PC Graphics

[ardankyaosen]

I'm a first time undervolter and just got a new 6700XT.  I've got my setting set as I want them in the latest Adrenaline and am looking at some manual undervolting.  At this point, overclocking isn't of much interest to me.  But, for undervolting, should I run benchmarks (Superposition) with all the frame limiters off in Adrenaline or keep them on (as I would normally use them)?

[ardankyaosen]

Looking at what HWINFO64 reports when I'm running Superposition, the internal framerate of the card runs above what my framerate limits are for the actual displayed image.  So, I just continued with those limits in place.  I guess I'm in as close to a stopping point as I need to get:

• Under Power Tuning, I pushed the Power Limit up the full 15% available to me
• Under GPU Tuning, I left the Min/Max Frequencies at their default 0/2664 and worked my way down the Voltage level to 1080mv (1070 crashed the display driver in Superposition).
• Under VRAM Tuning, I set Memory Timing = Fast Timing and Max Frequency to 2122MHz.  I could probably push that higher, but HWINFO64 is reporting an internal framerate over 500 frames/second running the Superposition Benchmark at 4K.

[ardankyaosen]

Just for posterity, I've moved over to Linux (Fedora 35) and have just gone through a similar undervolting process with this 6700XT there.  I'm running the stock, included kernel drivers without any framerate limiters (primarily because I haven't figure out if I can even apply a framerate limit in Linux yet).  I used Unigine Superposition again as the benchmarking software and got about the same results:  an undervolt offset of -110 mV.  CoreCTRL doesn't show me the starting, absolute power setting for the card, but I'm pretty sure I remember that as 1200 mV.  So, in Adrenaline terms, that's an undervolt down to 1090.  Here are my testing results:

Undervolting AMD 6700XT with CoreCTRL on Fedora and Unigine Superposition (4K)
Voltage Offset mV	Absolute Voltage mV	Percent Voltage Diff	Score	Percent Score Diff	Min FPS	Avg FPS	Max FPS	Min GPU C	Max GPU C
0	1,200	0.00%	9,229	0.00%
-10	1,190	-0.83%	9,059	-1.84%	58.63	67.76	81.27	41.0	79.0
-20	1,180	-1.67%	9,090	-1.51%	58.63	67.99	81.89	49.0	80.0
-30	1,170	-2.50%	9,144	-0.92%	59.06	68.10	82.28	49.0	80.0
-40	1,160	-3.33%	9,186	-0.47%	59.04	68.71	82.81	48.0	80.0
-50	1,150	-4.17%	9,230	0.01%	58.71	69.04	82.88	48.0	80.0
-60	1,140	-5.00%	9,291	0.67%	59.88	69.49	83.22	48.0	80.0
-70	1,130	-5.83%	9,339	1.19%	60.19	69.86	83.76	46.0	80.0
-80	1,120	-6.67%	9,408	1.94%	60.82	70.37	84.48	48.0	81.0
-90	1,110	-7.50%	9,438	2.26%	61.02	70.60	84.90	48.0	80.0
-100	1,100	-8.33%	9,465	2.56%	61.33	70.80	85.07	49.0	81.0
-110	1,090	-9.17%	9,474	2.65%	61.09	70.86	85.13	49.0	80.0
-120	1,080	-10.00%	9,466	2.57%	60.62	70.80	84.94	49.0	80.0

I tested to a -120 mV voltage offset.  But, since my performance actually went down from the -110 mV offset, I just went back to that and stopped.  Assuming my 1200 mV starting point is correct, that's a bit more than 9% voltage reduction with a bit over 2-1/2% performance increase.

EDIT:  If anyone ever sees this, maybe they can check my math on the following power savings.

P = V^2 / R.

So, the percent power savings would be:

(P2 - P1)/P1

or

(((V2^2)/R) - ((V1^2)/R))/((V1^2))/R)

((V2^2 - V1^2)/R)/(V1^2)/R)

((V2^2) - (V1^2))/(V1^2)

((V2^2)/(V1^2)) - 1

((V2)/(V1))^2 - 1

(1.09/1.2)^2 - 1

-0.17 or a 17% power savings?