ROG Crosshair X670E Gene + Artic Liquid Freezer II 280 and ROG Thor 1000P + Ryzen 9 7950X
I didn't give up and since my piece is not completely useless, some of the cores even have SP124 (CCD0 is very decent), so I tried to push the limit to the very edge of my cooling. In time I will measure it with a 360mm cooler, which I have already ordered from abroad and should arrive by the end of the month.
I'm deliberately using an older BIOS on this board, admittedly it won't allow as good RAM frequencies as say 0805 and newer, but then again for 24/7 it offers better single core boost performance on multiple cores.
The whole issue is just how you cool the processor relative to the voltage. The voltage itself that I used for the tests is not extreme even for a Zen4, and with a full conductor it would be perfectly fine to run long term. I went from 1.25V incrementally up to 1.285V. I would need say 1.3V for the extra MHz, but the heatsink would not be able to cool it effectively and the score would be worse... I used a combination of DOS (Dynamic OC Switcher) and PBO with added CO (Curve Optimizer) and FMax and Scalar functions. DOS will guarantee manual values for the allcore load and also a high less-threaded frequency when PBO is currently being used. Perhaps all the Ryzens I have encountered in the 5000 and 7000 series always have better CCD0 than CCD1. Therefore setting the clock rates separately for CCD0 and CCD1 can still increase allcore performance. In the end, I ended up with 5475 MHz on the CCD0 and 5375 MHz on the CCD1. However, the differences between the CCD0 and CCD1 are usually around 50-75 MHz, for me it's 100 MHz (CCD0 is very good, CCD1 is about average). Thanks to PBO+CO, my Ryzen 9 7950X alone boosts up to 5975 MHz in idle, or even 5980 MHz depending on how lightly bclk flies . The maximum effective clock speed is then 5900 MHz for single core (5910 MHz). Which is decent. FMax itself works a bit like Intel Velocity turbooost, i.e. under condition A and condition B the CPU clock speed can be up to XXX. AMD defines it up to 5850 MHz, just because of the curve optimizer it then moves even further. But in general it depends on the CPU piece (silicon quality), on your cooling and therefore on the CPU temperature.
The first example is Superpi, i.e. using DOS switching to PBO+CO+FMAX mode, as you can see the frequency was up to over 5900 MHz in the pass test
Next examples of multicore tests
I ran the older Cinebench R11.5 on a comfortable 5400/5325 MHz setting with 1.25V
I tried the 7-zip a bit more conservatively and had the frequency lower at the time, I think it was something like 5425 MHz for CCD0..
On Wprime I slightly increased the frequency again, so far everything is holding and not falling..
Cinebench R15 and the first test to see if 5475MHz+5375MHz can stand up to the more demanding R20 and R23 tests
Cinebench R20 netted 16,000 points, which I was happy about because previously the results were like 15 8xx or 15 9xx
Latest Cinebench R23 exceeded 41k points, 5475 MHz and 5375 MHz in CCD units
Generally speaking, Raphael is the best clocked AMD, it may sound ridiculous, because compared to the default you only get a few extra MHz (about 400 MHz for allcore and still only stable for a half-minute test), but the CPU runs at very high frequencies without suffering its IPC. Few people could imagine a CPU that goes further in frequency than Alder Lake. Reportedly, it's an interesting chip even for liquid nitrogen, where its only drawback is possible coldbug and coldbootbug. There it's a lottery, some chips have it practically no, others have it quite early (around -100C) and others have it up to maybe around -165C. Even a good chip on water can then ruin coldbug under nitrogen. However, the golden bits then have their limit around 6750-6850 MHz for heavier tests and around 7100 MHz for lighter benchmarks. It's not quite up to the extreme frequencies of the Raptor, but it's still a jump up by about 1 GHz for the modern era of Ryzen.
Bookmarks