Maybe the cores aren't thermal throttling then, perhaps the 1.44V they're getting isn't enough and the CPU is seeing errors and throttling down.
RGone, no, I leave all CPU-related adjustments at stock settings in the BIOS and use AMD OverDrive to raise the Turbo speed. I do adjust the NB/GFX voltage in the BIOS, and the GPU clock, because adjusting voltages in AMD OverDrive seems to have no effect.
As for the second part, I see what you're saying. Yes, raising Turbo to only 4.4-4.5 GHz seems to work just fine, without throttling much under load. 4.6 GHz is pretty much the limit with as much voltage as it's getting (1.44). So, like I said the throttling is probably not temp related (45C is as hot as it gets measured by OHM, and I don't think the sensor is 45 degrees off), and with a little more voltage might hang around those speeds more consistently. I mean, I don't know what to say, my temps have never really gotten out of hand. 50C is the hottest the CPU sensor has ever shown, and the room was 26-27C at that time.
Check motherboard VRM temps pretty please
The CPU can't sense errors and throttle...the CPU simply does a calculation and spits it out, if it is erroneous it will never know unless you check it against a reference like Prime95 or LinX does. (When you get an error, it's because you were supposed to get a certain "answer" to the "problem" and didn't get it)
Also, Turbo will not work with Prime95. LinX and other stress testers (except maybe AIDA which is a super easy test) probably won't work either.
Turbo raises TDP too, (I don't mean the MAX TDP rating, I mean heat output for a given usage scenario(!)) ...and because of the way Piledriver's APM/Turbo leans on TDP, it is possible that you are just above that threshold for current draw. Turbo is used to increase overall performance without increasing current draw, so that your single thread programs and low-threaded programs (4 thread, example) get the most performance possible that the architecture will allow while also giving the whole 8 threads and better overall performance to programs that can use it.
Turbo is very aggressive on Piledriver, most normal 8 thread workloads also benefit from Turbo too - AT STOCK VOLTAGE/SPEED. ...because there is wiggle room in the TDP/current draw at whatever time.
If APM is disabled, you're out of luck. If it's not disabled, then try it, because that's supposed to be what I'll call here the "power and CPU usage monitor".
If you're overclocking like this, then you shouldn't care about that. You should disable Turbo because of two things:
The turbo states have a much higher voltage than stock VID, configured as an offset of that VID. Lets look at a scenario:
Stock voltage 1.3v
Turbo VID 1.425v
Manual voltage: 1.45v
Theoretical turbo VID = 1.575v
This CAN happen. If it doesn't, "phew".
because adjusting voltages in AMD OverDrive seems to have no effect. << Good Ole AOD.
Using CHV and FX-8120 can set APM to OFF and the cpu does not throttle by power states. Turbocore also off.
I know AMD had no idea till the end that temps would be so high. They barely shoe-horned the BD or PD into their TDP rating. So they will give you a boost for just a little but then it will reduce speed it seems. They can say the setup will give you a boost if X and Y are right but they don't say it will not last for long. Hehehe.
BB2 after "redduc900" had us go into the past said some 'guy' at ExSys was testing Vishera and Trinity and gaming loads would allow the near full upclock, but P95 would drop the speed almost immediately. I am not surprised. What is said about that is why I lock APM OFF and turn Turbocore OFF and then handle the heat that is generated running 8 cores on and loaded with no throttle back. After BB2 spoke I did a little looking and found that Turbocore on PD even works differently than on BD. So a lot has changed. For the good? Not at all sure. By the way as I understand it now, Turbocore can respond to calculated TDP, to what it sees as 'the' load and also to current draw. So if that is accurate, for us to really know why or exactly when the cores swap P-states is going to be hard to do. I doubt software monitoring is fast enough to catch all the swapping. Maybe and maybe not.
