Edit: Anyway very clean and nice looking build and work. From my experience related with cross-flashing 2070s BIOSes only Stock(MSI) and FE have similar results clock for clock. All other are giving worse results even having better TDP limits.
Yeah - that Ryzen 1700 is mostly for productivity. It's definitely holding back the gpu in applications like this. -as you can possibly tell from the fact that my machine is the fastest with that combination of hardware. Besides that your sons 2070 super is running at almost the same clock frequency as my 2070, but with more cuda cores. Ofcourse it will be faster ;-)
I'm also a fan of silent running computers, so that water cooler is as much a noise reduction measure, as a performance enhancing measure.
The Gigabyte bios gives about 3-4% extra performance on my card. It will simply hold boosts way longer due to the higher TDP. This is also true without water cooling. :-)
... in this dx12 particular application your [email protected] is not holding the GPU.
I get what you are saying - I'm sorry if my point didn't make it across. English is not my first language. I meant, that My Time Spy score is being held back because it takes both GPU and CPU into account. Thus a more powerful CPU will yield a better overall score, even if the GPU is a bit slower. My overall RTX 2070 rank is #394, yet my gpu score is 300 points higher than #3 overall 2070 benchmark. :-)
Boost clock of my overclocked Palit 2070 Super mostly depends on GPU temperature. I can push over 11k GPU score in Timespy (and keep boost over 2000) when I set the fans to 100%, but that's quite noisy. Using the BIOS 2 (quiet) fan-curve it more around 10.5k and much less audible. This is at a +85 MHz GPU core OC and 500 - 700 MHz memory OC.
Keep in mind that even when Superposition Benchmark reports no errors and you don't see anything funky on screen that your higher overclocks may not be fully stable. You need to run several hours of Realbench to get more reliable results, because its hash calculation really catches even single bit errors of both GPU and memory.
I totally agree on the benchmark thing. Over the years I've dialed in several "stable" overclocks in benchmarks, only to have them fail in real games. I have not idea what kind of Time Spy scores I could be getting if I would only go for a "Time Spy-stable" overclock. Perhaps that is a project for another day :-)
So - I got curious and did a run with fans on full blast, and a boost clock speed that I know won't fly in the real world. A GPU score of 10.700 on a non super 2070. Almost a clean 100 points above what is actually stable. That's pretty decent in my book. Boost clocks only momentarily dipped below the 2.100 mark. A shame it won't fly in the real world :-)