Maximus XI gene, 9900KS, episode 2 degradation??

[edhunterbg]

This is kind of next episode of my story.
First episode is dead cpu - https://rog.asus.com/forum/showthread.php?115747-Maximus-XI-Gene-qcode-CC-dead

So I received my new cpu - it is 9900KS.
Initially I had no problems. But few days later it started giving me troubles. I'll explain shortly the issue:

After installing I started to seek the lowest possible voltage for my new cpu at 5ghz without avx offset.

5.0Ghz,
XMP disabled
avx offset 0
mce - remove all limits
Voltage manual - this is what changes
LLC level 5

I am using prime95 40k fft size avx load to quickly test for stability and temperature. (For me it has proven to be the best overall test - if I can pass this, I can pass everything).

Day 1: vcore 1.345v - stable, max temp 93
Day 2: vcore 1.36v - stable, max temp 96
Day 3: vcore 1.37v and yet no stable

Of course I have not been running prime all day. If I pass 15 minutes of it I go on with the next step. I noticed the need for voltage increase when I tried to enable my xmp profile, and then I see that it is unstable, do bios reset and enter the last stable settings and each time it requires more and more vcore to pass the same test that yesterday is been stable.

I cannot cool it any more. (Ek p360 performance water cooling kit). It is not normal to require more and more vcore - something is definitely wrong. On this board i have one failed cpu, now the new one is probably "degrading" fast.

Could I have two bad cpus in a row or the board is killing them??!

Need help.

[edhunterbg]

Hello again,
While waiting for reply from Asus tech support, I did some more tests today.
Using offset mode and ac/dc settings, I marked down few stable configurations.
Common settings: Multiplier - 50x, AVX offset - 0, MCE - remove all limits, LLC - 4, Voltage mode - offset, xmp is disabled

Offset mV, ac mOhm, dc mOhm -- temp C, vcore, current A (avg values under load p95 40k)
A. 10, 0.50, 1.3 -- 95.0, 1.186, 190.5
B. 20, 0.45, 1.3 -- 94.6, 1.182, 189.0
C. 30, 0.40, 1.3 -- 93.8, 1.182, 188.8

So far so good, but ... In the late afternoon I did retest the same settings, and each of them is giving the same voltages, temperature and current, but they fail (usually in the first 5 minutes)!!!!
The ambient temperature at home is about 24-25 degrees all the time. The VRM temperatures are also close. The PC has been running from the morning and I think that the issue may be related to the temperature of the motherboard component that is making problems when it gets hot.

Any thoughts?

[Falkentyne]

edhunterbg wrote:
Hello again,
While waiting for reply from Asus tech support, I did some more tests today.
Using offset mode and ac/dc settings, I marked down few stable configurations.
Common settings: Multiplier - 50x, AVX offset - 0, MCE - remove all limits, LLC - 4, Voltage mode - offset, xmp is disabled

Offset mV, ac mOhm, dc mOhm -- temp C, vcore, current A (avg values under load p95 40k)
A. 10, 0.5, 1.3 -- 95.0, 1.186, 190.5
B. 20, 4.5, 1.3 -- 94.6, 1.182, 189.0
C. 30, 4.0, 1.3 -- 93.8, 1.182, 188.8

So far so good, but ... In the late afternoon I did retest the same settings, and each of them is giving the same voltages, temperature and current, but they fail (usually in the first 5 minutes)!!!!
The ambient temperature at home is about 24-25 degrees all the time. The VRM temperatures are also close. The PC has been running from the morning and I think that the issue may be related to the temperature of the motherboard component that is making problems when it gets hot.

Any thoughts?

Set VCCIO to 1.20v. VCCSA to 1.25v, PLL Terminiation to 1.10v, Internal PLL Voltage to 1.0v, PLL Bandwidth to level 0, Standby voltage to 1.05v then test again.

[edhunterbg]

Falkentyne wrote:
Set VCCIO to 1.20v. VCCSA to 1.25v, PLL Terminiation to 1.10v, Internal PLL Voltage to 1.0v, PLL Bandwidth to level 0, Standby voltage to 1.05v then test again.

