Ryzen 1600x not running at expected frequencies

[DarkRogue]

Hey guys,

I'm trying to set up a new PC to use since my current one is dying. I've gotten all the parts and it's built and stuff, but I'm doing preliminary stress/stability testing before I shift over to it as a daily driver. Unfortunately, I'm running into some weird quirks and I'm not too sure what's going on, or how to deal with it.

Relevant Hardware:
Mobo: ASUS ROG Strix X370-F with latest BIOS
CPU: Ryzen 1600x at stock
Heatsink: Scythe Kotetsu (with Arctic MX4)
RAM: G.Skill FlareX 2x8GB DDR4-3200 (14-14-14-34-1T)
GPU: Not really relevant for this issue, but borrowing an R9 Fury X until GPU prices calm the hell down
PSU: Seasonic Focus+ Gold 850W
OS: Windows 10 x64 v16299.98 with 'High Performance' Power Profile

So, the issue: As far as I am aware, the 1600x is supposed to come at 3.6GHz base, have an all-core XFR of 3.7GHz, a stock boost of up to 4.0GHz, and a single core XFR of 4.1GHz, similar to the 1800x. Correct me if I'm wrong here.

Anyway, I was confused over why when loading CPU-Z and the RyzenMaster program that my CPU was listed as basically being pegged at 3.7GHz no matter what. When idle it was 3.7GHz, but this is explained by the fact that the power profile sets minimum CPU performance at 100% -- if I drop this minimum to 5%, it will drop itself down to ~2.1GHz when idle.

When loading up the CPU with 12 threads via Prime95, it also stays pegged at 3.7GHz, which I suppose is also normal given that 3.7 is its all-core XFR speed. Temps are also fine as far as I can tell, as RyzenMaster reports CPU temps not exceeding 65C during Prime95.

So far it seems normal. The problem comes when I try to only load a single core/thread - Running one thread of Prime95 also pegs my CPU at 3.7GHz. What gives?!

I thought it might have been a sensor reading issue, so I tried yet another program -- HWinfo. I set it to update twice per second (500ms) to try to catch fluctuation more easily. However, it basically showed the same thing -- I was stuck at 3.7GHz. It showed that various cores were individually attaining a maximum of 4.1GHz, which seemed to indicate the CPU was trying to turbo up, but they would never stay there for more than a quarter of a second, and the 'average' speed was reported to actually be closer to 3.6GHz, even for the core that received over 80% of the workload.

I decided to instead try running Cinebench R15 so I could compare against benchmark values from AT, and also Guru3D.

For Cinebench's MT test, I got a score of between 1233-1242, which is in line with the numbers in the reviews on both sites.

However, when running the ST test (whether limiting it to 1 thread, or simply using its Single Core test) I consistently came up with a score of 149-150, which is noticeably below the 1600x ST benchmarks of either site (~160,) and is in line with the 1700x's ST performance instead.

HWinfo's sensors once again reported during Cinebench's ST test that each core was able to very briefly hit 4.1GHz, but the average speed over the entire test (I reset HWinfo's sensor numbers between each test) was closer to 3.6GHz -- for a ST test!

I'm out of ideas on what to try right now, and I'm not sure why my ST tests are underperforming.

I should mention that I've strived to leave as many settings related to overclocking and such on default/auto/off in the BIOS. I've changed settings relating to system boot priorities, logo showing during POST, etc, but none of these should impact the CPU.

The biggest thing is that I had to enable "D.O.C.P. Standard" in order to get my RAM running at the rated 3200MHz of its kit. I couldn't find any other way of enabling XMP (and it turns out that DOCP is supposed to be ASUS' XMP implementation on AMD systems.) I don't think DOCP should impact CPU OC settings, but I'm not entirely sure; At the least, I don't feel like the CPU would be boosting to 4.1GHz at all if 'OC Mode' was enabled.

I know the 'higher' DOCP settings (level 1~5) seemed to try for mild levels of OC (whether it's restricted to the RAM or not, I'm not sure) but as far as I can tell, DOCP Standard simply loaded the XMP data straight from the RAM without trying to OC anything else.

