Steam Deck 2's Gainsborough APU: 16 RDNA 5 CUs Explained

Four and a half years after the original Steam Deck launched on TSMC’s 7nm node, we finally have a plausible silicon target for its successor. Moore’s Law is Dead is reporting that an AMD APU codenamed Gainsborough — the surname of Final Fantasy 7’s Aerith, which was itself the codename for the original Deck’s chip — is the leading candidate for a true second-generation Steam Deck. The headline detail isn’t the GPU configuration. It’s the process: TSMC’s N3P.

Treat this as a rumour, because that’s exactly what it is. MLID has a decent hit rate on AMD silicon and a long tail of misses, and Valve hasn’t said a word. But the shape of the leak is worth taking seriously, because it lines up almost suspiciously well with what Valve has publicly demanded from a Deck successor.

Why N3P Matters More Than the CU Count on a 15W Handheld

The original Aerith chip — and the mildly revised Sephiroth silicon in the OLED Deck — is built on TSMC’s 7nm. Gainsborough, if the leak holds, jumps to N3P, the most refined member of TSMC’s 3nm family. That’s two full node generations of density and efficiency gain, and on a handheld locked into a fixed power envelope those gains don’t just buy frames. You can spend them on battery, on acoustics, or on a bigger GPU at the same wattage. Every portable PC in this category is fighting the same fight inside a 15W budget, and a node shrink is the only lever that moves the whole curve at once.

Valve’s own framing makes the rest of the leak legible. Steam Deck designer Pierre-Loup Griffais has said publicly that Valve isn’t interested in “20 or 30 or even 50% more performance at the same battery life” — it wants a demarcated generational leap. Gainsborough’s rumoured spec sheet is aimed straight at that bar: Zen 6 CPU cores, RDNA 5 graphics, and potentially 16 compute units where Aerith and Sephiroth packed eight RDNA 2-spec CUs. Because RDNA 5 CUs should be substantially more capable per unit than RDNA 2’s, you’re stacking a doubling of the count on top of per-CU generational gains, plus whatever clock headroom N3P unlocks. That is the step change Griffais described — on paper.

Late 2028, and the Memory Market Is the Real Gatekeeper

MLID puts Gainsborough’s own readiness at late 2027 to early 2028, which realistically pushes a shipping Steam Deck 2 into late 2028 — roughly six years after the original. The delay isn’t a silicon problem, it’s an economics problem. The AI-fuelled memory crunch has already disrupted Valve’s other recent hardware plans, and a handheld is unusually exposed: LPDDR is not a rounding error in a $500 portable, and the wider memory bus that next-gen handheld APUs are reportedly targeting only raises the bill. If a Deck 2 and Sony’s rumoured PS6 handheld are drawing from the same memory supply, pricing pain is the likeliest outcome for both.

None of that dims what Gainsborough would mean competitively. The ROG Ally X, MSI Claw 8 and Legion Go have all tried to out-spec the Deck and lost the argument anyway, because Valve’s edge was never raw silicon — it’s SteamOS, a decade of Proton tuning, and suspend-resume that actually works. N3P arriving in 2028 rather than on TSMC’s newer N2 isn’t a compromise; for a volume handheld it’s the cost-conscious, high-yield choice. Which leaves a practical read for anyone buying today: nothing in this leak pulls a Deck 2 into 2027, so buy the handheld that plays your library now. The next one will be worth waiting for precisely because Valve is willing to.