AMD RX 9070 XT, 새로운 DirectX 기술로 프레임률 50배 이상 향상 및 레이트레이싱 메모리 사용량 90% 감소
🇰🇷 한글 번역
(이 부분 번역 실패)
AMD GPU 사용자에게는 또 다른 제한 사항으로 하드웨어 지원이 꼽힌다. 현재 미리 보기 드라이버는 26.20.12.01 버전을 통해 RDNA 4 GPU,其中包括 RX 9000 시리즈를 지원한다. 오스발도는 AMD의 구현이 BVH8 및 방향성 경계 상자(Oriented Bounding Boxes)와 같은 아키텍처 개선 사항에 크게 의존하고 있어, 이전 세대 Radeon GPU의 지원은 어려울 것이라고 믿는다.
그러나 마이크로소프트는 표준화된 API가 근본적으로 새로운 레이 트레이싱 하드웨어를 필요로 하지 않는다고 밝히며, 광범위한 호환성은 여전히 벤더의 구현과 드라이버 지원에 달려 있다고 밝혔다.
결국 오스발도의 발견은 마이크로소프트의 DXR 개선 사항, 특히 이전에 비교 가능한 공개 최적화 경로가 없었던 AMD GPU에게 지닌 막대한 잠재력을 강화한다. 개발자들이 이러한 극적인 합성 성능 향상을 의미 있는 게임 성능 개선으로 전환할 수 있을지는 아직 지켜봐야 할 문제다.
뉴스 출처: 오스발도 피날리 도데를레인의 X(트위터) 스레드
더 많은 뉴스 커버리지를 피드에서 받아보려면 Wccftech를 Google에서 팔로우하세요.
📄 원문 (English)
Microsoft's latest DirectX Raytracing (DXR) upgrade appears to be delivering on its remarkable performance claims, with independent testing on an AMD Radeon RX 9070 XT revealing massive improvements in acceleration structure efficiency and memory consumption.
As we reported last week, Microsoft's DirectX 12 Agility SDK 1.722 Preview introduces Clustered Geometry, Partitioned Top-Level Acceleration Structures (PTLAS), and GPU-driven acceleration structure builds. Together, these features could enable more efficient ray tracing of highly detailed environments, similar to NVIDIA's proprietary RTX Mega Geometry technology.
Following AMD's November 3 announcement confirming support for the new features on RDNA 4 GPUs, graphics enthusiast Osvaldo Pinali Doederlein decided to put Microsoft's claims to the test, with some fascinating results.
AMD's RX 9070 XT Goes From 1.7 FPS To 91 FPS With Clustered Geometry
Using Microsoft's publicly available D3D12RaytracingClusteredGeometry sample and AMD's dedicated Agility SDK Developer Preview driver, Osvaldo compared conventional Bottom-Level Acceleration Structure (BLAS) rebuilding with the new Clustered Geometry implementation.
His testing was conducted on an RX 9070 XT at 1600 × 920, using a fixed camera position and a geometry-heavy configuration intended to approximate Microsoft's original demonstration.
The results were extraordinary:
That's approximately 54 times higher frame rates and nearly 90% lower acceleration structure memory consumption, broadly corroborating Microsoft's earlier demonstration, which achieved 1.2 FPS versus 79 FPS on an RX 9070.
Crucially, the improvements primarily come from drastically reducing the overhead of rebuilding ray tracing acceleration structures rather than making ray traversal itself substantially faster.
Actual Gaming Performance Gains Will Be Far More Modest
Osvaldo also investigated several important limitations. With the demonstration's default geometry configuration of 100 additional BLASes, Clustered Geometry was approximately twice as fast. Increasing that figure to 10,000, effectively multiplying the additional geometry by 100, produced the spectacular 50X-plus gains.
Rendering resolution also matters. At 4K, the geometry-heavy configuration improved from 1.6 FPS to 28 FPS, a still impressive but substantially smaller 17.5X improvement. Furthermore, Osvaldo observed that ray dispatch performance could occasionally regress by as much as 40%, even though Clustered Geometry remained faster overall in his tests.
The demanding configuration featured approximately 91 million triangles. For context, Osvaldo pointed to Remedy Entertainment's Alan Wake II, whose most complex scenes can reportedly contain around 200 million triangles before culling, although the workloads are not directly comparable.
He also emphasized that these results do not mean an RX 9070 XT will suddenly run demanding path-traced games at 90 FPS. Real games involve many additional rendering workloads, and integrating Clustered Geometry requires substantial engine-level changes.
Indeed, our previous Gears of War: E-Day performance analysis demonstrated that NVIDIA's RTX Mega Geometry implementation could carry considerable performance and VRAM costs despite improving geometric detail.
For AMD GPU users, another limitation is hardware support. The preview driver currently supports RDNA 4 GPUs, including the RX 9000 Series, through driver 26.20.12.01. Osvaldo believes AMD's implementation relies heavily on architectural improvements such as BVH8 and Oriented Bounding Boxes, making support for older Radeon generations unlikely.
However, Microsoft states that the standardized APIs do not fundamentally require new ray tracing hardware, meaning broader compatibility remains dependent on vendor implementations and driver support.
Ultimately, Osvaldo's findings reinforce the enormous potential of Microsoft's DXR improvements, particularly for AMD GPUs that previously lacked a comparable publicly available optimization path. Whether developers can translate these dramatic synthetic improvements into meaningful gaming performance gains remains to be seen.
News Source: Osvaldo Pinali Doederlein's X thread
Follow Wccftech on Google to get more of our news coverage in your feeds.
▶ 칩 라이브 24 홈으로 — chiplive24.com