Does anyone have any ideas? I'm pulling my hair out over here and it is frustrating the hell out of me.

 

[StinkyPinky]

Did you see this?

https://community.amd.com/thread/214796

 

[coercitiv]

Does anyone have any ideas? I'm pulling my hair out over here and it is frustrating the hell out of me.

First things first: yes, stock MT frequencies should be 3.7Ghz and ST should be 4 or 4.1Ghz. I use a 1600X machine but honestly never bothered to see if sustained ST was 4 or 4.1Ghz.

My advice would be to reset BIOS to defaults or "optimized defaults" as they're usually called. Only change the settings necessary to boot properly. Leave RAM at default for starters, no A-XMP. You want to isolate the BIOS setting that stops the CPU from going over 3.7

• If optimized defaults works properly, you know what to do: only change one setting at a time towards your desired configuration and observe when ST clocks become stuck at 3.7Ghz
• If problem stays the same at full default... then at least you know you need to look into BIOS options you haven't yet changed, most likely related to overclocking or power management / CPU features.

This is a motherboard & BIOS specific problem, it's bound to have a solution within BIOS.

 

[TheELF]

Opening post is not very clear so I will just add this,you will have to use affinity on prime or cine to keep them running on more then one core,windows shuffles threads around cores as a way of wear leveling,if you run something that has only one thread but multiple cores peak up to max clocks than that's the "problem" with affinity it should only peak that one core.
Also make sure nothing else is running because if an other cores gets some activity it will go to 2 or 4 core turbos.

 

[DarkRogue]

Ugh. All right, this is driving me insane.

@StinkyPinky
No, I haven't seen that thread, thanks for linking it. Judging from the responses there, it sounds like they are suggesting power profiles may be playing a part. However, I have gone ahead and tried/re-run the CB15 benchmark repeatedly on every power profile except for Power Saver (so, Balanced, Ryzen Balanced, and High Performance.) I even went ahead and created two new custom profiles using the Balanced and High Performance options as its base.

Every combination ended up giving me the same results -- 3.7GHz Multithreading with expected scores, <3.6GHz Singlethreaded tests with underperforming scores.

@TheELF
Assgning an affinity to the Cinebench process via Task Manager made no difference. I tested setting it to CPU0, CPU2 and CPU4 individually; Single threaded scores were all exactly the same (or +/-1 variance, which is within the margin of error) and HWinfo continued to show that the workload was bouncing around between cores/threads regardless.

@coercitiv
As much as 'resetting everything to default' annoys me because I'd have to go back over everything again, I went ahead and did so just to rule everything out.

Unfortunately, it still made no difference, relatively speaking. Multithreaded benchmark still pegged the CPU at 3.7GHz across the board, and Singlethreaded benchmarks was still underperforming with average clock speed under 3.6GHz. Multithreaded scores dropped to around 1201-1210, while Singlethreaded scores dropped to 145-147, I believe largely because of my 3200MHz RAM now running at 2400MHz at default BIOS settings.

Literally everything is stock and default now.. I'm not sure what else I can do.

 

[DarkRogue]

All right! I think I've fixed it and gotten it working. Damn, that was annoying.

Just in case anyone else runs into this problem in the future and winds up over here, or if anyone is simply curious:

Setting 'DOCP Standard' to enable the XMP of the memory does not have any impact on the CPU's ability to turbo up and down, as far as I can tell. However, based on my observations, there was clearly some issue with getting Windows to leave a single thread on any one core long enough for it to turbo up to its XFR, which was why running any ST test resulted in average clockspeeds being lower than even 3.6GHz on average.

I went looking around for solutions to this, and came across this reddit thread which explained what was going on, as well as providing a solution/workaround.

I gave it a shot and followed its instructions, using the RightMark PPM Panel to set those settings he suggested, with the exception of 'Core Override' since that might be a Windows 7 setting; no such setting appeared for me in Windows 10.

