Thentilian's R5 1600 OC thread

[Thentilian]

Just completed over 7 hours of P95 while I slept. I dunno if that's normally 'good enough' to be considered stable? I want to work on overclocking the RAM next. That seems to be a tougher nut to crack then the CPU overclock. I've tried all sorts of timings and certain timings, even when losening them up just won't boot. When I put the timings tighter it will. For example I went for a 2800 OC on the RAM with 15-15-15-36 timings, didn't like that, but 14-16-16-31 it did, which is the same timings it runs at 2400. I crank the voltage up from 1.2 to 1.35 and my SOC is at 1.0v. I'm going to keep the SOC at 1.0 or higher even without the RAM OC as Johan mentioned. It did seem to keep things more stable just bumping it up to 1.0v.

Temps aren't an issue, I hit 67C on die, and 61C to 63C on CPU during that overnight run. Ambient in the house was pretty cool though, during the day I have seen Die temps as high as 73C, but still not a concern, as those loads were generated by IBT, which makes my CPU hotter than running P95.

EDIT: Always forget something... so the voltage is 1.412 Vcore, with LLC 1, gets me stable at 3.9Ghz for at least 7 hours of P95 anyways. Using offset voltage, but can't remember the precise number.

 

[Johan45]

Was that P95 blend? If so before you try overclocking the ram I would run P95 V29.1 on custom and set it to use 13000 MB of ram. The blend only uses maybe 2 GB this will incorporate about 75%

 

[Thentilian]

Was that P95 blend? If so before you try overclocking the ram I would run P95 V29.1 on custom and set it to use 13000 MB of ram. The blend only uses maybe 2 GB this will incorporate about 75%

Yes, it was on blend. I have V29.1 already, so custom with 13000 MB, anything else need to be set?

 

[Johan45]

Nope. That's it. If you start dropping threads that's memory related. Black screen is voltage.

 

[Thentilian]

Nope. That's it. If you start dropping threads that's memory related. Black screen is voltage.

Yeah, I was wondering, when I went to start my burn last night I had the 2800 OC on the RAM, but it dropped threads after 9 minutes. Figured it was the RAM OC it didn't like. Reset to 2400 DOCP settings and ran the 7 hours no issue. I ran the 13000 MB setting for about an hour while I was out this afternoon, but I'll run it again all night when I go to bed. Thanks for the tip on the custom settings.

 

[DaveB]

I know I have the headroom as far as temps go now. Even IBT only pushes 67 to 70C on the die. I'm just really reluctant to push past 1.4 or 1.41. I might give it a go just to see if I can get 4Ghz stable, but as you said 100Mhz more isn't really going to do much for me but push better benchmark numbers, and I'm more concerned about stability, game performance, streaming, and video editing.

Stability was not an issue, but longevity was a concern.

 

[Thentilian]

Stability was not an issue, but longevity was a concern.

Yes, that's why I don't want to push more than 1.4 or 1.41 volts for the sake of longevity. I might try to get 4ghz stable at some point just to see what my chip needs to be stable. Unless it's over 1.45 volt. I won't pass that even for short term curiosity.

 

[Thentilian]

Was that P95 blend? If so before you try overclocking the ram I would run P95 V29.1 on custom and set it to use 13000 MB of ram. The blend only uses maybe 2 GB this will incorporate about 75%

I ran 13000 MB custom profile as you suggested. It ran for 9h49m without any errors or warnings all cores. 3.9Ghz 1.412 volts LLC 1. RAM DOCP 2400 14-16-16-31 at it's DOCP volt of 1.2 with SOC 1.0v.

Don't want to post in the OC results just yet, as I want to try to dial in some OC on the RAM, since that seems to be the best way to gain some FPS while gaming.

 

[Thentilian]

Ok, so currently I'm running a P95 custom test with 13,000 MB of RAM being tested with my RAM OC to 2933 18-19-19-46 with 1.45v. RAM is warm, but certainly not hot. I had 3066 with an hour of memtest 86 no errors, but once I got into windows P95 shut down threads, and had a couple of blue screens. one was an sata.sys failure, and another was IRQ not less or equal.

