Steam Frame vs. Quest 3 Specs: Better PC Streaming, Power & Hackability

A core by itself may not be optimized for high burst alone, but a SoC most certainly is. And in reality the reason why SoC now have several types of cores is exactly because they are optimized for either burst, sustained or low power performance.

The SD8 Gen 2 had a 1/2/2/4 high performance/performance/efficiency/low power core configuration, while the equivalent XR2 Gen 2 had a 0/2/3/0 configuration. There is no way you can drive all cores of the SD8 Gen 2 at the same time, the TDP simply doesn't allow it, and you cannot simply overclock and cool it with liquid nitrogen. The XR2 Gen 2 is also produced on an older Samsung process that is sufficient for its significantly lower core clocks and reduces the price.

So in effect you can run the highest performance level possible on the XR2 Gen 2 with "normal" cooling for a longer time. You cannot do the same with the SD8 Gen 2, where this would quickly lead to thermal throttling, but that is by design, as the high performance core is only used for burst activities. And the main reason for this is to allow the whole phone to be able to return to a low power state faster, which more than compensates the higher power draw of the burst core.

I'd expect Valve to have come up with decent cooling for the SD8 Gen 3, and easily outperform the Meta headsets that underclock the CPU. But there is only so much you can do in a 195g core module, not allowing for some crazy overclocker cooling setup. So it is very unlikely that even Valve will be able to run the X4 high performance core at its max speed for more than a few minutes, let alone in parallel to the performance and efficiency cores. The SD8 Gen 3 with a 17W peak/8W sustained power TDP is simply not designed for that.

TL;DR: No, they cannot easily drive it at 17W sustained. They don't even try, because that would be stupid, and the Gamers Nexus video explains why.

The Steam Frame is not a PC where you can add lots of airflow and cooling, and phone SoC work very differently from PC CPUs that are mostly designed for sustained load. Apple uses similar core configs on both Macs and iPhones, but Qualcomm relies a lot on alternating between burst and low power cores.

And the rules for max power are pretty universal. On high core CPUs like Xeon the per-core clock goes down a lot when you run all of them in parallel. People using old Xeons in gaming rigs regularly disable extra cores to allow the remaining ones as higher speed, because many games need more single core peak than multicore performance, possible as the disabled cores no longer count against the total TDP. And while PCs and consoles can just throw more power at things to improve performance (up to the chips limits), in mobile devices everything is about saving power, and running anything at max TDP for a longer time is just plain stupid, as relative performance gains decrease with higher power, while it kills battery life. With a Steam Frame 21.6Wh battery, just the SoC alone would burn through this in about an hour at max TDP, even without the RAM or very bright displays and other parts eating more power. If we assume that Valve is aiming for at least 90min play time, the whole system mustn't draw more than 14.4W sustained.

The SD8 Gen 3 is designed for a sustained TDP of 8W, with short bursts. And in the Games Nexus Steam Frame engineering deep dive the Valve engineers clearly state that to even test running the SD8 Gen 3's max thermal dispensation, they had to custom build a water cooled solution to drive it without throttling. Through clever PCB layout Valve was able to add a second heat pipe for improved cooling, but that still doesn't allow them to drive it at twice the designed sustained TDP, which would required for max load. They did a lot of clever things like very precisely measuring the separate power rails for the different SoC clusters to learn where exactly the power went, plus tons of thermal sensors all over the device, but this is to find the optimal performance/power balance, not to magically allow them to drive the SoC way beyond its physical capabilities.

Instead they will be able to configure a game specific optimal power distribution. One game with sustained load might only use performance cores with the GPU fully enabled, another turn-based game with regular compute bursts instead use the high performance core for computation, increasing heat, but then allowing the whole SoC to cool down afterwards. It does not mean running the SoC at full power, it means running the SoC at optimal power/performance ratio. Pretty much what made up "Valve magic" on the Steam Deck.

And they aren't even maxing out the cooling capacity of those two heat pipes because this would require a very large and noisy fan, with the Valve engineers saying they are aiming for a "sweet spot" for efficient cooling vs. noise generation. Just like on the Steam Deck, TDP will be user adjustable, so you will be probably be able to drive the fans at max speed and the SoC at 17W, but that will nonetheless quickly lead to thermal throttling. And they state that a lot of components like the Wifi chips and displays are thermal sensitive, so their performance will suffer when driving the Frame outside of the recommended TDP. Not to mention that wearing a hot HMD would suck.

Just like on the Steam Deck, the user adjustable TDP will be most useful for lowering power consumption for specific apps to allow for longer battery run times. And Valve has invested a lot of effort to run the SD8 Gen 3 at the highest performance that makes sense for different games while keeping a lot of other factors like battery and comfort in mind. Not the highest performance that it could technically achieve (with a water cooler). If you didn't get that from the Gamers Nexus video, I'd suggest watching it again.

Nobody said it would be hard to cool down 17W, that was never the question. What you claimed (before you edited your post) was that you were sure that Valve would drive the Frame at max performance all the time (17W TDP for the SoC plus more for RAM, displays, etc.) And that is just nonsense, and everything I wrote was regarding this being nonsense, both regarding technical feasibility with a phone SoC and the impact on battery life on a mobile HMD.

And of course you don't need watercooling for a 17W SoC, but if you listened to the DF video, you'd have learned that they build one anyway. The reason was not to run the chip at 17W, but to observe thermal creep throughout the system when running all the cores at max load (like you suggested) while preventing the SoC from throttling down, thus changing the thermal profile. Basically driving the SoC beyond what would regularly be possible due to it throttling once it runs too hot, which would prevent to exactly locate the source/path of the heat from the chip segments to the surrounding system, as some parts would constantly clock up or down. This is how they found where most of the heat went and came up with the second heat pipe solution on the other PCB side.