November 30, 2025
Neural speed, user snark
I Tested the M5 iPad Pro's Neural-Accelerated AI, and the Hype Is Real
Blazing iPad AI, but commenters say it won’t matter till Mac
TLDR: The M5 iPad Pro blasts through AI prompts far faster than the M4, even topping Apple’s claims. Commenters love the speed but argue it won’t matter until Mac apps and iPadOS catch up, while developers see a big win for future offline, private assistants—if the software ecosystem shows up.
The tester says Apple’s new M5 iPad Pro doesn’t just meet the AI hype—it beats it, slicing the “time to first token” (aka how fast the first word appears) from minutes to seconds on long prompts. The crowd? Split. One camp is cheering the numbers; the other is rolling their eyes and blaming the iPad’s locked‑down vibe. Bigyabai’s snarky mood captured it: great hardware, but iPadOS is the fun police. Translation: without desktop‑style apps, most people won’t feel the turbo. Meanwhile, devs are quietly drooling. Faster prompt processing means smarter offline assistants, quicker cross‑document search, and beefier “project” features—if apps like Locally AI, OfflineLLM, and Craft tap into it. Joakleaf drops receipts with Apple’s own deep‑dive: Exploring LLMs with MLX and the Neural Accelerators in the M5 GPU, stoking the “this will shine on Mac first” chorus. The meme energy? People joked that TTFT now stands for “Time To First TikTok,” and compared M4 vs M5 waits to pizza delivery vs instant noodles. Bottom line: the iPad got insanely fast at starting answers, but the fight is over whether that matters today—or if it’s just a teaser for the Mac crowd and future iPad apps.
Key Points
- •Pre-release MLX optimized for M5 shows local AI prompt processing gains exceeding Apple’s 3.5× claim.
- •For a 10,000-token prompt, TTFT improved from 81 seconds (M4) to 18 seconds (M5), a 4.4× speedup.
- •For a 16,000-token prompt, TTFT improved from 118 seconds (M4) to 38 seconds (M5).
- •Token generation speeds improved modestly (~1.5×) compared to significant prefill-stage gains.
- •Developers are advised to integrate MLX and design long-context features (e.g., RAG, MCP) to leverage M5 acceleration; some iPad apps may benefit once public MLX supports neural acceleration.