AMD RYZEN Builders Thread

[Valantar]

Because there's always some sort of activity. Processor is never truly idle. This is normal thermal variance. Check out core temp and look at how there's some slight load on your cpu. Any sort of load, even if it's 1-5% will cause this. Even from moving your mouse cursor.

It doesn't look like normal temperature variance - it's far too systematic and non-random for that. It never stays at the same temperature for more than a second or so, and always changes in the same pattern, never ramping up, always jumping up by ~10 degrees instantly and cooling down again over the next 7-8 seconds.

I noticed that too, at idle. I'm also on water. Not sure what it is. Sensor anomaly? It doesn't seem to happen at nearly-full load.

Yeah, that's what I've noticed too. The fluctuations are too large and systematic for simple thermal variations, so I guess it must be something like that.

Edit: another oddity I've noticed is that when I keep Ryzen Master open for temp monitoring, after a while it pegs the 12th CPU thread at ~90% usage and keeps it there seemingly indefinitely. It goes away after a restart of the app (or, rather, drops to ~20-30%). Has anyone else seen this? In general, though, RM seems a bit unstable (some times refusing to un-minimize after the computer has been sleeping, for example).

 

[Despoiler]

Because there's always some sort of activity. Processor is never truly idle. This is normal thermal variance. Check out core temp and look at how there's some slight load on your cpu. Any sort of load, even if it's 1-5% will cause this. Even from moving your mouse cursor.

Probably this. Is your loop PWM in anyway? Pump, rad fans? If so it could be them adjusting as well.

Ryzen master eats more CPU than it should when open. Don't leave it open.

 

[Topweasel]

Ryzen master eats more CPU than it should when open. Don't leave it open.

This. Whomever was quoted as saying that influence an experiment by observing it, must have been using Ryzen Master.

 

[Valantar]

Probably this. Is your loop PWM in anyway? Pump, rad fans? If so it could be them adjusting as well.

Ryzen master eats more CPU than it should when open. Don't leave it open.

Again, the fluctuations are far too systematic and short (a 40-30-40 degree cycle takes perhaps ten seconds, which is far too quick for fan or pump speed to affect it unless fluctuations were huge and very audible) for that to be the case. The fan on my 120mm rad (right after the CPU in the loop) is PWM, but it's set to run at ~900rpm until the CPU reaches 45 degrees, so it's not budging during this. The two other fans (on the 240mm intake rad, after the GPU in the loop) are controlled by a manual fan controller (for now). This happens regardless if I run those at either of the three settings of the controller (although the exact temps might differ). The pump is set at 100%, as it's already on the weak side for a CPU+GPU+two rads loop.

This. Whomever was quoted as saying that influence an experiment by observing it, must have been using Ryzen Master.

Haha, that made me laugh. Although perhaps we are affecting Ryzen Master by observing it?

 

[ScottAD]

Used the 1.64 Beta BIOS on my MSI B350 Tomahawk, this is the AGESA 1.0.0.6 update. 2933...FINALLY!

 

[JimmiG]

This. Whomever was quoted as saying that influence an experiment by observing it, must have been using Ryzen Master.

RM can easily load one full core on my system on its own.
Temp fluctuations are normal with Ryzen due to SenseMI. It depends on the motherboard, which sensor it uses and how it reports the temp.

 

[noneis]

Ryzen temperature reporting issue: (R7 1700)

It looks like there is thermal offset that is applied if at least one core is at P0 state and thermal dissipation (or cpu utilization?) for that core is above certain threshold. This offset is then gradually decreased if thermal dissipation (or cpu utilization?) for that(all) core(s) is below certain threshold.

If only P0 is allowed, offset is applied almost instantly on load and offset isn't gradually decreasing during load.
If only P2 or P1 or P1&P2 are allowed offset is not applied.
If single thread is utilized to 100% and jumps between cores, offset is gradually decreasing.
If single thread is utilized to 100% and assigned to single core (to one or both threads), offset is not gradually decreasing
If same workload is given to fewer cores there is much higher probability, that offset will be applied.

 

[Indus]

Apparently I won the silicon lottery. My chip boots up at 4.1 @ 1.50v on air and runs windows and web surfing fine but its not stable under cinebench. I may need more voltage or better liquid cooling.

But so far my most stable overclock keeps coming back to 3.85 (38.50 x 100) at 1.360V and I think 1.36 is good enough for a 24/7 overclock.

 

[richierich1212]

Apparently I won the silicon lottery. My chip boots up at 4.1 @ 1.50v on air and runs windows and web surfing fine but its not stable under cinebench. I may need more voltage or better liquid cooling.

But so far my most stable overclock keeps coming back to 3.85 (38.50 x 100) at 1.360V and I think 1.36 is good enough for a 24/7 overclock.

1.5v will require a better-than-water cooling solution unfortunately.

 

[Topweasel]

Some users have noticed at 1.45v and 1.5v that after extended runs that their CPU degraded a bit and no longer run stable at numbers they hit before that run. Even if it looked like my CPU could take it I wouldn't chance it. You probably need extreme cooling (like some kind of refrigeration system) to avoid permanent damage at that voltage level.

 

[Indus]

So what's the absolute max safe voltage level?

 

[Reinvented]

So what's the absolute max safe voltage level?

AMD said that nothing more than 1.4v or something along those lines for extended periods.

And per your other post - 4.1 @ 1.5v isn't the lottery. If it's not stable, it's not really hitting the jackpot period.

 

[Indus]

Something strange happened.. my overclock is not stable anymore even 3.85 at 1.36v but I noticed something.. the cpu voltage seems to fluctuate from 1.28 to 1.36. I'm thinking the power supply is failing. When it stays steady the oc is stable.

I might have to replace the Power Supply.

