As the Senior Analyst, Graphics at Tom's Hardware, Jeff Kampman covers everything that has to do with graphics cards, gaming performance, and more. From integrated graphics processors to discrete graphics cards to the hyperscale installations powering our AI future, if it's got a GPU in it, Jeff is on it.\u00a0 ","collapsible":{"enabled":true,"maxHeight":250,"readMoreText":"Read more","readLessText":"Read less"}}), "https://slice.vanilla.futurecdn.net/13-4-25/js/authorBio.js"); } else { console.error('%c FTE ','background: #9306F9; color: #ffffff','no lazy slice hydration function available'); } Jeffrey Kampman Senior Analyst, Graphics As the Senior Analyst, Graphics at Tom's Hardware, Jeff Kampman covers everything that has to do with graphics cards, gaming performance, and more. From integrated graphics processors to discrete graphics cards to the hyperscale installations powering our AI future, if it's got a GPU in it, Jeff is on it.
Kindaian There are some significant issues with using a mac though, specially in terms of service setup and containers (no gpu passthrough for containers under macos due to cpu/hardware limitations). If those are not important to you, then the comparison is fair i would guess. BUT also one consideration is the price in relation to the performance. Reply
splus I don't get it – why would you test the old M4 Max Studio when there's a new M5 Max MacBook, which is better and faster, especially for AI? And, as you said, the M4 Max Studio configuration you tested basically isn't even available any more? Reply
JamesJones44 splus said: I don't get it – why would you test the old M4 Max Studio when there's a new M5 Max MacBook, which is better and faster, especially for AI? And, as you said, the M4 Max Studio configuration you tested basically isn't even available any more? Prior to Apple limiting the M4 Max Studios RAM I would have argued 256 GB of VRAM would be better for a pro user than the M5 Max 128 GB. However, with Apple cutting that option, I agree, the M5 Max 128 GB MBP would have been a better comparison Reply
splus JamesJones44 said: Prior to Apple limiting the M4 Max Studios RAM I would have argued 256 GB of VRAM would be better for a pro user than the M5 Max 128 GB. However, with Apple cutting that option, I agree, the M5 Max 128 GB MBP would have been a better comparison It's good to have more memory, but this hardware and its memory bandwidth is simply too low for larger models. It would be too slow to be useful. 256 or 512 GB would be useful only with much faster chip (like M5 Max or Mx Ultra) and with much higher memory bandwidth. M4 Max doesn't have either. Same applies to the Spark and Ryzen 395. 128 GB is their practical and usable max limit. Reply
Dmtrii I don't understand how you're testing Geekbench multi-core! Either Windows is broken (I don't know, I haven't used it for 10+ years) or you chose the wrong power profile. Here's my result of Geekbench 7 for the Strix Halo 128GB (Framework, Linux, Performance profile) – https://browser.geekbench.com/v7/cpu/1327 More than 30 000 for multi-core! The same was for Geekbench 6 – every time I got 15% better results in multi-core 🤷♂️ Reply
Bikki @Jefferey Lamma.cpp does not natively support mlx, it supports metal but not Mlx. Ollama or Llm studio do. On Ollama new mlx backend, Qwen prefill speed increases by 1.5x and decode 2-3x on m4 pro compared to old version that uses llama.cpp backend. Ps: I’m currently an AI engineer, gladly contributr to AI article’s process unpaid. contact me at bik.huynguyen gmail Ref: https://ollama.com/blog/mlx Reply
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://www.tomshardware.com/desktops/SPONSORED_LINK_URL
- https://www.tomshardware.com/desktops/exploring-apple-silicons-local-ai-performance-with-the-mac-studio-and-m4-max-m4-max-beats-gb10-and-strix-halo-in-decode-throughput-but-memory-bandwidth-isnt-everything#main
- https://www.tomshardware.com/membership
- $21 Nintendo Wii U upgrade lets you add TBs of storage to your console with an M.2 or SATA SSD — upgrade makes console slightly faster, but requires Homebrew an
- NVIDIA Vera Rubin Maximizes Intelligence per Dollar for Post-Training Workloads — a Key Metric for Agentic AI
- New open source printer has 7 toolheads that swap in 5 seconds for fast, zero-waste multi-color 3D printing — Sovol M1D 3D printer is the first open-source IDEX
- 32GB Corsair Vengeance DDR5 is $369 in this Woot sale — the lowest standalone RAM price right now, thanks to a $132 discount
- Firm that uses AI to locate ancient lost shipwrecks is hiring a literal pirate to salvage sunken treasure, paying up to $500,000 a year — AI mines 500 years of
Informational only. No financial advice. Do your own research.