I try to run a 9900k on my Gigabyte z170x-ud5 th. Unfortunately end up with quite serious stability problems. The computer boots with stock settings, but then shuts down under heavy load. I have already tried two different bios. One was modded with CoffeeTime and the other with revlaays tool. Both times the same result. It also seems to me that Turbo Boost doesn’t work, the CPU permanently clocks with 5ghz on all cores. Powertarget settings in the bios are apparently also ignored. I made the pinmod using the picture in the attachment, red and blue are isolated with capton tape. What could be the error here? Does anyone have a similar problem?
I would suggest setting PL3 to ~50W to limit Package TDP and see if you can stabilise it - PL1/PL2 may be too slow or not set aggressive enough so your VRM is pushed to the limit = shut down.
You might also need to limit IccMax to ~125A to stop excessive power draw (especially on AVX loads).
[Edit] Your motherboard only has 4 Phases for the CPU with 4 x Vishay SiRA12 MOSFETS on the low side which are rated for 20 Amps each @ 70°C for a combined total of 80 Amps, this is going to limit your max all core frequency substantially.
Depending on the application you may be able to run 5GHz all core on really light loads (like hash checks, browsing etc) but on any heavy loads (gaming, benchmarks, video rendering etc) you may only be able to run at ~3 GHz all core depending on how good your VRM cooling is.
Hi chinobino, thanks for your answer. I have found a stable configuration with 3.6ghz at 1 Volt. With turbo disabled the system was stable even at prime95 with avx. The VRMs are water cooled, that was also the reason why I chose this upgrade route. Is there a way to push the limit somehow? Maybe bridge some shunt resistors or modify the bios to prevent the shutdowns?
I’m going to play a little bit with the power and amp limits tonight. I had the feeling that the settings have no influence but I’ll report again.
By the way, I saw in your thread you recommend to isolate the yellow pads on Gigabyte as well. Is that mandatory?
This is good progress, I could also run 3.6 GHz all core at around 0.960 V on a Gigabyte Z170 with the same VRM.
I would think using a shunt resistor would be dangerous given Intel’s spec for LLC and the purpose designed voltage fluctuations (even when vcore is set to a ‘fixed’ voltage on override). If you cleared CMOS and the board went back to adaptive voltage I think that would be the end of your VRM, so you’d need to mod the BIOS to be in override by default and I don’t know what the default override voltage is (probably 1.20 V). I don’t know why you’d want to stop it shutting down - it protects your VRM from permanent damage. You need to monitor your power/current limits and find out why it is shutting down.
They should still work, have you installed Intel’s XTU to see if the BIOS changes are being applied and also to check which limits are being exceeded? You can also use HWInfo64 or Throttlestop if you don’t like XTU. You need to be checking for any power limit throttling, current limit throttling and motherboard VR Thermal Throttling.
No, not mandatory just recommended. When comparing Skylake pin diagram to Coffee Lake those pins showed changes. These pins were also tested in some Z170/Z270 motherboards and not all of the Vendors left those pins unconnected and some were found to have power connected to them. If you’re worried then check your CPU pads for scoring and if you see brown marks check the motherboard pins, they may be missing xD.
Hey, thanks so much for taking the time to help me.
After messing around with PL3 and IccMax I came to the following conclusion.
PL3 doesn’t seem to have a big influence on the stability of my system for whatever reason. So I set the value back to auto for now. I will check it again with the software you recommended as soon as I have some time.
IccMax on the other hand works very well. I started with a value of 60. The system never shuts down. So I gradually increased the value until I reached a maximum of 135amp.
With this setting I reach in prime95 with 22-85 K FTT (Threads / Core Clock) 16 / 3.9 ghz 4 / 4.2 ghz 2 / 4.7 ghz sometimes even short 5ghz
Without AVX and FMA3 I reach 16 / 4.5 ghz 4 / 4.7 ghz 2 / 5 ghz
@davarga Your results look very good - much better than mine, as expected as you are using water cooling for the VRM (I was using the stock heatsink).
PL3 has to be set very low to have an effect.
I found that setting 67 W limited all core frequency to ~4.8 GHz (AVX load), so to limit it to 4.7 GHz all core I needed to set PL3 to less than 67000 in the BIOS.
PL3 of 50000 in BIOS = 50 W limit, still allowed 5.0 GHz on light loads.
I also found that there seemed to be ‘steps’ or increments for PL3 where the power draw was not changed until a certain threshold had been reached, IccMax may be a bit more granular and allow better fine tuning - either way just go with whatever works best for you :).
