Intel Arc Pro B60 Creator 24GB: Workstation GPU for 3D, Video, and AI
Intel's Arc Pro B60 Creator 24GB is the company's most capable workstation GPU built on the Battlemage (Xe2) architecture. Positioned below NVIDIA's RTX 6000 Ada and AMD's Radeon Pro W7900 in price while offering 24GB of GDDR6 memory on a 384-bit bus, the B60 targets content creators, 3D artists, and AI researchers who need a large frame buffer without the premium pricing of top-tier professional cards.
This guide synthesizes publicly available specifications, third-party benchmark coverage, and community performance data to evaluate where the Arc Pro B60 fits — and where it falls short.
For an AI inference-focused look at the same hardware, see the companion guide: Intel Arc Pro B60 24GB: AI Workstation GPU Guide 2025.
Arc Pro B60 Specs at a Glance
| Specification | Intel Arc Pro B60 Creator |
|---|---|
| Architecture | Xe2 (Battlemage) |
| VRAM | 24GB GDDR6 |
| Memory Bus | 384-bit |
| Estimated TDP | ~160W |
| PCIe Interface | PCIe 5.0 x16 |
| Display Outputs | 4× DisplayPort 2.1 UHBR10 |
| Target Street Price | ~$699 |
Per Intel's official product brief, the 384-bit GDDR6 configuration delivers meaningfully higher peak memory bandwidth than competitors at comparable price points that use narrower bus widths. The PCIe 5.0 interface future-proofs the card in platforms that support the newer slot standard, while four DisplayPort 2.1 UHBR10 outputs accommodate high-refresh 4K multi-monitor workstation setups.
How Much VRAM Does the Arc Pro B60 Have — and Who Actually Needs 24GB?
The B60's 24GB frame buffer is the same capacity as the consumer RTX 4090 but delivered on a workstation-certified driver stack with professional ISV certifications.
In practice, 24GB GDDR6 comfortably covers:
- 8K video editing in Adobe Premiere Pro and DaVinci Resolve, where uncompressed RAW timelines can consume 8–12GB of GPU memory per cached frame set
- Large-scene 3D rendering in Blender Cycles and Cinema 4D, where geometry caches, HDR environment maps, and texture atlases compete for VRAM
- Local LLM inference at 13B–34B parameter scales under quantization (see Ollama on Intel + AMD via IPEX-LLM: Docker Compose Guide for container setup)
- High-resolution Stable Diffusion pipelines — SDXL with ControlNet and multiple LoRA adapters loaded simultaneously can exceed 12GB under batch generation at 1024×1024 or higher (Intel Arc Stable Diffusion WebUI: Setup & Performance 2026)
The ceiling the B60 runs into is 16K compositing and offline VFX scenes with very high polygon counts that approach or exceed 24GB. For those workflows, AMD's Radeon Pro W7900 (32GB) has a clear capacity advantage.
Arc Pro B60 vs. AMD Radeon Pro W7900 and W7800: VRAM, Bus Width, and Power
| GPU | VRAM | Bus Width | TDP | Est. Street Price |
|---|---|---|---|---|
| Intel Arc Pro B60 Creator | 24GB GDDR6 | 384-bit | ~160W | ~$699 |
| AMD Radeon Pro W7800 | 32GB GDDR6 | 256-bit | 260W | ~$999 |
| AMD Radeon Pro W7900 | 32GB GDDR6 | 256-bit | 300W | ~$3,499 |
| NVIDIA RTX 6000 Ada | 48GB GDDR6 ECC | 384-bit | 300W | ~$6,800 |
Per specifications published by Intel, AMD, and NVIDIA respectively, the B60's 384-bit bus is wider than both the W7800 and W7900's 256-bit configuration, which translates to higher peak memory bandwidth in sustained compute workloads despite the B60 carrying 8GB less capacity than the W7800. For tasks where bandwidth saturation — rather than raw capacity — is the bottleneck (GPU denoising, real-time grading, convolution-heavy AI passes), the wider bus can close or reverse the gap.
The power comparison is equally significant: at approximately 160W TDP, the B60 draws roughly half the sustained power of the W7900 (300W) and substantially less than the W7800 (260W). This affects PSU sizing, thermal budget, and ambient noise in sustained workflows.
For a direct AI inference comparison between Intel and AMD GPU stacks, see Intel IPEX-LLM vs Ollama on AMD: AI Rig Guide 2025.
Arc Pro B60 in 3D Rendering: Blender, Maya, and Cinema 4D
Blender Cycles
Blender's GPU rendering backend supports Arc GPUs via Intel's oneAPI Level-Zero compute path starting in Blender 4.2. Community benchmark threads on Blender Artists and third-party GPU benchmark compilations on TechPowerUp indicate that the Xe2 architecture's improved compute-per-tile efficiency over the original Alchemist (Xe) design is measurable in tile-based rendering workloads.
