MacBook Pro with M5 Pro and M5 Max
Apple’s new MacBook Pro lineup with M5 Pro and M5 Max chips is seen as a strong incremental hardware upgrade, bringing 18‑core CPUs, higher unified memory ceilings up to 128GB, modestly better memory bandwidth, and GPU “Neural Accelerators” that greatly speed up LLM prompt processing but not token generation. Commenters are split between excitement over local AI and pro workloads on unified memory, frustration at Apple’s RAM and charger pricing (especially in the EU/UK), and disappointment that laptops still top out at 128GB when many want 256GB+ for serious local models. Many note their existing M1–M3 machines remain more than fast enough, and several say they’re delaying hardware upgrades primarily because of dissatisfaction with the new macOS Tahoe UI and performance rather than CPU limits.
CPU & GPU Architecture
- Discussion centers on the new 18‑core CPU layout (6 “super” + 12 “performance” cores) and the move to a CPU+GPU chiplet (“Fusion”) design.
- Some think “super” is just a rename of old performance cores; others argue Apple now has three distinct tiers (super / performance / efficiency), with Pro/Max dropping E‑cores entirely.
- Several note this is a major architectural change but reserve judgment until independent benchmarks, especially around latency and any hidden chiplet interconnect costs.
Memory, Storage & Pricing
- Base storage bumps (1 TB on most configs, 2 TB on some Max) are welcomed, but many see it as a stealth price hike: base models went up ~$100–200 and RAM upgrades remain very expensive.
- Max RAM remains 128 GB, disappointing those hoping for 256–512 GB in a laptop for large local models.
- Some argue that, relative to rising PC RAM/SSD prices, high‑RAM Macs are currently “good value”; others point out Apple’s upgrade margins were already huge.
Local LLMs & AI Positioning
- Big debate around Apple’s “up to 4x faster LLM prompt processing” claim: clarified in the thread as improved prefill / time‑to‑first‑token, not tokens/sec.
- Memory bandwidth on M5 Max (614 GB/s) is only modestly up from M4 Max, so decode remains bandwidth‑bound; prefill benefits from new GPU “Neural Accelerators.”
- Split views on viability of local LLMs: some happily run 20–70B (and even ~120B MoE) models on high‑RAM Macs; others report poor speed and quality and stick to cloud APIs.
- Many see Apple’s unified memory as uniquely attractive for large models, but note they still lag high‑end NVIDIA GPUs in bandwidth and ecosystem (CUDA).
macOS Tahoe & Software Concerns
- Hardware widely praised; software is not. A sizable subset calls Tahoe “a disaster” and delays upgrading hardware until macOS 27 or beyond.
- Complaints include UI aesthetics (Liquid Glass, border radii), performance regressions (Safari, Spotlight, system apps), and general OS instability; others report zero issues and “buttery” performance.
- Several say they’ll freeze on older macOS or eventually move to Asahi Linux once Apple drops support.
Upgrade Decisions & Longevity
- Many M1/M2/M3 owners feel no compelling need to upgrade; Apple Silicon laptops are described as “too good,” still fast and cool under real workloads.
- Some recent M4 buyers are annoyed by the quick M5 Pro/Max release; others accept this as the usual Apple timing risk.
- Enthusiasm is highest among people coming from Intel Macs or those who specifically want 128 GB and faster LLM prefill; others plan to wait for rumored M6/OLED redesigns or M5 Ultra/Studio.