[Edit] You might want to use HWinfo64 to monitor your VRM temperature (on the sensors page it’s labelled ‘VRM’) and not let it go above ~90°C continuously.
VRM temperature settles somewhere around 65°C under full load. May reach 75°C if I start furmark and the water gets very warm. But to be honest I’m more worried about the CPU temperature in that case. It has moved towards 95°C. Which is 55°C above the water temperature. Maybe I should delid the cpu and replace the solder with liquid metal. On the other hand, I can’t imagine a real world scenario that adds that much heat into the loop.
Bad news. Yesterday, I took apart the computer to delid the cpu. So I drained water out of the system and so on, only to find out that the CPU doesn’t fit into my Delid tool. After I put it all back together, it won’t start. Not even with the 6600k. I tried everything, reinstalled the ram, etc. Probably I bent the board too much while I mounted the water cooler.
Sometimes the board shows the error code 01, then restarts and shows no codes at all. I think the board is toast. Bad timing to get a new one as the new intel platform has just been launched and prices for z390 will drop in next weeks.
Sorry to bring this thread back to the dead, but I also have a Z170X-UD5 TH with a 9900k I just got so I figured I’d post here. I was able to modify the F22g BIOS and flash it with no issue. I think I modified it right, and it did flash successfully since the stock version I was running before was F22d and it now shows as F22g. I was able to tape over the pins as well. I didn’t do the yellow, just the red and blue ones. But when turning the board on after installing the 9900k, when I push the power button it clicks on and then clicks right back off in less than a second. It clicks off before any codes come up on the 2 digit display. I pulled it out and reseated it, same thing. If I try to turn it on with an empty socket though, it at least stays on for a few seconds and a bunch of codes flash across the display before it resets. I reinstalled the 6700k and the board boots just fine and is stable when stress testing.
Any idea what it could be chinobino? To me it seems like it’s a physical issue, rather than a BIOS issue. I thought maybe the tape job I did was bad and the VRMs were shorted out and shutting off to not burn up but I checked it again, all the pads are covered and there’s no markings or damage on the bottom of the CPU or melted pins.
Did you check the ME FW version after flashing the modified BIOS with the 6700K installed to make sure it was successfully downgraded to 11.6.x or 11.7.1?
The 9900K has two CPUID’s, 906EC (S-spec SRELS) and 906ED (S-spec SRG19), what is the S-spec of your 9900K and which microcodes did you add to the BIOS?
What did you use to paste over the CPU pads? Can you show a photo?
If the 6700K still works fine under load then I think your VRM is ok.
@chinobino Thanks for the reply. I used CoffeeTime .92. I just loaded up the modified BIOS I made back into it to take a look. I think I skipped over the Intel ME in the first go. In the ME box it says 11.8.50.3425 CON and Enabled. My options next to ‘Internal’ are 11.0.0.1168 Consumer and 11.7.0.1229 Consumer plus 11.7.0.3307 Corporate and 11.8.77.3664 Corporate . Should I replace 11.8 with one of those or load a custom one? Is that all that needs to be done with the ME or can I check it once I flash it?
It’s an SRG19 9900k, brand new. I thought I had applied the 906ED microcode since it’s the last step I did before clicking ‘Save image’ but now that I’m looking at the BIOS file loaded into CoffeeTime again i don’t think I clicked ‘Replace’ as it still shows the 3 default microcodes. 906E9 506E3 and 506E8.
So as of right now I just applied the 11.7.0.1229 ME and the 906ED and 906EA(I may be able to borrow an 8700k to test if I can’t get the 9900k going) microcode and saved the image again. I closed CoffeeTime and then opened it and loaded the new BIOS I made and it looks like everything is applied now. Do you think I should be good to go as it is?
I used 3M vinyl electrical tape. The pins are all for sure covered and the tape wasn’t pierced. It does seem soft though and I don’t want to use it long term. It seems like once a bunch of heat is applied it will probably become too soft and the socket puts a ton of pressure on it. I have some Kapton tape ordered which I will use that I’ll have at the end of this week.
@chinobino Ok, new development. I flashed the new BIOS I made with the 11.7.0.1229 ME and 906ED microcode and it posted. I heard what sounded like coil whine right away which was worrying. I got into the BIOS and the CPU was pegged at 5GHz. It looks like the same issue that @davarga had. The CPU heatsink got warm very fast. I cleared CMOS and booted again. Same coil whine and 5Ghz in the BIOS. I changed the CPU clock ratio to 8 to see if it would work and it did. No coil whine. Booted into Windows at .8Ghz. Restarted and changed it to 20, ran Cinebench to see if it was stable. It was, so I went to 30, 36, 40, and now 45.
