Currently Online 
bonzobanana said:
However I'm not sure I want to argue against your laptop battery runtime estimates because if anything they reinforce my view that the Switch 2 either will have far lower performance than stated in portable mode or much shorter battery runtime. If the RTX 2050 mobile only gets lets say 90 minutes battery runtime out of 70Wh then using the same fabrication process how low does the Switch 2 have to go to get 2 hours minimum out of 20Wh? I remember when modders/hackers claimed in a forum that the Switch was dropping to 90-140 Gflops based on its reduced power consumption when portable compared to peak power. That gives an average of about 115 Gflops when the maximum potential performance of the Switch when portable is 236 Gflops so roughly twice as fast as reality. Surely the Switch 2 will be doing the same with a sub 1 Teraflop gflops figure, probably 600-800 Gflops in reality in portable mode. Switch 2 has a maximum 6.5hours runtime so with reduced brightness that still only leaves 1-2 watts per hour. 20W shared by 6.5 is only 3W and at that point the screen is probably needing more power than the SOC. I really can't see how the figures given for portable performance of the Switch 2 stack up. Where is it getting all the additional power to deliver such performance. I think we will see a huge difference this time around between portable and docked performance maybe 3-4x the power. |
What you are missing (or ignoring) is that the Switch 2 doesn't have an underclocked GA107. The T239 is a chip designed for a 5-20W power curve. It is like saying, how could they fit 4.3 TFLOPs of performance in a 30W profile (RTX 2050) when an RTX 3090 (on the same manufacturing node and general micro-architecture) consumes 350W to achieve 35.58 TFLOPs? Shouldn't it be only 3 TFLOPs not 4.3 TFLOPs, if linearly scaled per watt?
Well chip design matters and this power-scaling isn't linear.
Even on the same node and architecture different chips have different power efficiencies at different watt ranges. Nodes mature over time. Professional versions of these chip families (which the T239 is closer to in terms of SOC architecture) tend to consume less power than consumer versions for the same performance even if manufactured on the same node. For example, and a500 Mobile (professional RTX 2050 equivalent) achieves 6 TFLOPs @30W vs. 4.3 TFLOPs @30W for the 2050.
Also the CPU in the Switch 2 is a very low powered ARM CPU, nowhere near as power-ineficient (especially at low sub-10W consumption) as the Raptor Lake Intel chip in the laptop you shared, and about half of the battery consumption there, is because of the CPU.
Last edited by sc94597 - on 27 May 2025
