF 1866 just to be sure. I don't want to have issues. Everyone has been saying they have issues with IF 1900 and ABBA. I don't want to find temp changes make me unstable.
bump up vddp and vddg to 1050 mV
Some info for those going the Micron E-Die route, the 3000Mhz C15 AES Ballistix DIMM(s) are slightly higher bin than the 3200Mhz C16 ones.
Since tRCD(RD) doesn't scale with voltage for Micron E-Die, the 3000 15-[16]-16 version is just slightly better than the 3200 16-[18]-18 one. Though it probably won't really make much of a difference either way, just get whatever is cheapest, imo.
Also got 4x 3000Mhz C15 DIMMs with an R5 3600 and ASUS X370 Prime Pro mobo (T-Topology), managed to get:
- 3733Mhz 16-20-20-16-38-60 stable with tight subtimings and tertiaries at 1.45v with 1T GDM=on PDM=off procODT=40Ohm, but sadly my chip can't seem to do 1866Mhz on the IF, otherwise I had 7+ hours of Karhu RamTest with 0 errors.
- 3600Mhz 14-20-20-14-36-58 stable with tight subs & tertiaries at 1.45v with 1T GDM=on PDM=off procODT=40Ohm (ran Karhu RamTest for a few hours, 0 errors)
- 3600Mhz 16-20-20-16-38-60 stable with tight subs at 1.4v with 1T GDM=on PDM=off procODT=40Ohm, which is what I'm using right now since 1.4v feels like a pretty good voltage for the speed & timings (0 errors after extensive RamTest also)
Only times I really got errors was when I pushed some subtimings a bit too hard. If I pushed the primaries too much without the needed voltage I would just get bootloops.
Need to test these out more to find their limits.
Since tRCD(RD) doesn't scale with voltage for Micron E-Die, the 3000 15-[16]-16 version is just slightly better than the 3200 16-[18]-18 one. Though it probably won't really make much of a difference either way, just get whatever is cheapest, imo.
Also got 4x 3000Mhz C15 DIMMs with an R5 3600 and ASUS X370 Prime Pro mobo (T-Topology), managed to get:
- 3733Mhz 16-20-20-16-38-60 stable with tight subtimings and tertiaries at 1.45v with 1T GDM=on PDM=off procODT=40Ohm, but sadly my chip can't seem to do 1866Mhz on the IF, otherwise I had 7+ hours of Karhu RamTest with 0 errors.
- 3600Mhz 14-20-20-14-36-58 stable with tight subs & tertiaries at 1.45v with 1T GDM=on PDM=off procODT=40Ohm (ran Karhu RamTest for a few hours, 0 errors)
- 3600Mhz 16-20-20-16-38-60 stable with tight subs at 1.4v with 1T GDM=on PDM=off procODT=40Ohm, which is what I'm using right now since 1.4v feels like a pretty good voltage for the speed & timings (0 errors after extensive RamTest also)
Only times I really got errors was when I pushed some subtimings a bit too hard. If I pushed the primaries too much without the needed voltage I would just get bootloops.
Need to test these out more to find their limits.
Last edited:
Looks like it changes everytime I change RAM settings. I am 1.050 volts Prem SoC and VDDP is 0.880 volts. Changing volts in the overclocking area of BIOS does nothing for me. So I can change only PERM SoC and VDDP. The thing is that there is a point of stability which is just one point in the spread of available voltages. This is were I can run vray over and over without cracking and resets. This all started with the ABBA. I have been make changes to my RAM and this means I have to get everything stable afterwards.
The funny thing is that this one point is also the point of maximum performance in vray.
The IF remains stable in memtest86 and other applications but as soon as vray is run then the isssues begain. It takes a few runs but as you change the voltages and get close to stable. Performance increases and you can run vray longer and longer without issues.
AmD Ryzen master shows VDDCR SOC, CLDO VDDP and CLDO VDDG. All as 1.1,1.0979 and 1.0979 volts. I can set them to 0.900 and 0.950 but BIOS will ignore me. My motherboard tuning software A-Tuning shows PERM_VDDCR_SOC Voltage 1.053 volts (default) and VDDP Voltage 0.896 volts (set to 0.880 volts in BIOS). CPU VDDCR_SOC Voltage 1.208 volts (default). This is want I have entered in the main area, the overclocking area is ment to over ride this but doesn't.
From what I can tell the BIOS overclocking area does nothing. Setting PBO and RAM etc there does nothing. The RAM timings are ignored. So getting stable became down to what BIOS autoed too after changing RAM settings. So one moment I am stable but then could not get stable with different RAM settings. I have no way of changing VDDG, just VDDP voltage. IF 1866 will just run without effort, like IF 1900 did before the ABBA BIOS.
