Post it. And even then if it works it only works on a few units that have miracle RAM at that voltage. And I seriously doubt how stable that might be. It could boot into a game and then later crash but you wouldn't show that. I could undervolt shit slightly on my Switch as well but it would only work for my unit because every unit is different. It doesn't mean much. Still doesn't change the fact that your 2400Mhz was getting lower performance then 2133Mhz.
Done. Sorry for keeping you waiting. I hope this is what you expected:
VHQ Video |
VHQ Mirror (watch or doianload) |
YouTube Video
I made this video, taking into account all your wishes. It was recorded in one 27-minute take without gluing, on it you can see at least every pixel of each frame. I was not too lazy to use a tripod and a camera.
Obviously, it is not mounted, there are no tricks in it, it is extremely unsophisticated. This video is a testament to the fact that I somehow managed to do what no one else could before (at least on Switch). You, CTC, PSam, even KazushiMe, whoever. None of you not only did not understand how to achieve a similar result, it seems impossible to you, and I know why:
Stable 2400MHz on Nintendo Switch on undervolted volages? - Soind like a joke.
Now you are faced with a choice of two:
A. You can go on picking on minor details, looking for the most improbable arguments - designed to devalue my work, looking for fraud and extremely unlikely explanations for how it happened.
In that case, I'll just end the pointless dialogue and move on.
But there is another option:
B. The explanation for the phenomenon lies on the surface, it is extremely unpleasant to agree with it. According to this explanation, I was able to do (not for the first time) what others could not, because I understand this better than other Switch scene developers.
It is unpleasant to agree with this, I agree, but this option is accompanied by a bonus. You can ask me to teach you what I know. Take for granted my expert, and ask to explain to show how this is even possible. That way, you'll find out almost instantly, was the decision correct?.
I've wasted some of my time trying to get you to the right conclusions. If I didn’t succeed, it’s okay, you don’t owe me anything.
I'll go to import the innovation into the mainline 4IFIR. I wonder what results will be achieved in tandem with other optimizations, like timing adaptation, etc.
Some of the conclusions that I made during the successful "proof of the viability of the hypothesis":
- The values obtained are far from the limit of what can be obtained for free, remaining even within the limits of the standard voltage, not to mention a slight overvolting.
- It is able to function perfectly stable.
- MEMORY MATTERS! It still remains the bottleneck of the system. It is not obvious to everyone, but the increase in memory bandwidth allows you to get the same performance as before, at lower GPU / CPU frequencies, and as a result - less power consumption.
- It seems that I will be able to reproduce the method on ERISTA samples as well.
P.s. If the first option is chosen, for God's sake, don't feel bad about reverse-engineering my waste. Just ask me how it's done and I'll explain, saving you time.