Soft clock versus BIOS clock

[PaulCa]

Here's my plan, I wanted to run it by you guys. The idea is to provide a minimal overclock in BIOS and then use MSI Afterburner and AMD Ryzen Master to apply profiles for gaming when needed. The logic behind it is removing unnecessary heat and noise and remove long term OC stress (on cooling and on me).

System: Ryzen 2700X, 16Gb 3600Mhz 8pack, Crosshair VII, Pilat 1070Ti shroud cooler.
Stock speeds (from memory): CPU 3.7Ghz, memory: 2133Mhz, GPU 1900Mhz/4000Mhz

In the BIOS perform a small overclock to something conservative like CPU: 3.9Ghz Memory, 3000Mhz. Out of interest 3600Mhz D.O.C.P profile is unstable

Then in Windows 10 I can use MSI afterburner to push the video card to it's limits when I intend to game and Ryzen Master to push the 2700X to 4.2Ghz or higher and the ram to 3400Mhz

So far I have tested the CPU up to 4.2Ghz at 1.4V as stable, memory at 3200Mhz stable, but I believe that 3400Mhz is the sweet spot. GPU I have added +200Mhz to both CPU and memory, at 250Mhz it became unstable. Needs a bit more work to see if the memory will go higher still.

At these higher speeds while the initial stress tests only produce temps in the low 70s the system does build heat over an hour or more of gaming producing mid 80s on the CPU and high 70s on the video card.

I have had a few overheats caused by AI Suite 3 being bugged and worse than not controlling the fan speed it takes control of them and leaves them at minimum. Further when bugged clicking the "Full Speed" button to get the heat out of the system has a 50% chance of a hard shutdown occurring. So I need to find an alternative to that.

One consideration not mentioned is the system will dual boot to Gentoo Linux, so I do want a sensible BIOS profile for CPU bound compiling and workstation style loads. I also need to make sure that the default BIOS fan speed profile is sane and not borked by AI Suite 3 bug effects.

The system is currently very nearly silent under light load, I'm hoping to keep it that way, I don't so much mind if it sounds like a tornado under gaming load though.

Cooling, BTW, is a cheap Corsair H100X AIO 240mm Rad with front mounted twin fan rad and stock Fractal Designs case fans, one intake on the bottom, one exhaust behind CPU.

 

[LuckyBenski]

I've been in a similar boat: Linux dual boot, windows only for gaming really. I'm a massive tweaker so Speedfan under Windows is where it's at for me. Once you get your head around how to control it, I've found it's very reliable and versatile.

But it doesn't work under Linux. Annoying. Still, I'd recommend you set a basic fan profile in your BIOS and then have a modified one under Windows via Speedfan. Then, arguably, you could apply a different overclock during the reboot into Linux.

Also, IMO stress testing temperatures are only valid when they stabilise, whether this takes 10 minutes or 45. I think a long stabilisation period just means your case has a lot of thermal mass i.e. size, air capacity, metal mass. Or your room's thermal capacity is also coming into play if it's small/sealed?

 

[LuckyBenski]

Just noticed your high soak temperatures are while gaming, do you use Furmark/Kombustor to stress test your GPU? It's over the top but I do my final stress/noise/temperature balance tests with 100% CPU and GPU loads via Prime95 and Furmark.

 

[PaulCa]

So I set up a 4Ghz BIOS profile with the memory at 3200Mhz. This works fine.

I stress test things with Furmark and Prime95 together, when the fans are working the temps stabilse to a flat line i a minute or two. Gaming is the only actual long temp test I've done so far. Rise the tomb raider played straight for 3 hours didn't produce any worrying temps @ 4.2Ghz CPU and +200Mhz GPU clock and ram.

I can actually go to 4.3Ghz and +200/+300 on the GPU, I haven't tried 3400 DDR yet though.

My main problem at the moment is control of the case fans. They are the stock Fractal Design fans that came with the case, I have one in take, one exhaust, in addition to the CPU radiator. The trouble I have is the two case fans don't seem all that high flow, quiet yes, high flow, no. So in games with low ish CPU temps the radiator fans don't spin up much and the case fans can't handle the thermal load of the GPU. The internal case temps are well in the 40s.

I tried lots of things, including uninstalling AI Suite for simply being broken. SpeedFan only identifies my fans every once in a while, gives me no CPU temp and a random temp at -103C for the motherboard.