Aida64 will draw 88 watts @ 1.408 volts all cores hit 69c and be stable but vray will hit 108 watts with less heat and become unstable very quickly at the wrong settings.
Hopefully this is useful to someone.
Last edited:
I tried bumping CLDO_VDDG to 1.05V and it reboots within a few seconds of OCCT at both 3666 and 3733 MT/s. Haven't tried playing with CLDO_VDDP yet.
That what happens, vray just a few runs will cause cracking when sound is played or reboots. vray pulls a lot of watts as well. Aida64 is 88-89 watts but vray will pull 105-109 watts and hammer the IF into the ground. VDDP has to be the right value, all others will cause issues.
CLDO VDDP voltage - voltage for the DDR4 PHY on the SoC. The DDR4 PHY or physical-layer interface converts information from the memory controllers to a format the DDR4 memory modules can understand.
Somewhat counterintuitively, lowering VDDP can often be more beneficial for stability than raising CLDO_VDDP. Advanced overclockers should also know that altering CLDO VDDP can move or resolve memory holes. Small changes to VDDP can have a big effect, and VDDP cannot not be set to a value greater than VDIMM - 0.1 V (not to exceed 1.05 V). A cold reboot is required if you alter this voltage.
Limit: up to 1.0 V.
Just setting VDDP to 1.05volts is a very bad idea. Most of the time these voltages need to be reduced from their auto values in BIOS. I auto to 0.900 or something close but need to set it to 0.880volts to be stable. Changing SoC voltage will cause VDDP to increase or decrease. Also SoC voltage has to be the right value for things to be stable. Not too high or too low. If I set PREM SoC to 1.035volts I will push VDDP towards 0.890volts.
https://www.amd.com/system/files/documents/ryzen-master-quick-reference-guide.pdf
SOC voltage is the voltage to the IMC and like with Intel, it's not recommended to leave it on auto. You typically want 1.0 - 1.1v as above 1.1v doesn't help much if at all.
On Ryzen 2000 (possibly 1000 and 3000 as well), above 1.15v can negatively impact overclocking.
AMD only state that VDDCR SoC and CLDP VDDG should typically start at 1.1v for memory overclocking but if I set SoC to 1.1 it will restart or start cracking in sound. Anything above 1.050 in bios will cause big problems. Having SoC at 1.050 volts will reduce VDDP and VDDG if they are set to 1.1volts.
On Ryzen 3000, there's also CLDO_VDDG (not to be confused with CLDO_VDDP), which is the voltage to the Infinity Fabric. I've read that SOC voltage should be at least 40mV above CLDO_VDDG, but other than that there's not much information about it.
Most cLDO voltages are regulated from the two main power rails of the CPU. In case of cLDO_VDDG and cLDO_VDDP, they are regulated from the VDDCR_SoC plane. Because of this, there are couple rules. For example, if you set the VDDG to 1.100V, while your actual SoC voltage under load is 1.05V the VDDG will stay roughly at 1.01V max. Likewise if you have VDDG set to 1.100V and start increasing the SoC voltage, your VDDG will raise as well. I don't have the exact figure, but you can assume that the minimum drop-out voltage (Vin-Vout) is around 40mV. Meaning you ACTUAL SoC voltage has to be at least by this much higher, than the requested VDDG for it to take effect as it is requested.
Adjusting the SoC voltage alone, unlike on previous gen. parts doesn't do much if anything at all. The default value is fixed 1.100V and AMD recommends keeping it at that level. Increasing the VDDG helps with the fabric overclocking in certain scenarios, but not always. 1800MHz FCLK should be doable at the default 0.9500V value and for pushing the limits it might be beneficial to increase it to =< 1.05V (1.100 - 1.125V SoC, depending on the load-line).
https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md
So with my CPU in A-Tuning I can see I am at 1.053 volts for Perm SoC and VDDP I am just under 0.900 volts. HWMonitor is the same and so is BIOS. AMD Ryzen master thinks its all 1.1 volts. I am more or less stock settings.
Last edited:
Most my chip & mobo combo (R5 3600 and X370 Prime Pro) can do without any issue seems to be 1800Mhz for the FCLK, 1866Mhz gives me random stuttering, which seem to be due to Fabric not being stable, seems ~1800Mhz is where most chips top out.
I tried that with 1.1v on the SoC and 1.05v on the cLOD_VDDG, 1v VDDP (past 1v on VDDP does nothing for stability). Increased load line on the SoC so the VDDG doesn't drop bellow 1.05v might help but that seems like a bad idea for longevity...
I tried that with 1.1v on the SoC and 1.05v on the cLOD_VDDG, 1v VDDP (past 1v on VDDP does nothing for stability). Increased load line on the SoC so the VDDG doesn't drop bellow 1.05v might help but that seems like a bad idea for longevity...