When the load is removed from my FX-8120 the speed drops from 4.5Ghz to just about 1450Mhz and the heat drops 'dramatically'. All of that is accomplished though with bios settings and Windows 'power manager'. I can only guess that an APU with IGP might have problems run in the manner I run my FX-8120, because the heat is ginormous. Heck you can feel the heat coming off the VRM sinks on my CHV from supplying the power to 8 cores with no throttling.
AMD does not have the resources to rewrite AMD OD every week and add support for next weeks VRM controllers and every sensor from every manufacturer too. If AMD were to force a standard on manufacturers, ie. use last year's model of everything that puts them in a box and can't happen. If AMD were to have a crack team on the software development and be in constant contact with all manufacturers about what they all use in specific for every little thing on every board, there wouldn't be an issue.
That just can't happen. First off that costs a lot more money and us enthusiasts make up like 0.01% of the overall market.
If you think AMD had no idea till the end that temps would be so high, then you're strongly mistaken. Who do you think designed the CPU? Working stepping B0 BD Engineering samples have been available since the availability of the 32nm process...the first months of December 2010 AMD already had functioning BD CPUs and had an instant idea of how things would be. They knew that month that temperatures would be high and had an idea of power consumption. They have to shoehorn ANY CPU into a given TDP. First off, TDP means "Thermal Dissipation Power", while it's often correlated with power usage, it's really about how many watts (joules per second) must be dissipated by a heatsink, not the electricity kind. Without providing a giant heatsink so they can release with a higher TDP (and making themselves look very bad in them needing it and vs intel TDP), they MUST fit that TDP.
There is a reason why that boost is "a little", as explained above the quote.
Not just some guy, I told you all who it was...the same guy ("The Stilt") that wrote
TCI K^2 (Trinity Control Interface).
Here are the posts I am referencing, and while talking specifically about Trinity, it should apply to all Piledriver based platforms. I'll ask him to test it to make sure as I don't have access to CPUs right now. (Pb1 is the "turbo" power state)
http://www.xtremesystems.org/forums....3GHz-on-LN2&p=5140475&viewfull=1#post5140475
http://www.xtremesystems.org/forums....3GHz-on-LN2&p=5140494&viewfull=1#post5140494
Again, Turbo Core does not fundamentally work differently than BD, nor does it really than Intel Core i7 2x00 and 3xx0 CPUs. AMD's just works in 200 MHz steps (because it has to) and intel's works in 100 MHz because they have a lower BCLK, like FM2 sockets. FM2 works a little more like those platforms as far as the interconnect layout/design but is not optimal for enthusiast platforms, it is designed to be lower cost for their mainstream APUs.
The only difference between BD and PD Turbo is that APM is tuned more aggressively towards TDP, because they have a bigger window to play with now due to increased power and thermal efficiency. (Thermals don't matter in 50% workloads, but do matter in 100% workloads, and if power target is 160w, and 8 thread workload uses 130w, then turbo will engage, when before, BD was using the 160w up already).
In reality, Turbo Core was always based off of TDP, except when you load up all threads what do you think happens to the power consumption that is monitored to calculate whether to engage turbo or not?
So, not much has changed, PD is just much more efficient.
Software can't catch the changes in real time? Blasphemy
...there is a program called
TMonitor by the folks over at
CPUID that spits out the frequency data as fast as your eye can read it.
Windows' power consumption stuff works in correlation to the stuff you set in BIOS. Both use the same hardware registers (they talk to the same place) and they both accomplish the same thing.
If you select high performance in Windows even with C&Q and all that Jazz enabled, I believe it won't engage (not 100% sure) because Windows tells your hardware not to use it. If I'm wrong, then you can still go into the advanced power options and configure it yourself, and it is guaranteed to work.
If you force disable it in BIOS, you don't leave the option available, so it really doesn't matter what option you choose in Windows, the only things it may possibly effect would be Core Parking and "Turn off display" timer, sleep state timer, etc.
FX-8120 and 8150 are BD based to very leaky, high power consumption CPUs...PD is not quite as bad and APUs are a whole different animal with a different footprint and different VRM/PWM...apples to oranges.
