I've tested AMD's new AI frame generation to see if it's worth the wait

AMD has shipped its long-awaited FSR Redstone suite in the Adrenalin 25.12.1 driver, bringing ML-powered upscaling and frame generation to RDNA 4 GPUs (supported in ~31 games) and rolling out Ray Regeneration now with a developer-facing Radiance Cache coming next year. In testing the ML upscaler and frame generator deliver clear image-quality gains over FSR3 with only modest performance costs (the upscaler accounts for most of the hit and frame gen is largely low-cost), while results vary by title—stronger in Black Myth, Cyberpunk and F1 25, mixed in GTA V—and AMD still lacks multi-frame generation and parity with Nvidia/Intel multi-frame solutions. The release closes some of the feature gap with competitors but is late to market, hampered by confusing naming and driver-level workarounds for enabling features; Radiance Cache could materially improve ray-tracing efficiency if developers adopt it, but its workflow and performance benefits remain unproven.

Analysis

AMD has shipped FSR Redstone in the Adrenalin 25.12.1 driver, bringing ML-powered upscaling and an AI frame generator to RDNA 4 GPUs and enabling FSR Ray Regeneration now, with a developer-facing FSR Radiance Cache slated to appear in engines next year; Redstone currently works on roughly 31 games and requires driver-level toggles to activate.

Independent testing on an Asus Radeon RX 9070 XT with a Ryzen 9 testbed shows the ML upscaler delivers clear image-quality gains versus FSR3 while accounting for most of the performance hit, and the AI frame generator adds little additional overhead; title-level outcomes vary — Black Myth, Cyberpunk 2077 and F1 25 show notable visual and motion improvements with small perf costs, whereas GTA V at 4K High RT exhibited material strain (reported averages near ~30 fps and 1% lows near ~25 fps).

Strategically, Redstone narrows AMD’s feature gap with Nvidia and Intel but is late to market (Nvidia introduced AI frame gen in Sept 2022; Intel in Dec 2024) and still lacks multi-frame generation parity; the combination of confusing naming, driver-only workarounds and limited native game support makes developer adoption of Radiance Cache and broader Redstone integration the critical drivers of any sustained competitive or financial impact.

Key Decisions for Investors

• Maintain a cautiously constructive view on AMD: Redstone improves competitiveness but wait for broader native game support beyond the ~31 titles before increasing exposure
• Track developer adoption metrics and SDK uptake (GPUOpen releases, number of native Redstone integrations) and only update forecasts after measurable engine-level Radiance Cache implementation and performance data appear
• Account for competitive timing risk from Nvidia and Intel—use position sizing or hedges if AMD exposure is concentrated given AMD still lacks multi-frame parity
• For near-term trades, watch game-specific performance signals (e.g., GTA V’s ~30/25 fps outcome and F1 25 gains) and driver update cadence; consider holding core positions but avoid aggressive buys until adoption and multi-frame feature parity are proven