[SOLVED] Why are my cpu volatges so low?

[MXGamer80]

So I went to overclock my R5 3600 to 4.2Ghz at 1.38 volts. It started up fine, and went to stress test. For about 3 minutes it ran stable in prime95, then all of a sudden there is an error with thread 6 and 7 (they ran really unstable). I didn't think much of it as I thought 1.38 wasn't enough (forgot to mention that this was before I looked at voltages in hwmonitor). So i tried 1.39 volts, and the same thing. Then I tried 1.4 volts, here is when I decided to check my voltages. So I ran the test, and I noticed only 1.35 voltages were going through according to hwmonitor.The same thing happend,thread 6 and 7 were failing (running insanely unstable!). So I'm confussled now. I don't understand why only 1.35 volts are being sent to the cpu when in bios the voltage reads out 1.4 volts. My best guess to threads 6 and 7 failing is due to the low voltage.But I still don't understand why it' reading low voltage in hwmonitor. I have never had this happen to me before when overclocking processors, so I don't know how to troubleshoot this.I played BF1 to see if threads 6 and 7 were fine, and the percentage was normal (percentage matching other cores).

Specs:
Ryzen 5 3600
Adata XPG Z1 DDR4 2x8gb 3200MHZ CL16
XFX GTS Black Edition RX580 8GB 1405Mhz
MSI B450 Gaming Plus Mobo
Corsair Vengeance 650M PSU
HP S700 250GB SSD (Windows 10 1903)
WD Black 1TB 7200 RPM

 

[drivinfast247]

Use HWINFO64

 

[GilGrissom]

Try increasing LLC

 

[MXGamer80]

Try increasing LLC

What is LLC?

 

[GilGrissom]

What is LLC?

What is LLC?

On MSI board,set Load Line Calibration on Level 4 or 3 or 2,depends on how much core voltage you set.I had the 2600X overclocked to 4.2GHz with 1.4v core voltage and Load Line Calibration on Level 4.If you go with lower voltage,like 1.385,you might want to raise LLC to Level 3,and so forth. You probably have yours on Auto,so it's lowering the voltage to it's stock value,I think.Check that out anyway

 

[MXGamer80]

Ohhhh, that makes a lot of sense. Auto settings suck imo. But I'll check that out. Thx

 

[drea.drechsler]

Ohhhh, that makes a lot of sense. Auto settings suck imo. But I'll check that out. Thx

Three things: LLC on MSI's B450 boards kinda suck, so don't expect a whole lot of help from it. But what it's designed to do is provide a correction for VDroop...the tendency of the VCore voltage to drop from a low load condition to a high load condition.

Second... use HWInfo64 and when you do monitor the (SVI2 TFN) core voltage, that's the more correct voltage the processor is seeing when under load. Even if you think it's 1.35 V it could be something more like 1.25 V because it's not the true internal voltage the processor is seeing.

Third... beware Prime95. It can present a highly unrealistic power load because it uses tightly nested, AVX instructions in looping routines, way more tightly nested and longer loops than ever occur in a "real" program. An H.264 or H.265 encode is much more realistic use of AVX. P95 is widely considered a power virus... good for de-rating an overclock but not much else. Use something RealBench if you want a much more realistic way to stress test your overclock for stability.

Fun Fact; Intel processors can be set up to reduce clock speed when running AVX instructions to protect themselves. AMD's don't; instead what Ry3K does is throttle back when it hits 95 C. But if you manually overclock you frequently don't give it enough voltage to keep it stable at that temperature so it crashes well before it gets there instead. But that's OK, you don't really want it to operate there anyway (with that high a voltage and high a temperature). That's why you just avoid P95 and use RealBench to stress.

 

[MXGamer80]

Try increasing LLC

I have increased my LLC and the ame thing keeps happening, RealBench, indicates it's unstable. And when watching CPU-Z Some cores would go down to 99 or 98 percent. So my cpu still runs unstable.

 

[drea.drechsler]

I have increased my LLC and the ame thing keeps happening, RealBench, indicates it's unstable. And when watching CPU-Z Some cores would go down to 99 or 98 percent. So my cpu still runs unstable.

When running CPU-Z Stress Test, what is the SVI2 - TFN core voltage your seeing in HWInfo? And what is the SVI2 core voltage you're seeing at idle?