The B60's 24GB frame buffer is particularly relevant for scene caching: complex Blender scenes with high-polygon assets, volumetrics, and multi-layer HDRIs can exceed the 16GB available on previous-generation Arc cards or mid-range consumer cards, forcing tile-to-system-RAM fallbacks that significantly extend render times. The B60 avoids that bottleneck for most studio-scale production scenes.
For Stable Diffusion rendering on Arc specifically — a useful proxy for GPU throughput in image-generation pipelines — performance data across Arc generations is covered in Intel Arc Stable Diffusion Benchmark: A770, B580 & B60 (2025).
Autodesk Maya and Cinema 4D
The Arc Pro B60 ships with DirectX 12 Ultimate and Vulkan 1.3 support, which covers the viewport rendering paths used by Maya and Cinema 4D for interactive preview. Professional driver programs typically include validation testing for Maya and 3ds Max viewport stability — a key differentiator between Pro-tier and consumer cards from any vendor. Prospective users should verify the current ISV certification matrix in Intel's driver release notes for their specific application version.
Video Editing Performance: DaVinci Resolve and Adobe Premiere Pro
DaVinci Resolve
Blackmagic Design's professional color-grading and editing platform added Intel Arc support in DaVinci Resolve 18 and expanded it in subsequent versions. Per Blackmagic's published GPU acceleration documentation, supported acceleration includes Noise Reduction (Temporal and Spatial), Fusion GPU rendering, and the color science transform pipeline.
The B60's 24GB frame buffer provides headroom for 8K uncompressed timelines and multi-layer Fusion compositions without dropping to CPU fallback — a practical bottleneck on 8GB–16GB cards in high-end post-production environments.
Adobe Premiere Pro
Premiere Pro's Mercury Playback Engine uses GPU acceleration for color effects, Warp Stabilizer, and certain codec decode paths. The B60's Xe2 media engine includes hardware AV1 encode and decode, which is directly relevant for YouTube and streaming delivery pipelines that target AV1 output. Per Adobe's published GPU acceleration requirements, Arc cards with current drivers are supported under the Mercury GPU Acceleration (CUDA/OpenCL) framework.
Workflow-specific gains in Premiere depend heavily on effect chain complexity. Puget Systems' published benchmark methodology notes that GPU acceleration benefits in Premiere are most pronounced in export-heavy and effects-saturated timelines. Users with Premiere-centric workflows should benchmark their specific project templates before drawing broad conclusions.
Power Efficiency and Thermal Design
At approximately 160W TDP, the Arc Pro B60 sits in a meaningfully lower power tier than the AMD and NVIDIA workstation cards it competes with:
- PSU requirements: A 650W–750W system PSU accommodates the B60 alongside a current-generation desktop CPU. The W7900 (300W) typically warrants 850W+ PSU sizing.
- Thermal load: Lower heat output simplifies airflow in mid-tower workstations and is relevant in thermally constrained rack-mounted or near-silent studio setups.
- Sustained workloads: For AI researchers and creators running overnight Stable Diffusion batch jobs or LLM fine-tuning tasks, the B60's lower thermal output reduces fan curve strain and ambient noise during multi-hour runs.
Per Intel's specification sheet, the B60 requires standard PCIe supplemental power connectors appropriate for its TDP tier. The card does not require the 16-pin (12VHPWR) connector used by higher-TDP cards, simplifying cable management in older cases.
Software Ecosystem and AI Framework Support
IPEX-LLM and Local Inference
Intel's IPEX-LLM framework, maintained on GitHub and updated through 2025, enables quantized LLM inference on Arc GPUs using oneAPI as the compute backend. The B60's 24GB frame buffer supports:
- 7B parameter models at full BF16 precision
- 13B parameter models at 8-bit (INT8) quantization
- 34B parameter models at 4-bit (GGUF/AWQ) quantization
For containerized deployment using Docker Compose, the Ollama IPEX-LLM guide covers the full setup pattern.
Stable Diffusion
The B60 is supported by both Intel's IPEX backend for Automatic1111/ComfyUI and the DirectML backend. Community benchmark data for the B60 in Stable Diffusion pipelines — including SDXL and ControlNet configurations — is compiled in Intel Arc B60 Stable Diffusion: Setup & Benchmarks 2025. For a cross-GPU comparison including the Arc B50, Intel Arc B50 for Stable Diffusion vs RTX 3060 12GB: Which Wins in 2026 provides useful baseline context.
Building a Workstation Around the Arc Pro B60
The B60's 24GB GPU frame buffer benefits from a well-matched system. For 8K video workflows, system RAM above 64GB prevents OS-level memory pressure when the GPU pulls large asset caches into host memory alongside running OS processes and application caches.
For storage, a fast NVMe SSD handles active project scratch space, while a SATA SSD tier handles media archive and delivery exports. Options like the Samsung 870 EVO 1TB SATA SSD ($430.00) or the Samsung 870 EVO 500GB ($274.95) provide reliable sequential throughput for media delivery without bottlenecking on archive reads.