Hi Falkentyne. Thanks. I followed your advice and did those settings.
Tested the "C" option and it failed after 7minutes (prime95 40k, error).
Option "B" passes the 12minutes threshold.

[Falkentyne]

edhunterbg wrote:
Hi Falkentyne. Thanks. I followed your advice and did those settings.
Tested the "C" option and it failed after 7minutes (prime95 40k, error).
Option "B" passes the 12minutes threshold.

I've seen chips lose a little of their "absolute minimum" voltage right after use (meaning the voltage has to be raised after about a week), but it usually stops after about 15-20mv after a week or so.

However you have a "KS", so you should be stable FAR Below these voltages you tested Ex: LLC5 + 1.325v should be completely and fully stable in all stress tests. I would RMA that chip since it is not even fully stable at stock speeds (5 ghz all core).

What does thet "VID" report at full idle (Don't run a stress test just look at the VID using HWinfo64), with AC *AND* DC Loadline both set to 0.01? (set SVID behavior to best case scenario, that should set them both to 0.01 for you if they are on 'Auto'. If not, set then to 0.01 manually and set SVID Behavior to base case scenario). You should use a fixed vcore (LLC5 + 1.350v) when you do this to make sure you can load windows.

Please tell me what the VID shows for the cores at idle.

Also can you run a LinX 0.9.6 test? This is faster than prime and uses more current.
https://hwtips.tistory.com/1611

If the Korean really confuses you, you can use 0.9.5 (english) but it's very easy to figure out what's going on.

Try this test with manual voltage, LLC5, 1.350v set in BIOS.

Run it, select 35000 sample size on the left, then select 10 loops on the right.
Your residuals should be the same (matching) on each loop. (it's a math calculation so the solution must be identical each time--this changes based on sample size though)/

[edhunterbg]

Ok I reconfigured for fixed voltage again
BIOS: vcore 1.35, llc 5, ac/dc 0.01, svid best case
HWinfo: VID in idle 1.210, vcore in idle 1.332

Will do the LinX test, but I am pretty sure that it will not be ok on 1.35

p.s. see the attachment, it is different on the third run...
1.35v was ok the first day, but now I dont know .. 1.37v was not enough in p95, 1.38 have not tried. That is why I decided to explore the possibilities with offset mode.

[Falkentyne]

edhunterbg wrote:
Ok I reconfigured for fixed voltage again
BIOS: vcore 1.35, llc 5, ac/dc 0.01, svid best case
HWinfo: VID in idle 1.210, vcore in idle 1.332

Will do the LinX test, but I am pretty sure that it will not be ok on 1.35

p.s. see the attachment, it is different on the third run...
1.35v was ok the first day, but now I dont know .. 1.37v was not enough in p95, 1.38 have not tried. That is why I decided to move to explore the possibilities with offset mode.

Wow, you should RMA that chip.

A VID of 1.210v (with AC/DC LL of 0.01) puts it at the worst bin for 9900KS. 😕 The lower the VID shown, the better the silicon quality. That bin is equal to an average 9900k.
My 9900k is equal to that. That explains why you are having problems with stability there.

That being said, passing 4 out of 5 LinX loops is EXTREMELY good for passing prime95 rock stable, but the very low gflops might be what is helping LinX run better :(. LinX is the hardest torture test on the planet (it was mentioned years ago that intel uses Linpack binaries to validate their own chips, since they do release the binaries themselves, but I have no hard proof of that), although it is VERY RAM dependent on proper RAM timings--bad timings will torpedo your Gflops. You should not be crashing in Prime95 at all if you can get 4 of 5 correct residuals on LinX. I've tested that extensively on my 9900k. And you were doing AVX 40k, not FMA3 40k, right? You are correct that AVX 36k-40k (I found this out in my own testing, so that's very impressive that you discovered this too) is the hardest FFT size to pass, while 112k FMA3 is the hardest to pass for FMA3.