I was unstable too following the guides were you need to increase the voltage.
Take Ryzen master, it will tell you the range of voltages to help you get stable.
For my RAM
cLDO VDDG Voltage min 0.950, Rec. 0.950 and max 1.075
cLDO VDDP Voltage min 0.700, Rec. 0.900 and max 1.1
SoC min 1.050, Rec 1.1 and max 1,125.
My BIOS settings
There is no need to go for high voltages right away. Remember IF could be stable but there could be an issue with VDDP. Small changes to VDDP can have a big effect. You can pass memtest86 and still have issues. Then fix VDDP, restarts in vray and cracking in sound after running vray will go away.
Example, MAX-peaks with VDDG @ 800mv | VDDP @ 750mv | SoC @ 1.100V no issues with my 3800 1:1 15-23-17-32-66 timings.
https://www.overclock.net/forum/13-...zen-3000-memory-fabric-x370-x470-x570-22.html
Example two
Might be, or it depends on settings and memory?
I tried VDDP/VDDG 750/750 again in 3800 1:1 mode and it failed the benchmark.
Trying VDDP/VDDG 750/760 now and have 1 pass already, trying another seems to pass round 2 as well. Stress test next.
I've noted before VDDP @ 700 fails to boot. 750 for that seems ok.
EDIT: VDDG @ 760 failed the stress test. Testing 770mv next.
EDIT2: I might have set a memory timing wrong/too tight and was causing stability issues. Might need to retest 750mv etc if it was that causing the failure for these lower voltages.
EDIT3: YEP seems I had a timing issue, retesting 750/750 again.
EDIT4: After finally nailing down a stable timing I've been testing VDDG @ 770mv and been running ok 5 rounds off stress test in y-cruncher for ~60min++ now. Stable? seems so.
EDIT5: VDDG 750mv is unstable... it fails the VST test. AVX2 Float for Vector Transform.
EDIT6: VDDG 760 is unstable... 770mv is ok and I can also lower vSoC to 1.050 without issue with it. Seems I found my cpu limit for VDDG @ 770mv @ 3800Mhz(1900) MEM/FCLK/UCLK.
EDIT7: After a while I've come to reason it's not really worthwhile to use 770 as such. It basically only works with the timings I had for it but tighter/faster and it stops working. Need to increase VDDG then.
800mv seems to be the minimum as a whole for the speeds I'm running. I get problems with VST in y-cruncher with it too low when I keep tweaking settings.
Lesson here is that everytime you change a RAM setting you have to make sure you are still stable. memtest86 is not enough. Make sure your RAM timings are correct. Wrong VDDP and it will effort stablity on warm reboots. You will get higher and lower RAM performance.
What you should aim for is this,
See RAM latency 60ns
https://www.userbenchmark.com/UserRun/21065112
Once you get the best time then its time to stop.
Note that this also affects hard disks as well. UserBenchMark is great for testing. Hard Disks love to disappear and become no longer accessible with IO errors. This was caused by VDDP as well as far as I could tell,VDDG or SoC to low. Once VDDP was set to 0.880 in BIOS, PERM SoC 1.050volts I have no issues running UserBenchMark and my hard disk dying.
Was able to find the right values by running vray.
IF1866 and vray fine but hard disk issues in UserBenchMark. It's the same every time you change anything. You have to find the new stable settings.
Last edited:
Target latency is 65ns and ryren dram calculator easy test 104 seconds. Sometimes you get timings which give lower latency. Make a note of them.
I was just doing it manually without the calculator following this guide here
https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md
It's really interesting that severely undervolting VDDP and VDDG can apparently improve stability at high clock rates. Might be worth a try...I was unstable too following the guides were you need to increase the voltage.
Take Ryzen master, it will tell you the range of voltages to help you get stable.
For my RAM
cLDO VDDG Voltage min 0.950, Rec. 0.950 and max 1.075
cLDO VDDP Voltage min 0.700, Rec. 0.900 and max 1.1
SoC min 1.050, Rec 1.1 and max 1,125.
My BIOS settings
There is no need to go for high voltages right away. Remember IF could be stable but there could be an issue with VDDP. Small changes to VDDP can have a big effect. You can pass memtest86 and still have issues. Then fix VDDP, restarts in vray and cracking in sound after running vray will go away.
Example, MAX-peaks with VDDG @ 800mv | VDDP @ 750mv | SoC @ 1.100V no issues with my 3800 1:1 15-23-17-32-66 timings.
https://www.overclock.net/forum/13-...zen-3000-memory-fabric-x370-x470-x570-22.html
When you get a chance, can you elaborate a bit more? I assume you mean timings in Dram calculator?
Do you know which timings affect latency the greatest?