Apple M5 Ultra Mac Studio outperforms Nvidia RTX 5090 in specific 3D rendering benchmarks

The Apple M5 Ultra Mac Studio, featuring an 80-core GPU and 96GB of unified memory, demonstrates superior performance in specific 3D rendering tasks compared to the Nvidia RTX 5090 desktop GPU. In the Blender 3D rendering "Classroom" project, the M5 Ultra completed the task in 5.10 seconds, slightly faster than the 5.67 seconds recorded by the RTX 5090. Similarly, in the "Lone Monk" project, the M5 Ultra finished in 35 seconds, significantly faster than the 1 minute and 26 seconds required by the RTX 5090. The M5 Ultra's performance advantage is attributed to its high-capacity unified memory architecture, which allows for more efficient data handling between the CPU and GPU compared to the discrete VRAM architecture of the RTX 5090. Additionally, the M5 Ultra shows significant performance gains in AI-related tasks, such as text-to-image generation and large language model inference, where its unified memory capacity provides a substantial throughput advantage over systems with less VRAM.

The Apple M5 Ultra Mac Studio, featuring an 80-core GPU and 96GB of unified memory, demonstrates superior performance in specific 3D rendering tasks compared to the Nvidia RTX 5090 desktop GPU. In the Blender 3D rendering "Classroom" project, the M5 Ultra completed the task in 5.10 seconds, slightly faster than the 5.67 seconds recorded by the RTX 5090. Similarly, in the "Lone Monk" project, the M5 Ultra finished in 35 seconds, significantly faster than the 1 minute and 26 seconds required by the RTX 5090. The M5 Ultra's performance advantage is attributed to its high-capacity unified memory architecture, which allows for more efficient data handling between the CPU and GPU compared to the discrete VRAM architecture of the RTX 5090. Additionally, the M5 Ultra shows significant performance gains in AI-related tasks, such as text-to-image generation and large language model inference, where its unified memory capacity provides a substantial throughput advantage over systems with less VRAM.

The M5 Ultra Mac Studio completed the Blender "Classroom" rendering project in 5.10 seconds, outperforming the RTX 5090's 5.67 seconds. In the more complex "Lone Monk" Blender project, the M5 Ultra finished in 35 seconds, compared to 1 minute and 26 seconds for the RTX 5090.

The M5 Ultra's unified memory architecture allows for more efficient data processing between the CPU and GPU than discrete VRAM setups. The M5 Ultra achieved 158 tokens per second in Ollama Qwen3.5:35b-a3b inference, significantly outpacing the M5 Max.

The M5 Ultra's 96GB of unified memory provides a distinct advantage for large-scale AI model inference tasks. The M5 Ultra Mac Studio is priced lower than a standalone RTX 5090 GPU, according to the video's retail comparisons.

Chapter guide

Worth noting

  • The video contains a promotional segment for Merlin AI, which is a paid sponsorship.
  • The comparison between the M5 Ultra and the Nvidia RTX 5090 is based on specific, limited benchmark tests and may not reflect performance across all real-world applications.
  • The video references unreleased or speculative hardware and software versions, such as 'M6' and 'macOS 27', which are not currently available.

Watch the original video ↗