 

[Reinvented]

Something strange happened.. my overclock is not stable anymore even 3.85 at 1.36v but I noticed something.. the cpu voltage seems to fluctuate from 1.28 to 1.36. I'm thinking the power supply is failing. When it stays steady the oc is stable.

I might have to replace the Power Supply.

Probably because Cool n Quiet is still enabled, that's why it fluctuates.

 

[DrMrLordX]

1.45v is classified as the breakdown voltage point for Summit Ridge. Approach that voltage at your own peril.

FWIW mine will bench stable @ 4075 MHz with less than 1.5v but won't boot 4100 MHz to save its life. Oh darn!

 

[Malogeek]

So I decided to replace my 10-year old case and got an NZXT S340 for $50 on sale, which for the case seemed to be a nice bargain. In addition, I bought some copper heat sinks for my RD400 NVMe drive which I've had issues with overheating.

Swapped over all the parts tonight and whilst I had a little drop in overall CPU temps with much better airflow, the main reason was to get ready for an AIO. My NVMe drive however dropped 15 degrees! Solved that problem.

 

[Eric1987]

Gonna try to get my 3400MT RAM up to par tomorrow morning. So on my Asus X370 Pro put the SOC up to 1.15 and DRAM to 1.4? I saw that suggestion.

 

[JimmiG]

Gonna try to get my 3400MT RAM up to par tomorrow morning. So on my Asus X370 Pro put the SOC up to 1.15 and DRAM to 1.4? I saw that suggestion.

Will be interesting to hear about your results. I can't do 3600, 3466 or even 3333 with my LPX 3600 RAM on the X370 Prime Pro. The AIDA64 memory stress test is what kills it. It fails after anywhere from 2 - 5 seconds at 3600 and 3466, to ~2 hours at 3333.
You can go as high as 1.2V SOC and 1.45V DRAM, but 1.15/1.4 are what I would consider "safe" for 24/7. Of course if you can get by with 1.1/1.35 that's even better/safer.

 

[repoman0]

Got my build up and running yesterday after grabbing that $260 chip deal. I swapped out a 6700k with Asrock Z170M Extreme4 for a 1700 and Asrock AB350M Pro4. Wouldn't boot at first but it was because I had failed to fully seat the memory, which I reused, then was stuck in single channel because the slots are labeled weird .. I would have thought A1 and A2 go together, but I guess A1 and B1 makes sense too Once it was working I immediately updated the BIOS via internet flash before even booting to Windows. Then Windows booted right up after spending 15-30 seconds recognizing and installing the new hardware, and even remained activated.

Next step is to figure out this P-state overclocking business, because I want the voltage and clock speed to drop at idle and the normal overclocking options on this board don't allow for that. Anybody get this to work on this board or another B350 board? I messed with it a little bit but got no boot and had to clear the cmos.

 

[Space Tyrant]

Next step is to figure out this P-state overclocking business, because I want the voltage and clock speed to drop at idle and the normal overclocking options on this board don't allow for that.

You can also ensure that "Cool & Quiet" is enabled in Advanced Options(?). That will drop idle cores to 1550.

 

[repoman0]

You can also ensure that "Cool & Quiet" is enabled in Advanced Options(?). That will drop idle cores to 1550.

Yep, it works fine at stock/auto settings but if I change to manual voltage and frequency mode it seems to disable CnQ ... could be a quirk with the board but I was curious if I could use the P-states to get around it for now.

 

[Space Tyrant]

Yep, it works fine at stock/auto settings but if I change to manual voltage and frequency mode it seems to disable CnQ ... could be a quirk with the board but I was curious if I could use the P-states to get around it for now.

I'm also using Asrock AB350M Pro4 and have CnQ running with my overclock of my R5 1600. Oddly, it didn't disable it when I set manual mode. I *think* I was running BIOS 2.40 at the time, but currently 2.50.

Edit: have you tried manually reenabling CnQ?

 

[repoman0]

I'm also using Asrock AB350M Pro4 and have CnQ running with my overclock of my R5 1600. Oddly, it didn't disable it when I set manual mode. I *think* I was running BIOS 2.40 at the time, but currently 2.50.

Edit: have you tried manually reenabling CnQ?

Hmm, good to know. I'll give it another try after work.

 

[Indus]

Got my build up and running yesterday after grabbing that $260 chip deal. I swapped out a 6700k with Asrock Z170M Extreme4 for a 1700 and Asrock AB350M Pro4. Wouldn't boot at first but it was because I had failed to fully seat the memory, which I reused, then was stuck in single channel because the slots are labeled weird .. I would have thought A1 and A2 go together, but I guess A1 and B1 makes sense too Once it was working I immediately updated the BIOS via internet flash before even booting to Windows. Then Windows booted right up after spending 15-30 seconds recognizing and installing the new hardware, and even remained activated.

Next step is to figure out this P-state overclocking business, because I want the voltage and clock speed to drop at idle and the normal overclocking options on this board don't allow for that. Anybody get this to work on this board or another B350 board? I messed with it a little bit but got no boot and had to clear the cmos.

Holy shit.

I've heard horror stories about blue screens like CPU is not registered on this install.

 

[SK10H]

Using Asus ROG Hero 1401 beta BIOS, the GSkill F4-3200C16-16GVK double-sided hynix 16-18-18-38 memory can finally be run at 3200 strap, using auto ProcODT will not boot. I set it to ProcODT=60 Ohm with custom timing that I grab by running 2933 strap and overclock the bclck to mirror the secondary timing.

vcore 1.46 at 3.9GHz
4 x 16GB memory at 3120 14-14-14-16-34-72-6-9-33-4-12-23-33, 2T
Soc 1.15v
vdimm 1.475v
vttddr 0.75v
vddp 1.035v