| numberwang said: Steam Deck has 1.6 TF with TSMC 7N/6N process and the SOC can reach >90°C in complex 3D games. Switch 2 has to be slimmer, cooler, lighter, longer battery life at a worse (Samsung 8nm) or comparable process node (Samsung 5nm). You can't make comparisons to home consoles with their high-juice SOCs. OG Switch has 0.15 TF in handheld. Going to 1 TF is an 8x increase and I am not convinced that we get 8x more raw performance. |
The Steam Deck goes for a high frequency, few CU design and has an x86 CPU that isn't going to be nearly as efficient at low wattage, regardless of node. It is far less customized than the T239.
Again, we already know how many SM's the T239 has. It's a matter of calibrating for the right frequency. 230 MHz is way too low for an Ampere chip, and iirc impossible at non-idle P-States. The minimum Ampere frequency is ~400MHz. That puts the minimum possible TFlops at about 1.2.
You can't compare TFlops across architectures like that anyway. Maxwell isn't Ampere isn't RDNA2. As Haxxiy noted earlier in this thread real performance per TFLOP is much higher on RDNA 2 than Ampere to the point that an Ampere "TFLOP" (for the purpose of extrapolating gaming performance from raw floating point compute) is something like 60 to 70% of an RDNA2 "TFLOP", if you want to extrapolate gaming performance (gaming is a mixed FLOAT - INT compute-load versus, say, Machine Learning or Scientific Computing.) 1.6 TFLOPs of the Steam Deck should give similar gaming performance to 2.3 TFLOPs of the Switch 2, all else being equal.
Last edited by sc94597 - on 04 January 2025
