Re: G-SYNC Pulsar: Blur Reduction or Black Frame Injection in cojunction with Variable Refresh Rate or Adaptive Sync
Posted: 10 Jan 2024, 01:12
Minimum Six Algebra Dimensions! Possibly Ten!
G-SYNC Pulsar is a almost certainly a very advanced six-dimensional algebraic overdrive implementation, I think, that utilizes:
- [+1] Scanline position (stronger OD at bottom of screen, due to scanout)
- [+1] Frametime (dynamic VRR overdrive)
- [+2] Gamma corrected overdrive to prevent R/G/B color shifts (3 vars instead of 1 variable)
- [+2] On top of usual A(B)=C overdrive formula
= [6 algebra dimensions]
This would be a minimum six-dimensional algebraic overdrive, probably generated out of advanced algebraic regression curve-fitting, since you cannot strobe-tune that many settings in one sitting, and the lookup table would be literally petabytes (or more). So it gotta be algebraic-formula-based overdrive, per subpixel, computed algebra formula ~4 billion times per second (2560 x 1440 x 360 x 3).
That doesn't even include potential multi lookbehind frametime history and pixel color history, because incorporating that additional information improves overdrive further. Sometimes clamping a pixel to 255 for 2 consecutive refresh cycles is needed to speed up the slowest "to near white" pixel transitions, when refreshtimes are so scarily fast (2.78ms) and some colors take more than two refresh cycles to finish. So you got multi-refresh-cycle-depth overdrive algorithms.
So we might even be looking at 8-dimensions or 10-dimensions!
Without even looking at NDA material and never even told what G-SYNC Pulsar does, I know how complicated overdrive tuning gets when you throw in all these extra variables. I've had to butt heads with scaler/TCON vendors already on this. This is a whole other ballpark that you will never find in generic chinese scaler/TCONs.
ELMB SYNC doesn't even have ANY of this, except standard A(B)=C overdrive formula in simple LUT. You get lucky and you may be able to realtime-compute the frametime too, in some of the more expensive FreeSync implementations.
So Pulsar seemed probably 100x+ better in crosstalk behavior consistency. Much, much, much, much less crosstalk yoyo-effect.
ELMB SYNC is mathematical bush plane league hack. This is a Concorde in mathematical brilliance.
But more testing is needed by reviewers to see if spending all this massive engineering money to improve the most difficult 10% of overdrive, provides enough dividends in this refresh rate race.
G-SYNC Pulsar is an engineering achievement in strobed VRR, whole other ballpark. Not even in the same solar system as ELMB SYNC, mathematically. I wonder if my whisper of "algebra regression formula" to Seth Schneider after ULMB2 announcement, gave some ideas to NVIDIA. I was egging them on my dissapointment that ULMB2 couldn't do flexible Hz. Ha, I wonder if I started that Chaos Butterfly flapping. Algebra regression curve-fitting is easy with Wolfram Alpha. The challenge is merging all the data together from a long automated photodiode oscilloscoping session, as there are possibly quadrillions of data samples there.
Who knows?
I'm at CES 2024.
G-SYNC Pulsar is a almost certainly a very advanced six-dimensional algebraic overdrive implementation, I think, that utilizes:
- [+1] Scanline position (stronger OD at bottom of screen, due to scanout)
- [+1] Frametime (dynamic VRR overdrive)
- [+2] Gamma corrected overdrive to prevent R/G/B color shifts (3 vars instead of 1 variable)
- [+2] On top of usual A(B)=C overdrive formula
= [6 algebra dimensions]
This would be a minimum six-dimensional algebraic overdrive, probably generated out of advanced algebraic regression curve-fitting, since you cannot strobe-tune that many settings in one sitting, and the lookup table would be literally petabytes (or more). So it gotta be algebraic-formula-based overdrive, per subpixel, computed algebra formula ~4 billion times per second (2560 x 1440 x 360 x 3).
That doesn't even include potential multi lookbehind frametime history and pixel color history, because incorporating that additional information improves overdrive further. Sometimes clamping a pixel to 255 for 2 consecutive refresh cycles is needed to speed up the slowest "to near white" pixel transitions, when refreshtimes are so scarily fast (2.78ms) and some colors take more than two refresh cycles to finish. So you got multi-refresh-cycle-depth overdrive algorithms.
So we might even be looking at 8-dimensions or 10-dimensions!
Without even looking at NDA material and never even told what G-SYNC Pulsar does, I know how complicated overdrive tuning gets when you throw in all these extra variables. I've had to butt heads with scaler/TCON vendors already on this. This is a whole other ballpark that you will never find in generic chinese scaler/TCONs.
ELMB SYNC doesn't even have ANY of this, except standard A(B)=C overdrive formula in simple LUT. You get lucky and you may be able to realtime-compute the frametime too, in some of the more expensive FreeSync implementations.
So Pulsar seemed probably 100x+ better in crosstalk behavior consistency. Much, much, much, much less crosstalk yoyo-effect.
ELMB SYNC is mathematical bush plane league hack. This is a Concorde in mathematical brilliance.
But more testing is needed by reviewers to see if spending all this massive engineering money to improve the most difficult 10% of overdrive, provides enough dividends in this refresh rate race.
G-SYNC Pulsar is an engineering achievement in strobed VRR, whole other ballpark. Not even in the same solar system as ELMB SYNC, mathematically. I wonder if my whisper of "algebra regression formula" to Seth Schneider after ULMB2 announcement, gave some ideas to NVIDIA. I was egging them on my dissapointment that ULMB2 couldn't do flexible Hz. Ha, I wonder if I started that Chaos Butterfly flapping. Algebra regression curve-fitting is easy with Wolfram Alpha. The challenge is merging all the data together from a long automated photodiode oscilloscoping session, as there are possibly quadrillions of data samples there.
Who knows?
I'm at CES 2024.