In either case, straight away after setting those, I ran my benchmarks again, and my numbers were right in line with the review scores - ST CB15 scores came back between 157-161, and HWinfo was reporting that two cores held, on average, close to 4.1GHz for the entire duration of the benchmark. The load was still bouncing around between the four threads in the first two cores periodically, but it was entirely localized to those two cores now.

I think I'm happy enough with this to just keep it and go with it, but I find it strange that I had to do this, because I had thought that AMD had already set these/fixed this issue with their official 'Ryzen Balanced' power plan, which I ended up using again, and subsequently modifying with the aforementioned tweaks. Apparently not, I guess?

The one wrinkle is that maybe the new Windows 10 update messed it up somehow, because since this is a new build and I'm coming from Windows 7, I decided to just go all the way to the newest Fall Creator's Update to avoid any 'in-place upgrading' nonsense. Perhaps this update messed with that power profile somehow, as I only noticed last night that AMD released a new set of chipset drivers a couple days ago, presumably for the FCU, although I did not see any option to install a new power profile while I was installing it.

As of now, though, I just have to decide what I want to do with the CPU's idling; Within the power profile I can set the minimum CPU performance down and have the 1600x idle at 2.1GHz, as opposed to idling at 4.0GHz at 100%. However, from a power-usage standpoint, it only amounts to a 2-3w savings at best, and ST performance is impacted slightly (155-158 CB15 vs 157-161 otherwise.) On the other hand, the CPU is eating close to 1.4v, and I'm not sure if it's safe to have it idling at that much voltage... it only uses about 0.8v while idling at 2.1GHz apparently.

 

[IEC]

You are seeing likely seeing the VID (voltage requested by CPU) and not the actual vCore being provided to the chip. It will be fine.

 

[DarkRogue]

You are seeing likely seeing the VID (voltage requested by CPU) and not the actual vCore being provided to the chip. It will be fine.

The VID is indeed listed as around 1.4v, but I was looking through HWinfo for other values that might be closer to the actual voltage being supplied, and the SVI2 TFN values are about the same, averaging around 1.385v and going up to 1.4v. The VDDCR CPU value fluctuates more frequently, jumping between 0.89v and 1.405v, with its 'average' value around 1.32v.

Being that they are all kind of around there, I was concerned about whether or not it might be too much voltage to have running constantly. At the least, I historically don't upgrade PC's for between 5-8 years (I mean, look at my Ivy Bridge,) so I needed to account for longevity; if 1.4v is fine and won't cause issues for that long, though, then I'll probably leave it, but I'm just not sure right now.

 

[PhonakV30]

My cpu : 1600x running at 3950@1.3 volt
ST = 161
MT = 1329
DDR4 2933 Cl 15-15-15-30-50
Just set manually your frequency, sometime I see 4100@1.55v when CB is running for ST score.

 

[DarkRogue]

AFAIK if I touch any of the settings to set frequency/multiplier/voltage/whatever, the CPU enters 'OC mode' and it loses all of its energy saving features and such, doesn't it? On top of having to stability test everything.. I did that to OC my 3770k all those years ago, but I really don't want to waste all that time doing that again, which was why I got the 1600x to just get "an OC'd 1600 without having to do any OC."

 

[Gideon]

Thank you so much for the info DarkRogue! I had the same issue with my 1700X being stuck @3.5 Ghz. I had tried a zillion different things in BIOS and with power profiles, but nothing helped.

With this I finally got my 1700X to turbo up to 3.9 Ghz in Cinebench ST and CPU-Z bench! IMO it's ridiculous that even with Ryzen chipset drivers and Power Plan and the latest bios, the Single Core Turbo just doesn't work flat-out in Windows 10. People should not have to edit stuff in Rightmark PPM panel.

Regardless a very good find indeed!

 

[DarkRogue]

Glad it was helpful, and yeah its kind of dumb how you still have to go through some hoops to get it working properly. On the other hand, I bought everything over Black Friday for pretty darn good discounts, so I guess I have to work a little bit for it.

I would monitor your voltages and temps for a bit though; I still haven't decided if I really like it idling at 1.4v or not.