For lighter creative workstations or systems where RAM slots are being expanded incrementally, the Crucial 16GB DDR4 3200MHz Kit (2×8GB) ($127.00) is a cost-effective system memory entry point, though demanding Blender or Resolve sessions warrant 64GB+ total installed RAM.
Workload Fit Matrix
| Use Case | Arc Pro B60 Fit |
|---|---|
| 4K/8K video editing (Resolve, Premiere) | Strong — 24GB covers most professional timelines |
| Blender Cycles rendering (mid-complex scenes) | Good — full scene cache avoids RAM fallback |
| Stable Diffusion / AI art pipelines | Strong — see B60 Stable Diffusion Benchmarks |
| Local LLM inference (7B–34B quantized) | Good with IPEX-LLM framework |
| 16K+ compositing or offline VFX | Limited — W7900 (32GB) more appropriate |
| Budget-conscious creator workstation | Strong — lower price than W7800 and W7900 |
| Scientific compute requiring ECC | Not suitable — no ECC support |
Frequently Asked Questions
Does the Intel Arc Pro B60 support ECC memory? Per Intel's official product documentation, the Arc Pro B60 does not include ECC memory — unlike the NVIDIA RTX 6000 Ada (48GB GDDR6 ECC). For creative production workloads (video editing, 3D rendering, AI art generation), non-ECC GDDR6 is the industry norm and is not a practical limitation in those pipelines.
How does 24GB GDDR6 on a 384-bit bus compare to 32GB GDDR6 on a 256-bit bus? A wider memory bus delivers higher peak bandwidth even with less total capacity. For bandwidth-saturating workloads — GPU denoising in Blender, real-time 4K grading, convolution passes in AI pipelines — the B60's 384-bit configuration can maintain competitive throughput against the W7800's 256-bit 32GB layout. Applications requiring the full memory pool (16K compositing, very-high-poly offline VFX) benefit from the W7900's larger 32GB ceiling.
Is the Arc Pro B60 suitable for running local LLMs? Yes, with appropriate software. Intel's IPEX-LLM framework enables quantized inference on Arc GPUs. The 24GB frame buffer accommodates 13B models at INT8 precision or 34B models at 4-bit quantization. The Ollama IPEX-LLM Docker Compose guide covers containerized deployment.
What PSU wattage does the Arc Pro B60 require? Intel specifies approximately 160W TDP. A 650W–750W PSU is generally sufficient alongside a current-generation desktop CPU — roughly half the PSU headroom required for a 300W workstation card like the Radeon Pro W7900.
Does the Arc Pro B60 work with DaVinci Resolve? Yes. Intel's professional driver program includes ISV certification for DaVinci Resolve, enabling GPU-accelerated noise reduction, Fusion compositing, and color science transforms. Users should review current driver release notes for any version-specific caveats before deployment.
How does the B60 differ from the Arc A770 for creator workloads? The B60 uses Intel's second-generation Xe2 (Battlemage) architecture, which improves compute efficiency and tensor throughput over the A770's Xe (Alchemist) foundation. The B60 also carries substantially more VRAM (24GB vs. 16GB) and a wider 384-bit memory bus, making it considerably more capable for memory-constrained 3D scenes and AI inference. For a historical benchmark comparison across Arc generations, see Intel Arc Stable Diffusion Guide: Setup & Benchmarks 2025.
Citations and sources
- https://www.intel.com/content/www/us/en/products/sku/graphics/arc-pro-b60-graphics.html — Intel Arc Pro B60 official product page (architecture, specifications, TDP, outputs)
- https://www.techpowerup.com/gpu-specs/intel-arc-pro-b60.c4234 — TechPowerUp GPU database (specifications, memory bandwidth)
- https://www.tomshardware.com/best-picks/best-professional-gpus — Tom's Hardware professional GPU coverage and competitive analysis
- https://www.pugetsystems.com/labs/articles/ — Puget Systems Premiere Pro and DaVinci Resolve GPU benchmark methodology and reports
- https://www.amd.com/en/products/professional-graphics/radeon-pro-w7800 — AMD Radeon Pro W7800 official specifications (32GB GDDR6, 256-bit, 260W)
- https://www.amd.com/en/products/professional-graphics/radeon-pro-w7900 — AMD Radeon Pro W7900 official specifications (32GB GDDR6, 256-bit, 300W)
- https://www.blackmagicdesign.com/products/davinciresolve/studio — DaVinci Resolve GPU acceleration documentation and ISV support matrix
- https://github.com/intel/ipex-llm — Intel IPEX-LLM framework: quantized LLM inference on Arc GPUs
- https://www.blenderartists.org — Blender Artists community GPU benchmark threads and scene compatibility reports
This piece is editorial synthesis based on publicly available information. No independent first-party benchmarking is reported.