I have SOC set to 1.1250 as well. Vcore is offset with 0.175 volts for 1.412 vore with LLC 1 set on both Vcore and SOC. I have GearMode on, and set CR to 2, but still boots with a CR of 1. I also had BIOS booting issues where it'd randomly like the settings or it wouldn't as well.

Went into a game of Dark and Light on a private server of ours and ran it fine for about 10 minutes... then exit game, bam, black screen. So, I'm not sure if I need more SOC voltage, or more CPU voltage with the higher RAM speed? Also, I am going to test backing down the voltage on the RAM from 1.45v. I see around the forums that's 'OK' but probably safer to do 1.4v. I read the OC article in Johan's links. It says that he has pushed as high as 1.6v on DDR 4 but I'm willing to go as far as 1.45v since the modules don't feel hot.

So, I'm just a bit confused as to why it seems stable one sec, then not, even after doing an hour of memtest. Right P95 is still going with all workers while I'm writing this, so hopefully I'm in the ballpark and then can start fine tuning.

EDIT: Also read that my Corsair Vengeance RAM is Dual Rank, which isn't as good for higher OC's and they weren't that good of overclocking RAM to begin with. So if I can manage 2933 or 2800 stable, I'll be happy until I can get some proper 3200 or higher RAM... Although, with the RAM prices the way they are right now, I'm probably gonna have to wait til next year after Apple has had their fill of RAM for their new phones and prices come down to sane levels!

 

[Thentilian]

Ok, I'm beginning to suspect something severely wrong with my PC. I've been having issues with the PC not wanting to turn on. I found it to be the RAM sticks are not being detected. I have to reseat them and BIOS will report only 8GB being detected sometimes. Other times I restart and the PC won't POST, just hangs with the power on. Problems really started after I checked the CMOS battery with a volt meter after pulling it out of the motherboard to see if it had juice or not, since I'm having issues with the PC keeping time correctly as well.

Anyways, hooked it back up and started getting the triple beeps, and it not liking something. The RAM was not being detected again. So I re-seated them again. Then my OC that was stable is now not stable even after I went back to DOCP settings on my RAM. Can't even pass standard IBT... and I noticed my Gflops are low... always been 60 to 62 at 3.9Ghz... I saw a Thuban 6 core scoring 78 or 79 Gflops recently. It was clocked to 4.2Ghz, but still I have 12 threads, shouldn't I be scoring higher?


EDIT: Sorry for all the information, just trying to sort this out... Looks like my board likes the RAM to be in certain slots... even though they are matched kit. A2 and B2 seem to be prefered and like a specific stick to be in each of those slots. Just after moving them around I got these results without tweaking anything in BIOS:


Such a touchy board with the RAM!!! Jeez...

 

[Johan45]

Your low score is because you're usinf the wrong version of IBT try the one linked in the intro of this thread http://www.overclockers.com/forums/showthread.php/782241-Post-your-AMD-Ryzen-Overclocks

 

[Thentilian]

Your low score is because you're usinf the wrong version of IBT try the one linked in the intro of this thread http://www.overclockers.com/forums/showthread.php/782241-Post-your-AMD-Ryzen-Overclocks

Oops, thanks... I thought I had the right version already. Yeah, this version was double the performance, whoa... ok. I also sorted out the RAM. It really likes one the RAM to be oriented a certain way, not sure why, but it does. I have the OC stable. 2800 on the RAM with CL16 1.36v 10+ hours of memtest 86 four passes completed. HPET turned off made a huge difference in gaming too. I noticed more FPS and smoother gaming. I'm glad someone mentioned that elsewhere. I know I've been all over the forums being a pain in the butt. Just want to dial this in and get every ounce of performance I can out of this system.

Clock is still losing time, I just don't get it at this point. If the DIMM slots keep being touchy I'll just have to RMA the board. I checked the voltage output of my PSU with my multimeter, 12v was 12.24 to 12.25 steady as heck, even while gaming. 5v was the same 5.06v steady. Not sure how to check the 3.3v rail though. I just had a molex connector with the probes to check voltage output.