 

[Gideon]

I would monitor your voltages and temps for a bit though; I still haven't decided if I really like it idling at 1.4v or not.

Yeah, that definitely doesn't sound right. I remember having problems with that as well. If you look at the Ryzen balanced power plan, you'll see the minimum processor state is set to 90% by default. You can lower this number and you should see the core speeds drop as well as voltage when you're idling. For reference, the default windows balanced power plan has the minimum processor state set to 5% at default.

Once I had:
1) Cool'n'Quiet (and all other power saving features) working in BIOS
2) Minimum processor state is set to 5% in the Ryzen Balanced power-profile.

It started downclocking properly to 2000 Mhz with voltage fluctuating between 0.9v-1.2v on idle. Another frustrating thing, that IMO should be working out of the box. In fact, it even works when overclocking

 

[IRobot23]

Maybe it is MB specific, since r7 1700 has 3,2GHz all core turbo on C6H

 

[DarkRogue]

Yeah, that definitely doesn't sound right. I remember having problems with that as well. If you look at the Ryzen balanced power plan, you'll see the minimum processor state is set to 90% by default. You can lower this number and you should see the core speeds drop as well as voltage when you're idling. For reference, the default windows balanced power plan has the minimum processor state set to 5% at default.

Once I had:
1) Cool'n'Quiet (and all other power saving features) working in BIOS
2) Minimum processor state is set to 5% in the Ryzen Balanced power-profile.

It started downclocking properly to 2000 Mhz with voltage fluctuating between 0.9v-1.2v on idle. Another frustrating thing, that IMO should be working out of the box. In fact, it even works when overclocking

Sorry, you misunderstand; I already know that you can adjust the idle speed by changing the minimum processor performance number. My issue was that setting it down lower (even 60% is enough to peg it at 2.1GHz, going down to 0% doesn't make any difference) does have it idle at 0.8v instead of 1.4v, but there's hardly any power savings, and it definitely does impact ST performance slightly, whereas leaving it at 90-100% causes it to idle at near 4.0GHz consuming 1.4v all the time.

My problem was more wondering if letting it idle at 1.4v (granted only on one or two cores?) was safe in the long term since it would keep it at the best performance right now.

Maybe it is MB specific, since r7 1700 has 3,2GHz all core turbo on C6H

I don't think it should be motherboard-specific, at least when we're talking about the CPU's XFR and boost frequencies. The different models have their own specific base clock, boost and XFR frequencies. 3.2GHz for a 1700 non-X model sounds about right, as it has the lowest clock speeds out of the 1600x, 1700, 1700x and 1800x. The 1700x has slightly higher clockspeeds compared to the 1700, and then the 1600x and 1800x have the same/highest clock speeds out of the box.

 

[IRobot23]

Sorry, you misunderstand; I already know that you can adjust the idle speed by changing the minimum processor performance number. My issue was that setting it down lower (even 60% is enough to peg it at 2.1GHz, going down to 0% doesn't make any difference) does have it idle at 0.8v instead of 1.4v, but there's hardly any power savings, and it definitely does impact ST performance slightly, whereas leaving it at 90-100% causes it to idle at near 4.0GHz consuming 1.4v all the time.

My problem was more wondering if letting it idle at 1.4v (granted only on one or two cores?) was safe in the long term since it would keep it at the best performance right now.



I don't think it should be motherboard-specific, at least when we're talking about the CPU's XFR and boost frequencies. The different models have their own specific base clock, boost and XFR frequencies. 3.2GHz for a 1700 non-X model sounds about right, as it has the lowest clock speeds out of the 1600x, 1700, 1700x and 1800x. The 1700x has slightly higher clockspeeds compared to the 1700, and then the 1600x and 1800x have the same/highest clock speeds out of the box.

No 3,2GHz is not correct for R7 1700. It should at max 3.05GHz.
3GHz is all core turbo + 50 MHz XFR for non X models.

 

[DarkRogue]

Well, I don't own a 1700, so I wouldn't be able to say definitively, but a cursory look at reviews seems to indicate that 3.2GHz is the standard all-core turbo frequency for that CPU.

