M5 Ultra vs 2 DGX Sparks… The Number You’re Not Looking At
Clip title: M5 Ultra vs 2 DGX Sparks… The Number You’re Not Looking At Author / channel: Alex Ziskind URL: https://www.youtube.com/watch?v=_yrw6c5gw3E
Summary
Certainly! Here is a summary of the video content:
The video discusses the performance comparison between the M5 Ultra Mac Studio and a cluster of two NVIDIA DGX Spark devices. The speaker tests both setups using various AI models, such as DeepSeek V4 Flash and Qwen3.8-Flash-Next, to evaluate their capabilities in tasks like token generation and prompt processing.
Key points discussed in the video include:
- Token Generation Speed: The M5 Ultra Mac Studio and the dual DGX Spark cluster showed similar token generation speeds when using DeepSeek V4 Flash, but the Mac outperformed the Sparks with Qwen3.8-Flash-Next.
- Prompt Processing Time: The dual DGX Spark cluster demonstrated faster prefill times for longer prompts compared to the M5 Ultra Mac Studio.
- Memory and Bandwidth: The M5 Ultra Mac Studio has 256GB of unified memory with high bandwidth, while the dual DGX Spark cluster combined offers 256GB of memory connected via a high-speed RoCE cable.
- Concurrency: The dual DGX Spark cluster scaled better with multiple concurrent users compared to the Mac Studio.
In conclusion, the video highlights that while the M5 Ultra Mac Studio is powerful and efficient for single-user workloads and specific models, the NVIDIA DGX Spark cluster excels in handling larger prompts and multiple concurrent users. The choice between the two depends on the specific use case, such as whether you are prioritizing single-user performance or multi-user scalability.
Video Description & Links
Description
It turns out the M5 Ultra needs something from you before it really pulls ahead.
🛒 Gear Links 🛒 ⌨️🏢 Cluster rack: 📦🎮 My gear: https://www.amazon.com/shop/alexziskind
🎥 Related Videos 🎥
🧳🧰 Mini PC portable setup - https://youtu.be/4RYmsrarOSw 🍎💻 Dev setup on Mac - https://youtu.be/KiKUN4i1SeU 💸🧠 Cheap mini runs a 70B LLM 🤯 - https://youtu.be/xyKEQjUzfAk 🧪🔥 RAM torture test on Mac - https://youtu.be/l3zIwPgan7M 🧠📉 REALITY vs Apple’s Memory Claims | vs RTX4090m - https://youtu.be/fdvzQAWXU7A 🧬🐍 Set up Conda - https://youtu.be/2Acht_5_HTo ⚡💥 Thunderbolt 5 BREAKS Apple’s Upcharge - https://youtu.be/nHqrvxcRc7o 🧠🚀 INSANE Machine Learning on Neural Engine - https://youtu.be/Y2FOUg_jo7k 🧱🖥️ Mac Mini Cluster - https://youtu.be/GBR6pHZ68Ho
- 🛠️ Developer productivity Playlist - https://www.youtube.com/playlist?list=PLPwbI_iIX3aQCRdFGM7j4TY_7STfv2aXX 🔗 AI for Coding Playlist: 📚 - https://www.youtube.com/playlist?list=PLPwbI_iIX3aSlUmRtYPfbQHt4n0YaX0qw
— — — — — — — — —
— — — — — — — — —
— — — — — — — — —
Tags
software developer, programmer, software development, programming, developer, developer tests, m3 chip, machine learning, llm, ml on m3, machine learning m3, m3 machine learning, openui, local ai, llama.cpp, nuc, m4, m4 pro, mac mini, apple, apple mini, mini, m4 mini, m4 pro mac mini, m4 max, mac studio, m3 ultra, m4 max mac studio, mac cluster, m5 ultra, m5ultra, mac ultra, m5 mac studio, m5 ultra mac, m5 ultra vs m3 ultra, dgx spark, dgx spark cluster
URLs
- https://www.amazon.com/shop/alexziskind
- https://youtu.be/4RYmsrarOSw
- https://youtu.be/KiKUN4i1SeU
- https://youtu.be/xyKEQjUzfAk
- https://youtu.be/l3zIwPgan7M
- https://youtu.be/fdvzQAWXU7A
- https://youtu.be/2Acht_5_HTo
- https://youtu.be/nHqrvxcRc7o
- https://youtu.be/Y2FOUg_jo7k
- https://youtu.be/GBR6pHZ68Ho
- https://www.youtube.com/playlist?list=PLPwbI_iIX3aQCRdFGM7j4TY_7STfv2aXX
- https://www.youtube.com/playlist?list=PLPwbI_iIX3aSlUmRtYPfbQHt4n0YaX0qw