I set the BIOS Q-Fan control to set the case fans to 100% when the mobo temp gets above 40*C but either it's not working or the fans are flat out already.

I think I need to invest in two or three high airflow PWN case fans for that to work.

No GPU temp sensor in the BIOS, although I could install a thermistor onto the SENSOR header and use that.

 

[PaulCa]

So a further bit of reading and I might just leave the BIOS is "AUTO" for CPU clocks. The reason being that in that mode the XFR2 will alter the voltage and clock frequency based on load. It claims (from memory) up to 4.35GHz boost clocks (with proper) cooling. Reviews are reporting 4.2Ghz, but I will do some testing myself.

If I run it with a manual 4Ghz overclock, it's not really an overclock when "AUTO" will happily run it up to at least 4.2Ghz. Running it constantly at 4.2Ghz is just going to produce more heat and waste power.

A manual, once in a while, overclock to 4.3Ghz doesn't really seem worth it though.

I will try the memory at 3400Mhz and obviously I have the option to offset clock the GPU still.

 

[PaulCa]

Small update. I tested "AUTO" and the CPU clocks as low as 3.3Ghz when idle and will "spike" up to 4.25Gz, however any sustained load and it quickly retreats back to 4Ghz or even 3.9Ghz, so it's thermal profile is very conservative. This is fine for normal desktop use and probably most gaming, so I'm leaving my BIOS profile set this way. Memory is stable at D.O.C.P profile with 3400Mhz manually set.

My high end overclock GPU+CPU is stable, except if I run Prime95 in FFT at the same time as a GPU stress test or bench mark when it lasts about 20 minutes before locking up, so I expect the CPU under that brutal load with case temps of 40+*C eventually overheats. I didn't get to see the temp as Valley was in fullscreen, but I have seen the CPU as high as 88*C with FFT. Intake was 24*C, exhaust 41*C. Small FFT not exactly a realistic load though, it designed to overheat your CPU. Interestingly though the CPU survives FFT for an hour without the GPU cooking the insides of the case. Yet as we know 90% of games will be barely loading the CPU and even if they were the "Blended load" from Prime95 runs the CPU 5*C or more cooler.

On fans, I have given up on the Fractal case fans. They are not PWM and I'm not sure my mobo is running them flat out. So I have invested in a 10K NTC temperature probe and 3 140mm Arctic PWM fans. The idea is to put two intakes in the floor, one exhaust in the rear and the temp probe on the case side near the video card shroud exhaust. The intent, subject to tests is:

Under 30*C case temps intake and exhaust at minimum RPM.
30-40*C case temps (remember it's in the exhaust of the GPU) 50% RPM.
Over 50*C 100% RPM.
So basically as soon as the GPU spools up it's fans the case fans should come up to full RPM to dump the heat out.

If this isn't enough I have the option to put 2 (or 3!) 140mm exhausts in the case roof.

The CPU I have the profile:
< 50*C minimum rad fan speed.
> 50*C 50% fan speed.
> 70*C 100% fan speed.
With a 5 second smooth change.

 

[LuckyBenski]

Personally I would set up fan curves (ramps really) so speeds don't jump and temperatures will balance against noise better.

But I'm concerned about the system locking up. If the CPU is overheating I'd be more expecting a shutdown than a lock up, unless it's voltage/overclock/stability related. Certainly at stock clocks I wouldn't expect instability even at 80-90°.

Welcome to the world of obsessive tweaking

 

[PaulCa]

Yea, but it might not be the CPU causing the lock up. With exhaust air temp of 41*C the actual mobo/ssd/ram might be much higher. I did see the M.2 temp at 55*C at one point, it's mounted directly between the CPU, 16xPCIe slot and VRMs, like that was a good idea.

I have also found that CPUs can and do lock up when overheated. They can only throttle so much. My previous FX9590 began locking up more and more frequently, it's at stock settings. Of course the air cooler was absolutely stuffed with dust. Hoover it out and it can again be nailed 100% without lock ups.

The other thing to remember is while an overclock is stable at 80*C it might not be stable at 85*C, the CPU physically expands etc.

I got the new 140mm fans last night, new temp sensor is still in the post. they add noise, not to much, but without overclocking playing ROTR for 3 hours produced exhaust temps not higher than 36*C. I will repeat with overclocking tonight.