Here are some stats according to HWiNFO64 v6.34-4300, in a closed case with sloppy reused thermal paste since I’ve been swapping CPUs around. I will redo the paste once I have the Kapton tape. 8 core/16 threads at 3.6Ghz during Cinebench used 56.5 watts CPU package power at .90V, maximum VRM temp of 47C, maximum CPU temp of 49C 8 core/16 threads at 4.0Ghz during Cinebench used 72.5 watts CPU package power, maximum VRM temp of 53C, maximum CPU temp of 54C 8 core/16 threads at 4.5Ghz during Cinebench used 101 watts CPU package power at 1.07V, maximum VRM temp of 64C, maximum CPU temp of 66C
101 watts seems very reasonable to me with the VRM on this board. Let me know what you think I should max out at. I think I saw on the Gigabyte board you had was a 4 phase VRM with single mosfets, I think this one is also a 4 phase but has doubled-up mosfets judging by the board pictures at Tweak Town, so it should do a bit better I think. I didn’t get the 9900k trying to hit 5.0Ghz on all 8 cores/16 threads, I knew that wouldn’t happen, and also the power consumption and heat when it’s doing that is stupid. It looks the 9900k will pull 170watts at full load 5.0Ghz all-core judging from reviews.
Time to start tinkering with setting some more. Is the only way to get 1-4 core turbos at 5Ghz by messing with the power limits while keeping my total package power around 100W? The behavior of the board/CPU at default settings worries me with how it was pegged at 5Ghz right away and creating a bunch of heat while it was in the BIOS doing nothing. Maybe I’m just better off setting it to 4.5Ghz as the limit like I have it. I also noticed a weird behavior at all frequencies I checked. After booting into Windows, the CPU speed would be at it’s max frequency and it’s peak voltage and wouldn’t downclock and downvolt like it normally would at idle, even after sitting for 4+ minutes. But after doing a Cinebench run, it would downclock and downvolt like you would normally expect. Any idea on how to resolve this?
The interesting thing is if I turn the CPU clock ratio back to AUTO and change the boost ratios to 5ghz for 2 cores and 4ghz for 8 cores it ends up worse than just setting the clock ratio to 45 and running all 8 at 4.5ghz. Single core performance in Cinebench is higher, but 8 cores at 4.0ghz leaving it on auto and changing the boost numbers causes all the cores at 4.0ghz to run at 1.1V and CPU package power ends up at 108 watts.
@4x4Mimo what are your actual voltages / vids under load at the various frequencies? Have you manually set IA AC/DC loadline? Not sure if it’s required, but I’d imagine it doesn’t know the proper values for 6c/8c cpus. Not sure what the units are for gigabyte bioses, but stock for 8c is 160mohm. So if you’re on auto voltages and ia loadlines, it may be pushing more voltage than needed.
Setting power limits is good if you want to be able to use higher clocks at lighter loads while keeping vrms cool, or really just in general if power consumption is a consideration. Stock for the 9900ks is 127w/159w@28s. If that’s too high for your tastes, stock otherwise is 95w/119w@28s. Of course, stability testing becomes more difficult
Glad to see you got it working and you are on the right track in regards to protecting your VRM.
You should use 11.7.0.1229 consumer ideally - the others are for running Xeons.
I can see in your later posts you got it working by using ME FW 11.7.0.1229 and adding 906ED microcode
I used Kapton tape and it is perfectly suited for high temperatures and is a solid insulator so I highly recommend it!
Excellent, this is exactly what I would do, test and slowly increment and re-test
Very interesting, good to see you’re monitoring the VRM temps, looking good.
Fair point, you may be able to clock higher if the cooling is sufficient.
Are you using Speedstep or Speedshift? Speedshift can change frequency much more dynamically and I have noticed lower temperatures (both idle and load) so I would suggest to enable it if you haven’t already.
Yep, Intel’s turbo settings will require higher Vcore to ensure that 5GHz is stable and that means more heat when running all core speeds as it will use the same max Vcore. I also found that setting turbo per core to the same frequency allowed better control of temperature and lower Vcore was required to maintain the all core frequency under load.
4.0GHz all core will still give you very good performance even though the 9900K is being underclocked.
As you say, a lot of testing to be done, if you have a good 9900K you may be able to undervolt (from stock VID) to squeeze out a bit more performance.
