So upon further tinkering and testing the card it is very different than even their last RDNA2 cards. With this card it almost doesn't matter what you set your "target" clocks at in the Radeon wattman overclocking utility in the AMD drivers. The card's core clocks all but ignore your requested targets entered in the driver, the only clock that behaves as expected is the memory clocks in that you can set it at whatever specific clock speed you want. The core however is really only manipulated by lowering the core voltage and then increasing the card's power budget. Lowering it from the default 1150mv to just 1100mv and setting the card to allow the max +15% power results in the core clock skyrocketing from holding roughly 2400-2500Mhz at default to holding 2700-2800Mhz when undervolted and allowing the card to draw more power.
Undervolted and higher power budget = Time Spy 26396 (which is a 400+ point increase from above)
https://www.3dmark.com/spy/34336442 Another thing is there is apparently no way to adjust or even see which clock you are manipulating be it the core clock or the shader clock. I assume the driver is reporting the core clocks and that's what you are able to adjust but I would have really liked to even see what the shader clocks are running at. AMD stated the RX 7900 XTX has shader clocks of 2.3Ghz and "front end" clock of up to 2.5Ghz. I remember back to the GTX 280 days when Nvidia tried a similar technique with the shader clocks and core clocks. The difference was that they were set on ratios and you could actually adjust them with tools like RIVA Tuner. No such luck with AMD's RDNA3 it seems. I've tried all the usual suspects like MSI Afterburner, Sapphire's Trixx and ASUS's GPU Tweak but all are the same, they only report one core clock speed.
EDIT: This is pretty nuts, it gave me an even bigger bump in Port Royal at 16681 (almost 700 points higher than above)
https://www.3dmark.com/3dm/86543548? EDIT2: The trend continues in Speed Way at 6359 (which if talking about % gain is the highest yet from the above results)
https://www.3dmark.com/sw/228610 Seems the RT accelerators really appreciate the bump in clock speed where as in rasterized workloads the core clock doesn't benefit performance as much. Very interesting.
Makes me actually wanna watercool this thing and see what it can really do.
EDIT3: This'll be the last update tonight but lowering core voltage to 1050mv gives yet another performance boost and has the card running at 2800-2900Mhz which is good for a Time Spy score of 26836. It held an average of almost 2850Mhz through the test.
https://www.3dmark.com/spy/34337491
post edited by CraptacularOne - 2023/01/04 20:58:39