 

[MXGamer80]

When running CPU-Z Stress Test, what is the SVI2 - TFN core voltage your seeing in HWInfo? And what is the SVI2 core voltage you're seeing at idle?

My SVI2 TFN is at 1.375 volts when stress testing. When at idle speed, I'm getting 1.381-1.387 volts (1.387 or 1.385 volts is what I set the voltage to in bios). And also Apparently earlier I was looking at VID voltages for each core. When stress testing with CPU-Z I have no issues, but with Prime95 and RealBench, those two stress testers detect instability.

 

[drea.drechsler]

My SVI2 TFN is at 1.375 volts when stress testing. When at idle speed, I'm getting 1.381-1.387 volts (1.387 or 1.385 volts is what I set the voltage to in bios). And also Apparently earlier I was looking at VID voltages for each core. When stress testing with CPU-Z I have no issues, but with Prime95 and RealBench, those two stress testers detect instability.

RealBench isn't a synthetic stressor: it uses Handbrake (as a background task) encoding video with H.264 or H.265, which uses a lot of AVX instructions. It's that AVX thing killing your processor. CPU-Z is a nice mild stress, something just to see what's going on with voltage but probably won't crash the system.

1.375 isn't too high yet... to get stable you'll probably have to raise voltage more. Upwards of 1.450 even to 1.50 in SVI2 is possible if you don't have really good silicon (and 3600's, kind of by definition, are going to be the worst silicon of all Zen2 chiplets). Thermals are going to go way up too so you better have good...no great cooling, like a 240mm AIO or a huge air cooler.

Now you start to see why people don't think manually overclocking Zen2 is the way to go. A lot of heat and noise, and very little gain.

I can get 4.3 Ghz all-core on my 3700x but only at 1.44 something volts. It's RealBench stable if I have fans at 100% before I start the test and it's under a 240mm AIO cooler. But start up prime95 and it literally pings it crashes so fast...first time it happened I thought I'd blown up the processor, I'm serious. But it was just a crash, it rebooted and I gave up on it. Even if I could deal with the aggressive fan curve it would need in use I can't abide the idea that it's so easily crashed: just start up something with tightly nested AVX loops and boom, it's down and out.

I'm just doing the PBO thing: I only lose about 2% in CB20 multi-thread score by letting the boost algortithm do it's thing and I pick up over 10% in CB20 single thread score. I'm happy with that.

 

[MXGamer80]

RealBench isn't a synthetic stressor: it uses Handbrake (as a background task) encoding video with H.264 or H.265, which uses a lot of AVX instructions. It's that AVX thing killing your processor. CPU-Z is a nice mild stress, something just to see what's going on with voltage but probably won't crash the system.

1.375 isn't too high yet... to get stable you'll probably have to raise voltage more. Upwards of 1.450 even to 1.50 in SVI2 is possible if you don't have really good silicon (and 3600's, kind of by definition, are going to be the worst silicon of all Zen2 chiplets). Thermals are going to go way up too so you better have good...no great cooling, like a 240mm AIO or a huge air cooler.

Now you start to see why people don't think manually overclocking Zen2 is the way to go. A lot of heat and noise, and very little gain.

I can get 4.3 Ghz all-core on my 3700x but only at 1.44 something volts. It's RealBench stable if I have fans at 100% before I start the test and it's under a 240mm AIO cooler. But start up prime95 and it literally pings it crashes so fast...first time it happened I thought I'd blown up the processor, I'm serious. But it was just a crash, it rebooted and I gave up on it. Even if I could deal with the aggressive fan curve it would need in use I can't abide the idea that it's so easily crashed: just start up something with tightly nested AVX loops and boom, it's down and out.

I'm just doing the PBO thing: I only lose about 2% in CB20 multi-thread score by letting the boost algortithm do it's thing and I pick up over 10% in CB20 single thread score. I'm happy with that.

