Top all-time · #4365
Exo Labs
Company blog · blog.exolabs.netaihardware
Exo Labs is the company blog at blog.exolabs.net, writing about ai and hardware. Since 2024, 18 different Hacker News readers have linked it 19 times, most often “Nvidia DGX Spark and Apple Mac Studio = 4x Faster LLM Inference with EXO 1.0”. No feed was found, so read it on the site itself.
Too few sampled comments (9) to say how often HN recommends it.?
Why HN links it
HN readers link Exo Labs posts for technical comparisons of local LLM inference hardware, especially the DGX Spark and Mac prefill versus decode speeds, and for distributed inference across machines.
Summarised by Claude from a sample of 9 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Experiments, research, and practical guides to running AI on your own hardware.
Most-linked posts?
- Nvidia DGX Spark and Apple Mac Studio = 4x Faster LLM Inference with EXO 1.08 people · 9 mentions · first linked 2025-10-15
- Running Llama LLM on Windows 985 people · 5 mentions · first linked 2024-12-29
- Running DeepSeek V3 671B on M4 Mac Mini Cluster4 people · 4 mentions · first linked 2024-12-27
What Hacker News says?
https://blog.exolabs.net/nvidia-dgx-spark/
Comment linking /nvidia-dgx-spark/
on “Apple introduces M6 and M5 Ultra”Submitted Nvidia DGX Spark and Apple Mac Studio = 4x Faster LLM Inference with EXO 1.0 as a story · 61 points
HN discussionI'd be interested in seeing numbers that split out the speed of reading input (aka prefill) and the speed of generating output (aka decode). Those numbers are usually different and I remember from this Exo article that they could be quite radically different on Mac hardware: https://blog.exolabs.net/nvidia-dgx-spark/
Comment linking /nvidia-dgx-spark/
on “1.5 TB of VRAM on Mac Studio – RDMA over Thunderbolt 5”Submitted Running DeepSeek V3 671B on M4 Mac Mini Cluster as a story · 8 points
HN discussionIt was about this use case: https://blog.exolabs.net/nvidia-dgx-spark/
Comment linking /nvidia-dgx-spark/
on “macOS 26.2 enables fast AI clusters with RDMA over Thunderbolt”Submitted Clustering Nvidia DGX Spark and M3 Ultra Mac Studio for 4x Faster LLM Inference as a story · 5 points
HN discussionThe author mentions it but I want to expand on it: Apple is a seriously good option here, specifically the M4 Mac Mini. What makes Apple attractive is (as the author mentions) that RAM is shared between main and video RAM whereas NVidia is quite intentionally segmenting the market and charging huge premiums for high VRAM cards. Here are some options: 1. Base $599 Mac Mini: 16GB of RAM. Stocked in store. 2. $999 Mac…
Comment linking /day-2/
on “Building a personal, private AI computer on a budget”Submitted Running Llama LLM on Windows 98 as a story · 4 points
HN discussionNo, they use the same definition of memory bandwidth as others, but Apple Silicon has a lot of memory channels. In previous generations, prefill has been compute-limited and decode is fast. https://blog.exolabs.net/nvidia-dgx-spark/ outlines a combination of a DGX Spark and an M3 Ultra that took advantage of fast prefill on the Nvidia hardware and fast decode on Apple Silicon.
Comment linking /nvidia-dgx-spark/
on “Qwen 3.8-Flash-Next releasing tomorrow (125B a6B)”Submitted Combining Nvidia DGX Spark and Apple Mac Studio for 4x Faster LLM Inference as a story · 3 points
HN discussionEven an M3 Ultra won't put up similar GPU compute to a DGX Spark: https://blog.exolabs.net/nvidia-dgx-spark/ Fill up the memory with a large model, and most of your memory bandwidth will be waiting on compute shaders. Seems like a waste of $5,000 but you do you.
Comment linking /nvidia-dgx-spark/
on “Asus Ascent GX10”Submitted Running Llama on Windows 98 as a story · 3 points
HN discussionFor training, yes, you will need to share the parameters (i.e., weights and bias); the number is huge. But for inference, you don't need that much high bandwidth to run it in a distributed manner. According to the author of Exo https://blog.exolabs.net/day-1/ : > When Shard A finishes processing its layers, it produces an activation that gets passed to Shard B over whatever network connection is available. In genera…
Comment linking /day-1/
on “Nvidia GPU on bare metal NixOS Kubernetes cluster explained”Submitted Running DeepSeek V3 671B on M4 Mac Mini Cluster as a story · 3 points
HN discussionCheck out exo labs blog. https://blog.exolabs.net/day-2/
Comment linking /day-2/
on “Ask HN: Deepseek v3 how can I check on local computer?”Submitted Running Llama on Windows 98 - 12 days of EXO as a story · 2 points
HN discussionThey're probably referencing this article: https://blog.exolabs.net/nvidia-dgx-spark/
Comment linking /nvidia-dgx-spark/
on “Dell's version of the DGX Spark fixes pain points”Submitted Running Llama on Windows 98 as a story · 2 points
HN discussionSubmitted Running Llama on Windows 98 – 12 days of EXO as a story · 1 points
HN discussion
