Got it stable. I had to disable the c-state and increase the voltage +200mV. Runs all Benchmarks fine... Without c-state the clock remains at 4.5Ghz all the time.
A lower cache ratio is probably the solution. Unfortunately Can't find this option in my BIOS.

@davidm71 updated microcode to DE in bios. Unfortunately adaptive voltage still isn't working. FWIW, I realized you've got a 906ec p0, vs I'm 906ed r0.

That said, it does seem to behave somewhat differently. My original LLC4 / 0.3 IA AC/DC loadline setting is no longer stable at all.

Testing auto LLC and 1.00 IA AC/DC loadline right now with full c-states. Deeper c-states would crash previously. Too early to tell, but so far so good (although I did bump cpu standby voltage up to 1.05v if that's making a difference)

4.7ghz: ~1.35vid 1.25v non-avx, ~1.41vid 1.3v avx

edit: nope. bumped adaptive vcore from 1.3 to 1.325 in case it is doing something + cpu standby from 1.05 to 1.15.
edit2: no bsod yet but got a status access violation in chrome. definitely not stable still :\
edit3: whea uncorrectable. bumped adaptive vcore again. looks like it is working now after all! It's pretty unfortunate pumping more and more voltage to handle these partial load / transition voltages. If driving a steady full load, I need way less voltage than I'm pushing now :\
edit4: 1.25v +0.1v offset. Really don't want to resort to fixed voltage, but might have to
edit5: adaptive voltage is indeed not working.

edit6: Ok, played with it a bit more. Really should be using a secondary OS. Had a big scare but seem to have recovered somehow..

The big issue is c-states, at least >c1/c1e. I did some testing with manual/fixed voltage (it works), and with all c-states enabled, need 1.325v for 4.7ghz. While on manual voltage, I had IA AC/DC loadline set to 0.01 so I could see the best scenario vid table. Looks like this chip asks for ~1.24-1.265v at 4.7ghz depending on avx.

Having c-states set to auto (c1/c1e only), I can go down to 1.25v if not lower. So I reverted back to my old preferred setting with past cpus (LLC4, IA AC/DC loadline 0.01). As expected, it's stable.

Again, adaptive voltage isn't working here (offset works). And >c1e requires massive voltage to stabilize. That'll bug the hell out of me, but at least reported idle consumption isn't noticeably different.

edit7: enabled c3. still stable

Tried yesterday the newest microcode for P0 and R0. Still problems with c-state. I have to disable C6 with the P0 version. In the R0 Version I can enable all c states (don't think it works at all)
Fixed and Auto voltage does not work stable at all. Have to use offset of 0.3V. In hwinfo I can see values between 0.9 and 1.2v.

@Monteaup you can use throttlestop or hwinfo64 to see c-state residencies.

CoffeeTime 0.9
Download
(Mirror)

Hello.
I'm back from vacation with a new CoffeeTime. =)
It's 0.9 version, which in fact is Advanced mode, without any automods (temporarily). So it requires some base knowledge about coffeemod.
At the moment CT 0.9 can help you to mod bioses for:
- 100/200 series to use SKL Xeon, KBL Xeon, CFL, CFL Xeon, KBL 1440, CFL 1440.
- Z370/B365/H310C to use SKL, KBL, SKL Xeon, KBL Xeon, CFL Xeon, KBL 1440, CFL 1440.
Check out readme for more about features CT 0.9.
Antivirus make warning about "avsr.exe" file (as for most compiled python scrips) - add it to exceptions in this case.

Come on, add some CML to that list :-)

Zitat von ziddey im Beitrag #1232

---

@davidm71 updated microcode to DE in bios. Unfortunately adaptive voltage still isn't working. FWIW, I realized you've got a 906ec p0, vs I'm 906ed r0.

That said, it does seem to behave somewhat differently. My original LLC4 / 0.3 IA AC/DC loadline setting is no longer stable at all.

