Z270 Max IX Formula/7700k cache OC

[Johan45]

Just got the new Kaby CPU and board on Sunday. First test drive seemed to be a decent CPU, set XMP in BIOS for Trident Z 3600 Cl17. It would run Aida64 @ 5.0 with 1.26v, XTU with 1.3v. I'm thinking this is pretty sweet until I tried the cache. Set cache multi to 48 an system wouldn't post. Upped core voltage to 1.4 eventually got into Win10x64 but it froze. To me that was way too much of a V_Core bump just to get cache to almost work at 4800. I have had multiple Skylake CPUs and never an issue with cache overclocking.
So I updated BIOS fro ver. 222 to ver. 701, re-installed Win10, used up to date chipset/ MEI drivers from support site and started over. The only other driver installed was GFX for the 980Ti. Still acted the same way so I backed dow nt 4.8 core which runs fine at 1.2v. Tried cache at 4.5 and no post. 4.5 cache is really not a large OC. for the cache and shouldn't require any additional voltage to the core/cache rail IMO. I did however up the PLL term. voltage to 1.2 and made it to windows but Aida froze while I was opening it. Bumped term voltage to 1.22 and core to 1.21v. This time Aida ran for 30 minutes and I shut it down. Restarted and set cache to 4800 and no post again. Got a little frustrated and shut it all down at that point.
So my questions
Is any one else having this issue with the New formula/7700k, I'm wonder if it's an immature BIOS issue or if I just have a dud CPU ?
This isn't my first rodeo http://hwbot.org/user/johan45/
TIA for any insight/advice

[Raja]

No issues here on my CPU. Can run 4.5G cache without pushing things. Probably a weak CPU for cache clocking. Not that it matters much.

[Johan45]

Thanks for the response Raja, I know it may not matter much in many situations but in benchmarking it can effect your scores quite a bit.
I was leaning toward the CPU/cache being a bit weak but typically a bit of extra voltage will straighten that out. This one just refuses to run higher.
Would you have any suggestions to secondary voltages that may help me with this or do you think that's a waste of time.

[Raja]

The secondary voltages would be SA and IO as you're pushing throughput when you increase the cache frequency. You can easily debug that by dropping the memory frequency, but, usually a good CPU will do 4.5GHz cache without needing too much cajoling. If you're into competitive benchmarking (no other reason to be so hung up on cache), you won't know how good the CPU til you take it cold. That said, the Formula isn't the best choice for sub-zero.

[Johan45]

I didn't think so either but it was what they had in the store, all others were too weak and I prefer the ASUS boards. What would you suggest for LN2. If the CPU doesn't straighten out then I'll likely sell the whole lot and try again.
After stripping all the plastic off wouldn't it be about the same as most other Higher end ASUS boards plus a few more voltage options?

[Raja]

Apex would be best for benching on Ln2.

[Tsvetan]

Great overclock, mate! Mine needs 1.36v for 5ghz and gets quite hot. I don't think I have that Cache issue though and I am with a much, much cheaper board. Will run a few tests right now at 4.8 as temps are better with this cheapo AIO at my disposal right now and let you know.

Edit: Cranked up the cache multiplier from 42x to 48x in Windows using Intel Extreme Tuning Utility - 10 minutes no problems on AIDA64 Stability test. Will reboot and set it directly in BIOS.

P.S. When I was first overclocking this CPU, cranking up the Cache didn't result in a higher Cinebench score for me. What benchmark benefits from this?

Edit2: I realised that despite I set the cache ratio to 48x in Intel XTU, Cache speed showed up at 4600mhz. So that's why it was stable. Setting it to 48x in BIOS didn't go well - no POST or freezing during loading Windows, even after cranking up the vcore to 1.25v. So yes, 46x cache is stable without any additional work at 48x core multiplier. I think I had it cranked up to 48x with 1.36v on the vcore @ 5ghz but I'm not sure.

[Johan45]

Just to follow up. It definitely was the CPU. I exchanged that one for another and it appears to be working much better. This is on a chilled loop.