Would this CPU be fine? https://www.amazon.com/gp/product/B...42e-e8f213ef368b&pf_rd_r=02TZ4RM9G57V3JTHR98P
or possible this?
https://www.amazon.com/FSP-Windale-...65239509&s=electronics&sr=1-7#customerReviews

 

[drea.drechsler]

Would this CPU be fine? https://www.amazon.com/gp/product/B...42e-e8f213ef368b&pf_rd_r=02TZ4RM9G57V3JTHR98P
or possible this?
https://www.amazon.com/FSP-Windale-...65239509&s=electronics&sr=1-7#customerReviews

Those are pretty big with a lot of fins, so probably OK. The air coolers I see most people using are Noctua's and Dark Rock... the Pro series I think. I'd suggest go looking through reviews https://www.tomshardware.com/t/cooling/review/ to get ideas. I prefer an AIO with 240 or 280 mm rad's; the water pumps in AIO's don't move enough water to get enough heat off a CPU to saturate a 360mm radiator, at least according to reviews .

But an air cooler is only as good as case ventilation: pay attention to case fans and air flow. Provide plenty of cool air coming in the front and extract it at the top and the rear. If you've an axial flow GPU it's dumping it's heat inside the case so that leaves the CPU cooler increasingly warmer air to work with and doing very little good unless you keep air moving through the case.

 

[MXGamer80]

Those are pretty big with a lot of fins, so probably OK. The air coolers I see most people using are Noctua's and Dark Rock... the Pro series I think. I'd suggest go looking through reviews https://www.tomshardware.com/t/cooling/review/ to get ideas. I prefer an AIO with 240 or 280 mm rad's; the water pumps in AIO's don't move enough water to get enough heat off a CPU to saturate a 360mm radiator, at least according to reviews .

But an air cooler is only as good as case ventilation: pay attention to case fans and air flow. Provide plenty of cool air coming in the front and extract it at the top and the rear. If you've an axial flow GPU it's dumping it's heat inside the case so that leaves the CPU cooler increasingly warmer air to work with and doing very little good unless you keep air moving through the case.

Ok, yeah since I'm on a budget and can't spend more than 60 dollars, I will just go with the FSP windale 6. And I can't fit a 240mm AIO in my case, only a 120mm AIO so that's off the list. Once I get the FSP Windale 6 today, I will test out high voltages, and see how that works. Thx for the help. I've have 2 intakes, and 3 exhausts, so I think airflow will be fine.

 

[drea.drechsler]

Ok, yeah since I'm on a budget and can't spend more than 60 dollars, I will just go with the FSP windale 6. And I can't fit a 240mm AIO in my case, only a 120mm AIO so that's off the list. Once I get the FSP Windale 6 today, I will test out high voltages, and see how that works. Thx for the help.

Hey, one thing I want to point out...

AMD's boosting algorithm is very agressive about boosting to 4.2Ghz (for 3600's) with sufficient thermal and VRM current headroom. But doing so they set voltage up to 1.5 V. That is a lot, but it also does not hold that boost for very long, just milliseconds, and is usually missed by monitoring software that's sampling at 2000 ms intervals.

You're trying to get a sustained all-core boost to your processor's max boost speed. That's not something I've read of many people accomplishing with Ry3k, and usually with sub-ambient cooling. If you're going to try to get even a RealBench stress test to pass at 4.2 G, and push voltage up to 1.5 V if needed to do so and think any "budget" cooling solution will help you. Just be careful cause I'm suspecting your budget can't fit another 3600 processor in it either.

 

[MXGamer80]

Hey, one thing I want to point out...

AMD's boosting algorithm is very agressive about boosting to 4.2Ghz (for 3600's) with sufficient thermal and VRM current headroom. But doing so they set voltage up to 1.5 V. That is a lot, but it also does not hold that boost for very long, just milliseconds, and is usually missed by monitoring software that's sampling at 2000 ms intervals.

You're trying to get a sustained all-core boost to your processor's max boost speed. That's not something I've read of many people accomplishing with Ry3k, and usually with sub-ambient cooling. If you're going to try to get even a RealBench stress test to pass at 4.2 G, and push voltage up to 1.5 V if needed to do so and think any "budget" cooling solution will help you. Just be careful cause I'm suspecting your budget can't fit another 3600 processor in it either.

Oof, might need to keep it at 4 to 4.2 ghz, I won't even try to go over then. I've had some failures with budget 120mm AIO's (For me, the CPU gets hot). So I guess the windale 6 is a step up. I remember checking temps on my old hyper 212 led, and my 4.1 ghz oc'd r5 1600 stayed cooler than with a thermaltake water 3.0 performer c. But I'll test some voltages, and maybe will try to get to 4.3 on my 3600, but I highly doubt temps will be good.