Intel unveils XeSS 3 with multi-frame generation for Arc GPUs

Intel introduced XeSS 3 with multi-frame generation up to 4x, bringing DLSS-like scaling to Arc graphics cards and the Xe3 iGPU in Panther Lake laptops. XeSS multi-frame generation will be usable for all Intel GPUs that already support framegen, with broader game availability coming via updates and an Intel Graphics override.

Intel announced XeSS 3 on October 9 at the Intel Tech Tour, adding multi-frame generation that can output 2x, 3x, or 4x on-screen frames. In Intel’s controlled demo on a Panther Lake laptop running Dying Light: The Beast and Painkiller, a 1080p scene rose from roughly 50 FPS native to about 200 FPS with 4x MFG enabled. Intel says players will toggle the higher MFG modes in supported games or through the Intel Graphics software.

Availability starts alongside Panther Lake, with Intel positioning XeSS 3 for Arc GPUs (and the new Xe3 integrated graphics card) while game updates roll out to implement the new XeSS version. As of today, Intel is characterizing MFG as an Arc-only feature, even though last summer’s XeSS 2.1 SDK brought cross-vendor frame generation and low-latency mode to compatible AMD and Nvidia GPUs.

Upscalers like Nvidia DLSS or AMD FSR render frames below the target resolution and then rebuild the frame at output resolution using motion vectors, depth, and a history buffer. The aim is to keep detail and stability while lowering the cost of shading. Frame generation is a separate tech stackable with upscaling. It inserts artificial frames between real ones to raise perceived smoothness, which can add latency and reveal artifacts like UI ghosting if motion data is noisy. Multi-frame generation inserts multiple frames.

XeSS follows this template with a temporal model. It runs most efficiently on Intel Arc using XMX matrix engines, and falls back to a DP4a compute path on other GPUs for broad compatibility. The SDK is public, and since XeSS 2.1, Intel has enabled cross-vendor frame generation and a low-latency mode on GPUs that meet the API requirements, while Arc continues to execute the neural nets on XMX for better throughput.

On September 18, 2025, Nvidia invested $5 billion in Intel (approximately a 4% stake) and announced a joint plan to co-develop chips for PCs and data centers, with Intel supplying custom x86 CPUs and advanced packaging to pair them with Nvidia GPUs.

Both firms emphasized that this is a product collaboration rather than a broader commitment, with Intel handling some packaging (Foveros/EMIB) while Nvidia continues to fab most GPUs at TSMC for now.