Testing auto LLC and 1.00 IA AC/DC loadline right now with full c-states. Deeper c-states would crash previously. Too early to tell, but so far so good (although I did bump cpu standby voltage up to 1.05v if that's making a difference)

4.7ghz: ~1.35vid 1.25v non-avx, ~1.41vid 1.3v avx

edit: nope. bumped adaptive vcore from 1.3 to 1.325 in case it is doing something + cpu standby from 1.05 to 1.15.
edit2: no bsod yet but got a status access violation in chrome. definitely not stable still :\
edit3: whea uncorrectable. bumped adaptive vcore again. looks like it is working now after all! It's pretty unfortunate pumping more and more voltage to handle these partial load / transition voltages. If driving a steady full load, I need way less voltage than I'm pushing now :\
edit4: 1.25v +0.1v offset. Really don't want to resort to fixed voltage, but might have to
edit5: adaptive voltage is indeed not working.

edit6: Ok, played with it a bit more. Really should be using a secondary OS. Had a big scare but seem to have recovered somehow..

The big issue is c-states, at least >c1/c1e. I did some testing with manual/fixed voltage (it works), and with all c-states enabled, need 1.325v for 4.7ghz. While on manual voltage, I had IA AC/DC loadline set to 0.01 so I could see the best scenario vid table. Looks like this chip asks for ~1.24-1.265v at 4.7ghz depending on avx.

Having c-states set to auto (c1/c1e only), I can go down to 1.25v if not lower. So I reverted back to my old preferred setting with past cpus (LLC4, IA AC/DC loadline 0.01). As expected, it's stable.

Again, adaptive voltage isn't working here (offset works). And >c1e requires massive voltage to stabilize. That'll bug the hell out of me, but at least reported idle consumption isn't noticeably different.

edit7: enabled c3. still stable

---

Yeah, I've been watching the vids, and the following has been stable:

* 47x100
* C6/C7 disabled, All PC-states enabled (via setting "cpu default" -- seems to cap at PC2 likely because of C-state cap)
* 0.01 IA AC/DC loadline
* LLC 1 (minimum)
* +0.05v offset

This gives me offset vids 0.680-1.300. Enough volts for light/transition loads, and some "nice" droop for load.

Unfortunately, these settings are only good for an all-core-synced 47x. E.g. for 4.9ghz, way more offset is required in order to stabilize transitions, resulting in an excessive 1.36v load vcore. With fixed voltage, it's stable at 1.29v.

Good call on the ram. I did actually drop to 2800 for all this. Running 4 sticks of hynix afr 8gbit. I've been able to push this ram up to 3400mhz with a 1600ae/b350, yet never been able to exceed 3000 stably on this z170. Even at 2933/3000, if it trains wrong on post, it'll be unstable. Yet decrease to 2800 and I can drop vccio/vccsa to stock, and tighten to 14-16-16-30-1T (from 14-17-17-30-2T).

And with modded 2000-series bios, can't post above 2133 with 4 sticks at all

Sorry for my bad English. I need help. Can anyone help me mod this main bios? I'm using: colorful BATTLE AXE C.B150M-D V23. Can not find bios on the homepage so I backed up bios on the mainboard
drive. google. com/file/d/1u24MLtb...iew?usp=sharing
when I use coffee_lake_mod_tool it gives an error:
imgur. com/1wPwgEo
also if using CoffeeTime it gives an error: "patch block size not known"

@abuken What exactly CPU you want to use?

Zitat von svarmod im Beitrag #1240

---

@abuken What exactly CPU you want to use?

---

@abuken Try this, flash with SPI programmer.
https://dropmefiles.com/lfIro

Zitat von svarmod im Beitrag #1242

---

@abuken Try this, flash with SPI programmer.
https://dropmefiles.com/lfIro

---

@svarmod - Good to see you back, and thank you from everyone for the update!

I was reading the change log, and have two questions when you have some time to answer
11.7.0.3307 | COR | Cut << What is "Cut" mean
What is NVME Fix used to do? Before, I thought this meant "Do NVME mod/insert", but now I see it's about removing something

@Lost_N_BIOS

Cut - means literally cut.)
Original ME_COR (about 7Mb at default) was cleaned with ME_cleaner and size was additionally reduced after - to fit in 2Mb space into most bioses (similar with this).

NVMe fix - now just deletes 1 module to solve NVMe and 8-threads+ compatibility problem. In v 0.85a it completely replaces internal NVMe driver to Express version (-3 / +1 GUIDS)