Valve Steam Deck 2: Reliable leaker alleges iGPU woefully inadequate for AAA games

[Redaktion]

Moore's Law Is Dead recently revealed some details regarding the AMD APU that could potentially power the Valve Steam Deck 2. Reliable leaker Kepler_L2 has now shared a key detail about the Steam Deck 2's proposed AMD APU that could disappoint many fans who were hoping for something much faster.

https://www.notebookcheck.net/Valve-Steam-Deck-2-Reliable-leaker-alleges-iGPU-woefully-inadequate-for-AAA-games.1414160.0.html

[QwertyChouskie]

"woefully inadequate" running at 1080p on Windows is very different than "woefully inadequate" running at 800p on SteamOS/Linux.  12 CUs vs 8 is 50% more, combined with the newer architecture brings probably a 2x or greater performance leap over the Deck 1, and even more when using FSR (FSR4 isn't even officially running on RDNA2 yet, and even when it will, it won't be as good as on RDNA 3.5/4).

[+2x perf either way]

Even the current crop of Intel-based handheld gaming consoles like the MSI Claw 8 EX AI+ with the Intel Arc G3 Extreme will likely be much more powerful than the Steam Deck 2.

True, Intel's Xe3 Panther Lake iGPU cores are more power efficient (but Intel is also using TSMC's next gen 3N node vs AMD using 5N) and so a Steam Deck 2 based on them would be faster. But it doesn't mean that AMD couldn't do it. FSR4 / upscaling would be especially helpful in handhelds.

So, only time will tell what APU the Valve Steam Deck 2 actually employs and how powerful the handheld console ends up being.

When using only a full node shrink:
Power reduction at same speed: ~25–30%.
Speed gain at same power: ~10–15%.

AMD Ryzen Z1 APU is using TSMC's 4N node (slightly improved 5N -- basically 5N), and by 2028 TSMC 2N node will be available in the sense that it's also going to be cheap enough by then, then using 2N node, the raw performance would have improved by 2 nodes (5N -> 3N -> 2N).

Using a 2-node improvement: 1.15 * 1.15 (= 1.15^2) = +32% FPS per Watt improvement.

Using a 3-node improvement / including TSMC A14 node:
1.15^3= +52% FPS per Watt improvement.

[+2x perf either way]

QwertyChouskie, good points, but an increase from 8 to 12 CUs is not a +50% improvement in practice (the performance increase doesn't scale linearly with the number of CUs).

brings probably a 2x or greater performance leap over the Deck 1

Using hardware-based RDNA4 upscaling with FSR4 on top of the nodes improvement is going to to certainly improve the perf by 2x (2 = 1.52 * 1.3 (what upscaling at good quality delivers).