@
DrMrLordX
Taking your advice on looking at the current side of things, I've logged a lot of data in the last few days. Hopefully it doesn't crowd this thread too much, if it does, maybe I could make a new one. I've been reading a lot about Ryzen in the last 12 months, but only finally got around to building my system in the last week or two. It's been fun just trying to figure out how it all works. And it's interesting to see how these Zen chips boost more like GPU's these days. But at the same time, I miss some aspects of the old Intel manual OC.. with the different frequencies for how many cores are loaded. And a bit of headroom to play with. Not sure how Intel does it these days, as my last Intel system was 4790K, which I'm typing this message on at the moment. My main complaint about stock Ryzen boost is the midrange temps and wattage. I like the idle and the full load, but don't like how jumpy and hot it gets in smaller tasks while idling. I also prefer to prioritize multi core over single core.
I found this picture below in a recent motherboard review from this site, which gives a sort of baseline for 3700X clocks and voltages.. the AGESA is pretty recent as well. I have to say I haven't tried the 'actual' manual OC setting so far. I've just been using per CCX, but using 42x 42x if I want 4200mhz for example. I'm not sure if it makes a difference, I just find it simple and easy to work with. The board is ASUS Crosshair Hero VI.
It seems my chip is better than this one at manual OC, but who knows. There's so many factors, maybe if I used that board with that BIOS, my numbers would be worse.
The wattage is measured from the wall for the entire system. And those voltage numbers were going for the lowest voltage possible to complete many CB R20 runs. To make it stable for linpack AVX or AVX2, it needs another .1 volts or more. For example IntelBurnTestV2 at very high for 10 runs. The easiest way is to make it stable for 4300mhz in CB R20, and then drop the frequency back to 4250 or 4200. For example, it can be stable at 4300mhz at pretty much anything at 1.286V under load (using 4350 voltages from above).
I've discovered two stress programs I didn't know about in the last few days. OCCT V7.2.0 and Realbench (great for running GPU and CPU at 100% at the same time, and it uses X264 which is great). The great thing about OCCT is HWINFO is a bit buggy with the sensor chip on my board (freezes a lot) but OCCT has the same data and it never freezes, so I can finally monitor all the Ryzen specific stuff while it is at 100% load. The great thing about OCCT is that it has all the Prime type stuff and the IntelBurnTest type stuff, and also Memtest type stuff all in one program and it monitors at the same time. And can automatically stop the test as soon as it detects an error.
Above 3300mhz, it gets a bit tough and the voltages and temps skyrocket. CB R20 can be stable but other stuff starts to get too hard to cool down. I didn't try 4400mhz, it felt too far out of reach without putting the voltages really high. The sweet spot is around 4250 - 4350mhz depending on how much wattage/temps you prefer. 4200mhz would be easy to run with low voltages and temps, but the SC score suffers quite a lot.
I recorded some current (A) numbers back at stock in OCCT and compared them to manual OC numbers. The stock numbers for my 65W TDP chip are..
PPT - 88W
TDC - 60A
EDC - 90A
Through all of my manual OC'ing.. the EDC stayed locked at 90A, not sure if there's a way to unlock that or if it's hardwired in. Or if that would be a good or bad thing. I mean, outside of the firmware isn't this just a poorly binned 3800X? The 3800X can do 140A EDC. I'm unclear about this. The EDC is supposed to be a VRM rating, but if I put in a 3800X, magically my boards VRM can handle an extra 50A?
During Manual OC, PPT went from 88W to 108W at the max, and current would sometimes go to 68A or 70A. So that's 10A more than stock. Is this unsafe or bad? That was mainly in linpack AVX. In regular tasks or even CB R20.. or X264 encoding, it stays around 58A to 62A. In Prime.. I don't even run Prime (small FFT). Nothing I use in regular real life usage comes close to it, AVX is about the most I ever do. Any run like that with manual OC isn't worth it. It feels like over stressing the chip for no reason.
The stock current numbers in OCCT 7.2.0 were -
Linpack 2012 (AVX) -
CPU Current - 59A
EDC - 82A
PPT - 88W
Vcore under load - 1.22V
Multiplier - 39.3X
Temp - 65C
Linpack 2019 (AVX2) -
CPU Current - 56A
EDC - 89A
PPT - 89W
Vcore under load - 1.29V
Multiplier - 40X
Temp - 69C
OCCT Small data set (AVX2) - (Prime 95 Small FFT type test)
CPU Current - 62A <--- (exceeding stock spec)
EDC - 89A
PPT - 89W
Vcore under load - 1.13V <--- wow much different than 1.325V
Multiplier - 38.25X <--- behaving almost like a built in offset
Temp - 67C
Manual OC - 3300mhz at 1.286V
CPU Current - 64A <--- (is this too high?)
EDC - 90A
PPT - 106W
Vcore under load - 1.264V
Temp - 72C
The setting above can be slightly unstable in RealBench after an hour or two if the temperatures are high. If I increase voltage even more.. it becomes super stable even over 80C.. and the current goes down a bit but the temperature and wattages goes up a bit more.
Manual OC - 3300mhz at 1.303V
CPU Current - 70A
EDC - 90A
PPT - 110W
Vcore under load - 1.286V
Temp - 74C
I'm not sure what is better, if if the current goes down or if the voltages goes down? The temps are between 70-75C. It only goes towards 80C in a stress test like RealBench after 30 mins when the GPU is at 100% load, at the same time the CPU is at 100% load. I've adjusted case fans to help reduce that and stop the motherboard/VRM from soaking up too much heat. The VRM max temp so far has been 52C.
I guess this would be safe? As long as I don't run any of those Prime small FFT type loads? Cooler is Noctua D15S with two fans and in CPU heavy stuff it is always below 75C. In CB R20 it's 69C. I was hoping to use voltage under load of 1.22-1.29 because those are the stock Vcore for Linpack AVX and AVX2 but yeah. The temps are ok and the voltage isn't exceeding 1.3V? The main thing is that under some loads, the current is 10A higher than usual.