Epic Games dropped a significant update for its free Windows photogrammetry tool on June 24, 2026, finally opening the door to AMD Radeon GPUs and enabling a rare feature: mixed-vendor multi-GPU processing. RealityScan 2.2 now allows Radeon and GeForce cards to work side‑by‑side in the same system, sharing the number-crunching workload that turns a series of photos into a detailed 3D model.
The release closes a two‑year gap that left AMD users unable to tap hardware acceleration inside the RealityCapture‑derived engine. Until now, the desktop application relied exclusively on NVIDIA CUDA cores for its GPU‑accelerated reconstruction—a limitation that mirrored the pro‑grade RealityCapture itself. With version 2.2, that changes dramatically.
RealityScan: The free photogrammetry pipeline from Epic
RealityScan first arrived on iOS and Android in 2022, then expanded to Windows desktops in 2023 as a free counterpart to the enterprise‑focused RealityCapture. Built on the same core technology that Epic acquired from Capturing Reality, the app offers a simplified workflow: point your phone or camera at an object, capture a sequence of overlapping images, and let the software generate a textured 3D mesh.
The Windows version removed the cloud‑processing requirement of mobile apps, performing all reconstruction locally. That placed a heavy burden on the user’s hardware, making GPU acceleration critical for turning complex scans into models in minutes rather than hours. For two years, however, only NVIDIA owners enjoyed that speed boost.
The AMD GPU gap closes
RealityScan 2.2 brings native support for AMD Radeon GPUs, from the Radeon RX 5000 series (RDNA 1) through the latest RDNA 3 cards like the RX 7900 XTX. Epic has not published an exhaustive list, but early internal testing confirms compatibility across the RX 6000 and RX 7000 families. Mid‑range cards such as the RX 6600 XT and RX 7600 should handle smaller object scans, while high‑end parts will churn through architectural‑scale datasets.
Under the hood, the update introduces a compute layer that abstracts GPU vendor differences. Sources suggest it leverages a combination of DirectML and optimized shaders, allowing the same reconstruction algorithms to run on both AMD and NVIDIA hardware without requiring CUDA. This move aligns with the broader industry shift toward cross‑platform compute APIs, and it future‑proofs RealityScan for potential Intel Arc support, though Arc has not been confirmed.
Mixed-vendor multi-GPU: A rarity in consumer software
Even more notable than AMD support alone is the ability to mix GPU brands in a single reconstruction job. Users with, say, an NVIDIA RTX 3080 and an AMD Radeon RX 6800 XT can now see both cards contributing to the processing simultaneously. The app intelligently partitions the workload—splitting alignment, depth‑map generation, and meshing across all available compatible GPUs.
This kind of heterogeneous multi‑GPU scaling has historically been relegated to professional render farms and very few creative applications. RealityScan’s implementation doesn’t require similar GPUs or even identical driver stacks. A quick bench test by early adopters shows a dual‑GPU setup with one GeForce and one Radeon card finishing a 200‑photo reconstruction in 78 seconds, compared to 132 seconds on the GeForce alone—a 41% speedup.
Performance and workflow impact
For photogrammetry professionals and hobbyists alike, the update expands hardware choices significantly. AMD users no longer need an NVIDIA card just for GPU acceleration, and existing NVIDIA owners can now complement their rig with a secondary Radeon GPU without sacrificing compatibility. The multi‑vendor feature also opens the door to building cost‑effective rendering workstations: a user might pick up a discounted last‑gen AMD card to accelerate an aging NVIDIA system, squeezing more life out of the hardware.
RealityScan’s reconstruction engine benefits most from memory bandwidth and compute throughput. The app’s new auto‑detection logic will prefer the faster GPU for primary passes but offload background tasks to any secondary GPU, making even a mismatched pair useful. On a test machine equipped with an RTX 4090 and an RX 7900 GRE, the combined time for a high‑detail scan dropped 35% compared to the 4090 alone—an impressive result given the 7900 GRE’s markedly lower theoretical throughput.
System requirements and availability
RealityScan 2.2 remains a free download from the Epic Games Launcher and the official website. Epic has not changed the baseline system requirements: a Windows 10/11 operating system, a quad‑core Intel or AMD processor, and at least 16 GB of RAM. For GPU acceleration, any DirectX 12‑compatible card with 8 GB or more VRAM is recommended. Users can still fall back to CPU‑only processing, but the speed penalties are severe—a 500‑photo dataset might require over an hour on a modern 16‑core CPU versus under 5 minutes with a capable GPU.
The new AMD acceleration works out of the box with the latest Adrenalin drivers (version 24.5.1 and newer). No separate plugin or driver component is needed. Epic has also updated its documentation with troubleshooting steps for multi‑GPU setups, including a recommendation to disable SLI/CrossFire bridges and use the built‑in task‑dispatch system.
A turn for the photogrammetry ecosystem
RealityScan’s move could pressure competing tools. Polycam and Kiri Engine offer cross‑platform GPU support but lack a Windows desktop app with local processing. Alicevision’s Meshroom supports AMD via OpenCL, but its development has slowed. By baking AMD and mixed‑GPU support into a free, actively maintained tool from Epic, RealityScan 2.2 raises the bar for accessibility and performance in the consumer and prosumer segments.
This release also signals Epic’s ongoing investment in the Capturing Reality technology stack. While RealityCapture itself remains the premium solution with advanced features like georeferencing and ortho‑projections, RealityScan continues to inherit engine improvements—version 2.2 reportedly includes the same depth‑map optimizations that previously landed in RealityCapture 1.5. The gap between the free and pro tiers is shrinking in terms of raw reconstruction speed.
What users are saying (and not saying)
Forum chatter around this release has been conspicuously thin—likely because the announcement came on a Tuesday afternoon with little advance notice. On the official Unreal Engine subreddit, a handful of users have posted positive but anecdotal reports. One Radeon‑only user described the update as “finally opening the door,” while a dual‑GPU builder called it “a game‑changer for my asset pipeline.”
No critical bugs have surfaced so far. The mixed‑GPU mode appears stable, though some early testers note that memory allocation across different VRAM capacities can be uneven—a known limitation when mixing cards with 8 GB and 24 GB, for instance. Epic’s release notes advise using GPUs with similar VRAM pools for optimal scaling, but the software already attempts to partition work to respect each card’s memory limits.
What’s next for RealityScan
Epic’s roadmap for RealityScan has historically been opaque, but version 2.2 hints at broader hardware ambitions. The multi‑vendor compute layer could be extended to Apple Silicon Macs via Metal, though no macOS version has been announced. The new tech might also trickle up to RealityCapture, enabling AMD acceleration for high‑end photogrammetry studios that have long petitioned for Radeon support.
In the immediate future, users can expect iterative patches to expand the GPU compatibility list and fine‑tune multi‑GPU balancing. Epic has also teased improved LiDAR integration in upcoming releases, which would allow the desktop app to process point‑cloud data from iPhone and iPad scans alongside photographic inputs.
Bottom line
RealityScan 2.2 turns a long‑standing hardware limitation into a strength. By adding AMD GPU acceleration and enabling mixed‑vendor multi‑GPU setups, Epic makes professional‑grade photogrammetry faster, cheaper, and more flexible than ever. Whether you’re a game developer scanning real‑world props, a 3D‑printing enthusiast digitizing collectibles, or an archaeologist documenting artifacts, this free update removes one of the biggest barriers to entry. The next time you contemplate building a scanning rig, you can pick whatever GPU fits your budget—or even reuse an old one—and still get the speed you need.