 

[IRobot23]

Well, I don't own a 1700, so I wouldn't be able to say definitively, but a cursory look at reviews seems to indicate that 3.2GHz is the standard all-core turbo frequency for that CPU.

Did reviewers use low end MB?
Maybe it the problem on asus side.

Anyway why doesn't he just OC that CPU?

 

[Topweasel]

Well, I don't own a 1700, so I wouldn't be able to say definitively, but a cursory look at reviews seems to indicate that 3.2GHz is the standard all-core turbo frequency for that CPU.

Yeah my 1700 has been doing encoding 24/7 for about 2 months now. Constantly at 3.2GHz.

3GHz is the base clock.

Also XFR is 200MHz on X chips, 100MHz on non X chips. 50MHz is the increment it can work in.

 

[IRobot23]

Yeah my 1700 has been doing encoding 24/7 for about 2 months now. Constantly at 3.2GHz.

3GHz is the base clock.

Also XFR is 200MHz on X chips, 100MHz on non X chips. 50MHz is the increment it can work in.

XFR for TR is 200MHz
Ryzen X is 100MHz
Ryzen non X is 50MHz

I will try to disable turbo on my asus board. I do get 3.2GHz turbo all core yet I get 3.75GHz on single core. So it depends on MB settings.

I don't know why OP does OC to 3,9GHz all core turbo (p-states or not).

 

[Topweasel]

XFR for TR is 200MHz
Ryzen X is 100MHz
Ryzen non X is 50MHz

I will try to disable turbo on my asus board. I do get 3.2GHz turbo all core yet I get 3.75GHz on single core. So it depends on MB settings.

You are right it's kind of mess because the 1500x does have a 200mhz XFR. I was thinking about the difference between the 1700x and 1800x being 200mhz even though the base clock is 100MHz different.

But XFR only works on 2 cores or at least early reviews say so. So for the 1700 for example it should be 3GHz base (doing some work), 3.2GHz all core turbo, 3.7GHz 2 core turbo, 3.75GHz XFR 2 core turbo if you have enough cooling.

 

[IRobot23]

Its like on lga 1151 z370 mce. Yet there is way above its tdp even at base turbo.

 

[DarkRogue]

I don't know why OP does OC to 3,9GHz all core turbo (p-states or not).

Uh.. I don't recall ever saying this. I haven't OC'd my CPU at all. I was only trying to get its stock XFR/turbo frequencies working as it was intended. The 1600x, like the 1800x, has an all-core boost of 3.7GHz, meaning all 6 cores run at that speed when fully loaded. When only a single core is loaded, XFR pushes it to 4.1GHz -- or was supposed to, but didn't until I tweaked the power settings.

All the various boost states for the different CPU's are a bit confusing, but no, the reviewers who got 3.2GHz with the 1700 I don't believe were using 'low-end' motherboards. It's just its standard boost, as far as I can tell.

 

[IRobot23]

Then your mb setting ( have own features) or cpu run to hot for xfr to work.


If you are thinking that your cpu is bad go and RMA it. After you will get new one same thing can happen, but when you get that one it works great go and OC it.

XFR is not limited to TDP. It works when your cpu has some heat headroom. What is your cooler and can you run CBr15.


I don't know why are you guys making scene out of this, but all it matter is if you won silicon lottery. Maybe you can run 4,1GHz with 1,4-1,425V on load

 

[cellarnoise]

OP, thanks! I had this same issue and that program you suggested fixed it. Also did not have all the options under Windows 10.
I had tried going back many bios and tried different power plan, bios settings, different benchmark software. Nothing would bring back the 2 core turbo boost that I had seen originally when I first built this 1700x , msi b350 pro carbon build.

I believe 2 core precision boost went away after a windows update also. I don't always want a bios /uefI overclock as this msi b350 board won't down volt the board voltage when idle in oc mode, I think for o.c. stability reason. Had i known this i woukd have bought a different board and still may when the newer chioset come out. It loses some power saving functions.