On fan curves, I'm suffering from bleeding edge syndrome with the X470 chipset. Not even ASUS's software supports it properly. The AI Suite only connects to the chipset about 1 in 3 boots. The other 2 boots it locks the fan control, so the fans never spin up from idle. Produced temp related lock up a few times that way. SpeedFan only sees the GPU and HDs, none of the mother board fans or sensors

Besides, it dual boots so I can't rely on windows software to control the fans. The BIOS only gives me Low, Mid, High temps to fan %. However it will allow me via the extra temp sensor to monitor the video card exhaust temp which is what I want to control the case fans from, giving something like:

Radiator fans - CPU temp controlled.
Exhaust fan - 50% from idle, 100% on GPU heat up
Bottom intake fan 1 - MINIMUM% from idle, 100% on GPU heat up
Bottom intake fan 2 - OFF from idle winding up with GPU temp.

When I say "GPU temp" I can locate the temp sensor anywhere I want. On or beside the GPU heat sink is my first go to.

Also the BIOS allows multilpe temperature sources per fan (3) so I can also add PCH, VRM, Motherboard temps.

The idea is of course for the PC to next to completely silent during desktop use and only generate noise when it needs the high flow cooling.

 

[Begbie]

Side note, Speed fan doesn't work with most Ryzen motherboards.

 

[LuckyBenski]

Side note, Speed fan doesn't work with most Ryzen motherboards.

Indeed, I always forget that Speedfan often hasn't been updated for modern regulation chips/motherboards. I'm used to using it on older stuff, my newest board is a Z97.

 

[PaulCa]

Are there any other options for modern motherboards?

For fans and temps etc? I've been using MSI Afterburner for metrics, but the only fan speed it shows is the GPU I believe (I haven't checked if I can add CPU and case fans speed to the palelte of charts yet).

 

[LuckyBenski]

Are there any other options for modern motherboards?

For fans and temps etc? I've been using MSI Afterburner for metrics, but the only fan speed it shows is the GPU I believe (I haven't checked if I can add CPU and case fans speed to the palelte of charts yet).

I think Speedfan will still monitor those things, it just won't be able to control the regulation. I think it can also do logs/graphs. Give it a try.

Also HWMonitor. https://www.cpuid.com/softwares/hwmonitor-pro.html

 

[PaulCa]

Speedfan is a non-starter unfortunately. SpeedFan shows me GPU temp and both HD temps and nothing more, doesn't even see any fans.

I am currently trying Argus Monitor. It's payware (30 day free trial), but it's cheap. It also works nearly perfectly. I say nearly as it reports the core temps wrong, at least I hope they are wrong as they are idle at 75*C and pinned at 110*C loaded. However the CPU socket temperature matches the AMD Ryzen Master software, so I'm trusting that.

I have been able to achieve what I wanted for cooling and noise with only a few caveats. Here is how:

CPU rad fans are BIOS controlled, I found that the BIOS (or the fans themselves) will enforce a minimum RPM on it, even if you set it to 0% it remains spinning at minimum and as the temperature of the CPU responds rapidly to load, so does the fan. At minimum 800rpm it's virtually silent.

I have an intake at the bottom front of the case, this responds to the motherboard temperature, I don't know exactly where that sensor is but it's indicative of the general temperature inside the case. This fan will eventually go off when everything cools down but the system has to be completely idle. The first caveat is that I need a quieter fan than my Arctic F14 for this slot as it runs most of the time on the Desktop at 40-50%, even when browsing . The curve controls provided by Argus monitor don't scale or zoom, and as this responds to a very narrow range of temp, 25*C - 35*C it's tricky to tune it right.

I then have a pair of fans, one intake below the video card, one exhaust above it (on the back of the case). These respond to a temperature sensor I slipped into the GPU heatsink. On the desktop these are usually off. When the GPU heatsink gets over 50*C these begin to spool up and hit maximum RPM around 70*C on the GPU heatsink. I actually had to tune these back a good bit as they were coming on before the GPU fan was and actually cooling the card enough that it's fan never came on! The GPU fan is quieter, so I backed the case fans off a bit so the GPU fan comes on first.

Under heat stress tests with FFT and Furmark the system builds up to a right tornado of fans but exhaust temps never get above 35*C. CPU is below 80*C and GPU below 70*C. When the load is removed the CPU rad fans drop off instantly, the GPU responding fans spin down after a few minutes and eventually the general intake fan spins down to silent.

So all I need now is a quieter PWM intake fan for the nearly always on bottom airflow intake fan.