I am running a 24x7 overclock at 4.7GHZ (4.3 uncore) @ 1.24v and memory overclocked from 1866 MHz to 2400 11-13-13-31-2T.
First to note, the 23h microcode has patches for Spectre (see the powershell reports at the end). I have been running 23h for a little while and I have noticed some stability issues. There is also a definite performance hit between 22 and 23.
I compared performance (Realbench 2.43) and stability (prime 95 28.10 @ 1344/1344 / 10GB memory) for 15 minutes. My method was:
1. I saved my OC BIOS profile to USB
2. Modified the BIOS for the microcode version I was testing
3. Flashbacked the BIOS
4. Set the BIOS to optimal defaults, rebooted
5. Loaded the saved OC BIOS and rebooted
6. Ran the tests
For both microcode versions I am running KB4056892 Windows 10 v1709 Build 16299.192 which is the Intel security patch. So performance differences are strictly due to the microcode revisions.
To cut to the chase, I am going to seriously be looking at a Ryzen 1800X for my next build, if I can find a suitable Motherboard.
I ran 2 passes of Realbench and averaged them:
ucode 23h 128,265
This is approx. a 9% drop on performance with microcode 23h. Note this is in addition to any Windows 10 performance drops.
For stability I had noticed I was getting a lot of WHEA errors with the 23h microcode. Here is the results of 15 minutes running 23h:
These are correctable hardware (internal CPU Northbridge) errors that also show up in the Windows event log. Each of these causees interrupts if nothing else.
For ucode 22h I get nada , zero, 0 of these errors !
To show that the 23h has patches for the security vulnerabilities, here are the powershell script results that test this:
22h no hardware BTI support:
and here is 23h with support: