New AGESA 1.0.0.6

Best what i can do, stable. Soc 1.15/DDR 1.43. Taichi 2.4 + Galax 3600 (b-die).


Edited by Lisanderus - 12 Jun 2017 at 2:02pm

wardog wrote: noneis wrote: ... As I'm aware, there's no need to alter P1 and P2. Only P0. P1 and P2 are very technical and can create instability if not done properly/correctly. Plus I've read they(P1, P2) can hinder other settings, only aggravating stability. Those/These are my understandings. Just sayin' .

Ryzen is voltage limited at around 0.575-0.6V which rises with temperature. It is also rather hard to test, because instability usually occurs during transition from P0 state to lower one. There are too reasons for this:
1.) At higher temperatures Ryzen needs slightly higher voltage for given clock. So Ryzen stable voltage for lower P-State is highest right after transition from P0 State (or Boost)
2.) During transition to lower P-state voltage drops below set value before vrms react.

At low clocks voltage curve is kind of flat all the way to ~2100. From ~2100 to ~3300 Ryzen needs ~25mV for 100Mhz. Necessary voltage above 3.3 Ghz is increasing much faster. Stilt analyzed Ryzen voltage curve from 2 to 4 GHz. https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/
For example my chip is stable at these clocks and voltages:
1.2Ghz @ 0.625 V @ 80C
1.55 @ 0.650 V @ 80C
2.1 @ 0.725 V @ 80C
2.7 @ 0.875V @ 80C
2.8 @ 0.900 V @ 80C
3.7 @ 1.1875 V @ 80C

2.1 GHz is sweet spot where "flat" voltage curve starts to linearly increase. This clock yields 35% higher performance compared to stock for only 20% lower power efficiency. (considering undervolted stock P2 voltage from ~0.825 V to 0.650 V). CPU doesn't have to use less power efficient P1 state as often providing higher peak performance under low load. It's optimal clock for P2 state because of this and Ryzens fast core parking. This is especially beneficial during highly multithreaded low utilization workloads such as video playback and/or when using power saving plans in windows.
3.3 GHz is probably optimal clock for P1 State because with higher clocks power efficiency rapidly deteriorates.

Edited by noneis - 11 Jun 2017 at 11:53pm

noneis wrote: datonyb wrote: in p states section i set a 3.9 cpu and a 1.306 voltage on first one and leave all others auto so pstates is   9c/9/27 i turn off cool n quiet, c6 etc then on advanced tab i set cpu fixed volts but leave voltage on auto there and llc level 2 when using monitoring software in windows is is very ridgidly fixed at 1.3 volts You are overclocking only P-State 0. That works for me too. I was talking about P-State 1 and P-state 2. That works in pervious BIOS but not in new BIOS with AGESA 1.0.0.6.

As I'm aware, there's no need to alter P1 and P2. Only P0. P1 and P2 are very technical and can create instability if not done properly/correctly.

Plus I've read they(P1, P2) can hinder other settings, only aggravating stability.


Those/These are my understandings. Just sayin' .

noneis wrote: datonyb wrote: noneis wrote: Does new AGESA 1.0.0.6 disables P-State 1 & 2 overcloking (with Offset mode voltage) or is it new bug? It doesn't work with X370 Gaming K4 with 2.5.0 BIOS. i use p states and set volts there i then use fixed volts on advanced tab  and use llc level 2 it seems to stay very firmly set at what i want I have just tried setting CPU Vcore Voltage to Fixed Mode and tried both Auto and specific stable voltage. (1.1875). Both gave me error code: "OC" as reported by Dr. Debug. Had to reset CMOS with jumper. Same thing happend before with Offset mode both Auto & specific offset. With 2.4.0 BIOS in offset mode a could overclock not just P-State 0 but also P-State 1 & 2. In 2.5.0 BIOS I can overclock only P-State 0. Did you set different (higher) clocks for P-States 1 & 2 or just changed volts? Changing volts works for me but changing clocks don't.

in p states section i set a 3.9 cpu and a 1.306 voltage on first one and leave all others auto
so pstates is   9c/9/27
i turn off cool n quiet, c6 etc
then on advanced tab i set cpu fixed volts but leave voltage on auto there and llc level 2

when using monitoring software in windows is is very ridgidly fixed at 1.3 volts

Your tRC is set at teoretical limit tRC = tRAS+tRP, try to set it higher. Stilt (B-Die) is using 54 for tRP=14 tRAS=28 that's 12T/29% above teoretical limit. My hynix kit at 2933 data rate requires 9T/20% above limit. So with Stilts ratio and your tRP,tRAS try something between 74 and 80+