Your Gflops is very low in LinX. You should be at 500 gflops. This is caused by RAM timings being too loose. Usually it's the TFAW timing being far too loose. If you have XMP enabled, try tightening this to "24" and set TRRD_S to 6 (TFAW should be TRRD_S * 4), and set TRRD_L to 6 also.

*edit*. Saw that XMP was disabled. Enable it if you have time then re-test. If your Gflops is still below 480 (500 is proper for 5 ghz), tighten the subtimings I mentioned.

That being said, with mostly correct residuals, you should be storming through 40k easily.

Note on my 9900k, I can't run LinX 35000 @ 5 ghz. It just crashes, unless I raise the voltage high enough that I get over 110C, (> 1.40v bios set + LLC5 (LLC High on my Aorus Master), then it's just too hot. I can do it at 4.9 ghz on all auto voltages and get correct residuals (Load vcore: 1.230v) and AC Loadline=1.6 mOhms + Vcore loadline calibration at Intel spec (Equal to LLC2 on your board), but it STILL Gets over 100C! I'm using air cooling however...so your CPU is still doing better than mine. But that's a very bad "KS".

[edhunterbg]

I dont use XMP yet. Just pure jdec 2133. First I want to get my cpu stable, then I'll deal with memory.
My dimms is in the QVL list - 2x16 gskill 3200 c14 ripjaws
Yesterday I had doubts about memory and tested with another 2x8 kit again without xmp. It didnt help at all.

About the avx or fma3 - I dont know - i think it is the FMA3 - see the attached screenshots.

Some years ago I also used the linkpack to validate max temperature, but I think nowdays the prime 95 is the most power and temperature stress test.
I discovered 40k ffts with many experiments with my previous cpus.

I dint knew about that method for evaluation of binning (vid on idle), usually i disregard the VID and use the vcore (or vrout) on GB board for comparing the cpus.

So at the end - my unit is just named KS, but actually is piece of sh*t 😞 ?

[Falkentyne]

edhunterbg wrote:
I dont use XMP yet. Just pure jdec 2133. First I want to get my cpu stable, then I'll deal with memory.
My dimms is in the QVL list - 2x16 gskill 3200 c14 ripjaws
Yesterday I had doubts about memory and tested with another 2x8 kit again without xmp. It didnt help at all.

About the avx or fma3 - I dont know - i think it is the FMA3 - see the attached screenshot - it is currently running.

Some years ago I also used the linkpack to validate max temperature, but I think nowdays the prime 95 is the most power and temperature stress test.
I discovered 40k ffts with many experiments with my previous cpus.

I dint knew about that method for evaluation of binning (vid on idle), usually i disregard the VID and use the vcore (or vrout) on GB board for comparing the cpus.

So at the end - my unit is just named KS, but actually is piece of sh*t 😞 ?

Got it.
FMA3 is a subset of AVX2. If you disable AVX2 in the stress test options, it will test AVX.

Try 112K in-place FMA3, and see if that fails faster.
112K is even more brutal. It draws more amps than 40k when I was checking.

For some reason, AVX1 (AVX) is more stressful on 36k-40k than the other FFT sizes.
But for FMA3, in my own tests it's 112K that is the hardest but I'll admit I did not extensively test 40k. I only looked at the power draw and a few failure tests and saw i had more problems with the 112K size than 36k. Try it and see what happens 🙂

But yes you should not be failing prime95 at all on a KS, with voltages that high.
Other people posted results in this thread on OCN.

https://www.overclock.net/forum/5-intel-cpus/1737290-9900ks-maximus-xi-extreme-oc-help.html

clkjoje (whatever his name was) had a 9900KS that was just as bad as yours. High VID, failing prime at 5 ghz. He RMA'd it and got a much better chip (I think the default VID on the new one was 1.17v at 5 ghz, with AC/DC at 0.01.

[edhunterbg]

Running it currently 🙂
It seems to draw a 5-10 amps more than 40k.
By the way when prime fails then I check the log of the hwinfo for each core/thread usage and see that almost always core 5 is failing and some times its core 2.

Wow thats an impressive information - thanks. I'll go for RMA most probably.