A real surprise – Intel Arc Pro B50 in the test and big workstation duel under 450 euros

1 - Introduction, unboxing and technical data 2 - Test system and equipment 3 - Teardown: PCB, topology and components 4 - Teardown: Cooling solution 5 - Teardown: Material analysis and ASTM TIM testing 6 - Autodesk AutoCAD 7 - Autodesk Inventor Pro 8 - PTC Creo 9 - Dassault Systèmes Solidworks 10 - Autodesk Maya 11 - SPECviewperf 15 (2025) 12 - Adobe Photoshop 26.10 13 - Adobe After Effects 2025 14 - Adobe Premiere Pro 25.41 15 - AI Benchmarks (AI Vision, Image, Text) 16 - Rendering 17 - Temperatues, clock rates, power draw and fan speed 18 - Summary and conclusion

Rendering with Blender 3.6

Blender is a powerful open source software for 3D graphics, animation, rendering, post-production, interactive creation and playback. It allows users to create, edit and render 3D models and animations. In addition to these 3D features, Blender also includes tools for video editing, sculpting, UV mapping, texturing, rigging, particle systems, physics and fluid simulations, and game development. Because of its extensive feature set and because it is free, Blender is used by both amateurs and professionals throughout the media and entertainment industry. However, I must first write something about CUDA, OptiX and HIP in order to better understand the comparisons. CUDA, OptiX and HIP are technologies that were developed specifically for the field of parallel and high-performance computing (HPC). Each of these technologies has its own advantages, target applications and underlying technologies.

Blender 3.6 LTS uses different compute backends in the Cycles renderer to efficiently address CPU and GPU resources. Which interface is used depends on the hardware used and the driver stacks. Whether CUDA, OptiX, HIP or OneAPI is used plays a central role for the performance measurement in the benchmark, as the range of functions and degree of optimization differ. The choice of compute backend significantly determines the results in the Blender benchmark. NVIDIA benefits from CUDA as a stable basis and OptiX as a high-performance option, AMD from HIP as the modern successor to OpenCL, while Intel offers complete GPU support in Blender for the first time with OneAPI. Differences in the degree of optimization explain why NVIDIA often leads in GPU rendering, AMD is in the midfield and Intel is currently heavily dependent on software updates to exploit its potential.

CUDA is NVIDIA’s proprietary API for GPGPU computing and has been the standard for CUDA or shader-based workloads in Blender for many years. CUDA is offered for RTX and GTX GPUs in the Cycles renderer. CUDA is stable, supports practically all cards from the Kepler generation onwards and delivers predictable results. In practice, CUDA is the basic option when OptiX is not used or older cards are in use. OptiX is NVIDIA’s ray tracing API that directly targets RTX cards with ray tracing cores in Blender. OptiX enables accelerated path tracing through hardware RT cores, which brings enormous advantages in scenes with complex light and shadow calculations. Compared to CUDA, OptiX offers significant performance gains in raytracing-heavy scenes, which is why RTX cards usually benefit greatly in the Blender benchmark.

HIP is AMD’s API for GPGPU computing, which is available in Blender as a backend for cycles. HIP replaced OpenCL, which is no longer maintained in Blender. As of RDNA2 GPUs, HIP is supported, allowing Radeon cards to remain competitive in GPU rendering. In practice, the HIP implementation in Blender still lags somewhat behind the maturity of CUDA and OptiX, but progress is visible from version to version. For the benchmark, this means that AMD cards show good raw performance in the HIP backend, but often do not come close to NVIDIA’s optimizations.

OneAPI is Intel’s open standard for heterogeneous computing, which is also supported in Blender Cycles. Intel GPUs such as the Arc series use OneAPI as the primary backend. This allows the Xe cores to be used for path tracing. The performance is still heavily dependent on the driver and the implementation in Blender, but is developing rapidly. For the Intel Arc Pro B50 in the Blender 3.6 benchmark, OneAPI is the decisive factor that enables the cards to perform GPU rendering at all and ensures that Intel delivers competitive values in the SPECviewperf and Cycles tests.

The NVIDIA cards use both CUDA and OptiX, the AMD cards use HIP and Intel uses OneAPI.

Conclusion

I have deliberately limited myself to a single benchmark for rendering, as this shows the limits of the tested cards. All three models belong to the entry-level class of professional workstation GPUs, which means that the pure raw performance is not sufficient for complex rendering scenes. Even more serious, however, is the limited memory: both the NVIDIA RTX A1000 and the AMD Radeon Pro W7500 only have 8 GB of VRAM. This memory reaches its limits at the latest with more complex meshes or high-resolution textures, which either leads to greatly reduced performance, rendering errors or even crashes.

The selected benchmark illustrates this problem very well, as it depicts typical rendering scenarios with extensive geometries and textures. The results are therefore representative of the entire class of these cards: the performance is sufficient for simple tasks, but they quickly reach their technical limits in professional workflows with higher model and texture complexity.