Many congratulations and thanks for the incredible research works and efforts you are doing in the field! Thank you Prof. Chen for amazing presentation! I'm not quite sure that I could understand correctly why sacrificing the efficiency is acceptable for CPU/GPU power supply application. It is true that the smaller power supply can get closer to the processor and cooling system, but we are producing more heat at the same time due to higher power loss, unless they are not linearly related. Thanks in advance if anyone from your team could get a chance and help me understand the philosophy better!
Just like real estate, the closer you are to Manhattan, the more $$$ each mm2 cost, and the air conditioner and electricity bill cost nothing ... reducing VRM size can open lots of opportunities in computing (e.g., faster chip to chip communication, faster memory access, better packaging), which worth much much more than electricity bill and cooling cost.
Thanks for the explanation! That was my confusion. If achieving 2x higher power density is possible at the cost of only 10% more loss, then that makes sense. I'm really curious to know what happens to this trade off if more computation power (more transistors) is added to the same processor owing to the saved space, since you would need higher rated power VRMs. Have you found any overhead limit for this trade off? Is there any restriction for maximizing power density at the cost of tolerating more loss for the entire cooling system, VRM, and CPU/GPU?
The typical thermal limit is about 10W/mm2, while the best 48V VRM density is about 1W/mm2, still far away from the cooling limit. Advanced cooling can push the thermal limit beyond 10W/mm2. @@saeedsharifi2588
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
Many congratulations and thanks for the incredible research works and efforts you are doing in the field!
Thank you Prof. Chen for amazing presentation! I'm not quite sure that I could understand correctly why sacrificing the efficiency is acceptable for CPU/GPU power supply application. It is true that the smaller power supply can get closer to the processor and cooling system, but we are producing more heat at the same time due to higher power loss, unless they are not linearly related.
Thanks in advance if anyone from your team could get a chance and help me understand the philosophy better!
Just like real estate, the closer you are to Manhattan, the more $$$ each mm2 cost, and the air conditioner and electricity bill cost nothing ... reducing VRM size can open lots of opportunities in computing (e.g., faster chip to chip communication, faster memory access, better packaging), which worth much much more than electricity bill and cooling cost.
Thanks for the explanation! That was my confusion. If achieving 2x higher power density is possible at the cost of only 10% more loss, then that makes sense. I'm really curious to know what happens to this trade off if more computation power (more transistors) is added to the same processor owing to the saved space, since you would need higher rated power VRMs. Have you found any overhead limit for this trade off? Is there any restriction for maximizing power density at the cost of tolerating more loss for the entire cooling system, VRM, and CPU/GPU?
The typical thermal limit is about 10W/mm2, while the best 48V VRM density is about 1W/mm2, still far away from the cooling limit. Advanced cooling can push the thermal limit beyond 10W/mm2. @@saeedsharifi2588
Great! Long way to go.
Thanks for your replies!@@princetonpowerelectronics168