Screencap:

 

[Johan45]

From what I have read you have the ram in the proper slots now. A2 and B2 are the recommended slots, pretty common for most motherboards.

 

[Thentilian]

From what I have read you have the ram in the proper slots now. A2 and B2 are the recommended slots, pretty common for most motherboards.

Yeah, that's what it said to have in the configuration. I tried the others because it seemed those DIMM slots were an issue. Plus it seemed to prefer a certain stick in each of the A2 and B2 slots. I had swapped them and it didn't like that. So now they are back how they were before.

Now I've had problems turning the PC on again, but don't think it's RAM. Might be the case power button or the plug for the front panel headers, for power, reset, etc... I had thought maybe the 24 pin power cable wasn't sitting all the way in, but it is. then I tapped the header connections. I'm using the Asus Q-connect thing.. it seems a bit lose, I'll ditch that and see what happens.

 

[Thentilian]

Ok, I did some final tweaking on my RAM OC. Looks like it was stable enough for 10+ hours and 4 passes of Memtest86, but soon as I went to the 'high' profile of IBT, errored. I'm not having much faith in Memtest86 anymore for making sure RAM is stable. I guess it's good enough to say if it's going to be an issue right away if it spits out a lot of errors off the bat, but IBT seems to weed out the pretenders.

So, I have 2800 with 1.37v, 16-18-18-42 CR1 that failed with high profile IBT. So, I was guessing it was close, so I changed CR to 2, and then ran IBT on custom with 12024MB, and this was the result:


I pray that this means stable. That took 57.5 minutes to complete. If not, I'm gonna put it back to it's DOCP of 2400 and just wait til I can afford some RAM meant to run at higher speeds.

 

[Johan45]

Looks stable enough to me

 

[Thentilian]

Looks stable enough to me

Yeah, it definitely feels more like it as well. Did over 3 hours of streaming Wolcen and I always crashed before. Wasn't sure if it was the PC or the game since the game is in an alpha state. But after dialing everything, I'm gonna say it was the PC not stable. Turning off the HPET in Windows helped gaming as well. Can't say if that helped stability any, but it's a smoother gaming experience now.

Also tossed the Q-connector. The powering on issue might of been caused by that. I put the front panel connectors directly on the motherboard. The Q-connector is nice, but I think it was on a bit too loose. Clock issue seems to be the only thing I'm having trouble with now. Although, after I set it this morning, I only needed to sync it once in the afternoon and it's been holding fine since... I'm still confused by this clock issue.

 

[Keith Myers]

Just noticed your thread here. I know I posted to you over in my thread but I thought I would jump in here with some info. Ryzen does NOT like uneven CAS settings. So only even clocks like CL14 or CL16 for first memory setting. The other settings can be odd with no issues. If you want to run odd CL for the first memory timing you must set Geardown to disabled and power off the system after changing in the BIOS to set it.

Extreme power phase is the most stable as it runs all the VRMs all the time. LLC4 or LLC5 allows you to drop Vcore a ton because you don't have to account for Vdroop. You should run Extreme for CPU and at least Optimized for the SoC. The SoC runs without much droop at LLC3. I would run at least 1.10V for any memory overclock greater than 3066. I run a +0.0125V offset at LLC3 for my Vsoc to settle around 1.106V. ProcODT at Auto or 53.3 or 60 ohms. Vdimm at 1.38V. That enables me to run 3333 Mhz at 14-14-14-30-52-1T Geardown enabled for the G. Skill F4-3600C16D-16GTZ memory kit I'm running. The PC is loaded 24/7 with SETI AVX apps and 3 GTX 1070s running full out all the time. Max Tdie temps have been 62-66° C. Cooled with a Corsair H110i AIO. Soon to change to a XSPC AX360 kit. Max memory temps have been 42° C. The CPU is at 3.9 Ghz at 1.344V which is achieved with a -0.00625V offset at LLC5. This is week stable with my full-time BOINC load.

