3 ms·

The solution of "huge beefy modular power supply" involves starting with a power supply that is way overkill; here, we can expand piecemeal. Nothing is more mod

by pxx 3y ago

The solution of "huge beefy modular power supply" involves starting with a power supply that is way overkill; here, we can expand piecemeal. Nothing is more modular than a separate unit. Modular power supplies don't allow you to "upgrade" them, only to reduce cabling. Re: voltages. I mean the opposite. The idea here is that we could supply, say, 50V for the graphics card to step down now that we're no longer constrained by system voltages. Now there is much less current going through the fiddly connector. A 2200W power supply constrained by 12V is delivering at least 180A.

oakwhiz 3y ago
I could get behind this idea. Open Compute has a 48 volt rack specification, negative 48V is common in telecom since forever, PoE standards use roughly 48V, and 48V is at the upper limit of what's often considered "safe" by various standards and many (but not all) regulatory bodies across the world. What hurts 48V adoption is typically component cost. In the past there was a large price jump between 12V and 48V-60V capable circuits, with a lot of competition and demand at 24V from industrial stuff but less at 48V. Part of the problem is the huge ratios of 10x-50x between the voltages and the idea that the semiconductor switching elements have to be able to handle the peak voltages encountered in the system. In order to realize a 48V system at minimum cost you kind of need a flyback style power converter, leaning on a transformer's winding ratio to help. Dealing with custom magnetics sucks and engineers have in the past related to me how much they try to avoid dealing with magnetics. The frequency you want to use is high and this makes the magnetics smaller but increases the stress on the switching element (MOSFETs). The power you get off a flyback-buck like this can be a little noisy if not postregulated so you might need clever tricks to regulate it better to reduce that in the first place. I have noticed that high powered PoE adoption and lithium battery pack demand (like for electric bikes etc) is driving some of the costs on these things down a bit.

by pxx 3y ago

oakwhiz 3y ago

I could get behind this idea. Open Compute has a 48 volt rack specification, negative 48V is common in telecom since forever, PoE standards use roughly 48V, and 48V is at the upper limit of what's often considered "safe" by various standards and many (but not all) regulatory bodies across the world. What hurts 48V adoption is typically component cost. In the past there was a large price jump between 12V and 48V-60V capable circuits, with a lot of competition and demand at 24V from industrial stuff but less at 48V. Part of the problem is the huge ratios of 10x-50x between the voltages and the idea that the semiconductor switching elements have to be able to handle the peak voltages encountered in the system. In order to realize a 48V system at minimum cost you kind of need a flyback style power converter, leaning on a transformer's winding ratio to help. Dealing with custom magnetics sucks and engineers have in the past related to me how much they try to avoid dealing with magnetics. The frequency you want to use is high and this makes the magnetics smaller but increases the stress on the switching element (MOSFETs). The power you get off a flyback-buck like this can be a little noisy if not postregulated so you might need clever tricks to regulate it better to reduce that in the first place. I have noticed that high powered PoE adoption and lithium battery pack demand (like for electric bikes etc) is driving some of the costs on these things down a bit.