 

[Thentilian]

Just noticed your thread here. I know I posted to you over in my thread but I thought I would jump in here with some info. Ryzen does NOT like uneven CAS settings. So only even clocks like CL14 or CL16 for first memory setting. The other settings can be odd with no issues. If you want to run odd CL for the first memory timing you must set Geardown to disabled and power off the system after changing in the BIOS to set it.

Extreme power phase is the most stable as it runs all the VRMs all the time. LLC4 or LLC5 allows you to drop Vcore a ton because you don't have to account for Vdroop. You should run Extreme for CPU and at least Optimized for the SoC. The SoC runs without much droop at LLC3. I would run at least 1.10V for any memory overclock greater than 3066. I run a +0.0125V offset at LLC3 for my Vsoc to settle around 1.106V. ProcODT at Auto or 53.3 or 60 ohms. Vdimm at 1.38V. That enables me to run 3333 Mhz at 14-14-14-30-52-1T Geardown enabled for the G. Skill F4-3600C16D-16GTZ memory kit I'm running. The PC is loaded 24/7 with SETI AVX apps and 3 GTX 1070s running full out all the time. Max Tdie temps have been 62-66° C. Cooled with a Corsair H110i AIO. Soon to change to a XSPC AX360 kit. Max memory temps have been 42° C. The CPU is at 3.9 Ghz at 1.344V which is achieved with a -0.00625V offset at LLC5. This is week stable with my full-time BOINC load.

I thought the VRM's were always all on. Does that cause extra heat on them to be 'always on'? I can barely get my RAM to 2800, but it's Hynix memory I think and Dual Rank, so it's not a very good set of RAM to OC to begin with from what I've read. I have my SOC set to 1.06v manually with no LLC. I have .13750 offset on CPU with 38.00 multiplier, and LLC 1, with optimized VRM settings. Maybe I'll play with those settings, I just didn't want to go aggressive and accidentally fry something!

I was aware of the CL timings. If you even set an odd CL, it'll change it to an even one, without GD mode on, at least that's what it did to me. Thanks for the information. Gives me something else to tweak and see if I can't get higher RAM or CPU OC's messing with those other settings.

 

[Keith Myers]

I thought the VRM's were always all on. Does that cause extra heat on them to be 'always on'? I can barely get my RAM to 2800, but it's Hynix memory I think and Dual Rank, so it's not a very good set of RAM to OC to begin with from what I've read. I have my SOC set to 1.06v manually with no LLC. I have .13750 offset on CPU with 38.00 multiplier, and LLC 1, with optimized VRM settings. Maybe I'll play with those settings, I just didn't want to go aggressive and accidentally fry something!

I was aware of the CL timings. If you even set an odd CL, it'll change it to an even one, without GD mode on, at least that's what it did to me. Thanks for the information. Gives me something else to tweak and see if I can't get higher RAM or CPU OC's messing with those other settings.

Both Auto and Optimized in the Power Phase settings utilize "phase shedding" per Elmor. That is, it turns off VRMs if the amp capacity is not needed. Unfortunately it does not respond fast enough to be stable when transitioning from idle to a high power state as when benchmarking, gaming or stability testing. Having them on all the time adds maybe 1-2° C. and maybe 1-2W in idle. Well tested by a lot of expert users. Also higher LLC settings have been proven with both digital multimeter measurements and O'scope screenshots to not cause overshoot or harm on the Prime X370 Pro. May not be the case with other motherboards. Extensive testing has been done at high load and stability testing to show that the voltage measured under the CPU socket on the caps, after the VRM output, is closest to what is set in the BIOS and read out by HwInfo64 SVI2 TFN readouts when LLC4 or LLC5 is used. This has been verified by several posters of high technical skill over at OC.net. HwInfo64 SVI2 TFN readouts are directly from the telemetry internal to the VRMs themselves. It is the closest to the actual delivered voltages other than directly measuring at the input to the CPU.