M5 Max 48GB vs NVIDIA RTX 5090 for local LLMs
67 reports on the M5 Max 48GB and 170 on the NVIDIA RTX 5090
Which is faster for local LLMs?
On Qwen3.6 35B · 3B active at 4-bit, the M5 Max 48GB has 137.7 generation tokens/s (2 runs); prompt processing 3338.1 tokens/s (2 runs); the NVIDIA RTX 5090 has 215.1 generation tokens/s (one run). Those are community reports whose engines and settings differ, so the gap is a rough sign, not a controlled measure of the two cards. Memory bandwidth, which caps how fast a card can write, is 2.9x higher on the NVIDIA RTX 5090 (614 GB/s on the M5 Max 48GB, 1,792 on the NVIDIA RTX 5090). Across 6 shared models with generation measurements on both cards, the M5 Max 48GB's median is higher on 0 and the NVIDIA RTX 5090's on 6.
- Memory
- 48GB unified
- Memory bandwidth
- 614 GB/s
- FP16 compute
- not on record
- Reports
- 67
- Memory
- 32GB
- Memory bandwidth
- 1,792 GB/s
- FP16 compute
- 419 TFLOPS
- Reports
- 170
Measured on both
Generation and prompt processing speeds from plain runs (one device, one request, no speculative decoding, the whole model in the device's memory) of the same model size at the same bit class. Each side shows the engines and reported context behind its median. Context can mean a configured window or prompt depth, so these are community comparisons with differing settings. How to read these
| Model | M5 Max 48GB | NVIDIA RTX 5090 |
|---|---|---|
| Qwen3.6 35B · 3B active4-bit | 137.72 runsPP 3338.1 (2 runs)llama.cpp, MLXcontext not stated | 215.1one runllama.cppcontext not stated |
| Gemma 4 31B4-bit | 25.2one runPP 626.5 (one run)llama.cppcontext not stated | 64.0one runOllama32K context |
| Gemma 4 26B · 4B active4-bit | 116.9one runPP 3423.2 (one run)llama.cppcontext not stated | 243.9one runPP 13809.2 (one run)llama.cppcontext not stated |
| Qwen3 30B · 3B active4-bit | 148.5one runPP 3496.9 (one run)llama.cppcontext not stated | 352.1one runllama.cppcontext not stated |
| Qwen3.5 35B · 3B active4-bit | 111.2one runPP 3378.3 (one run)llama.cppcontext not stated | 194.0one runllama.cppcontext not stated |
| Qwen3.5 27B4-bit | 29.2one runPP 730.5 (one run)llama.cppcontext not stated | 30.1one runPP 1515.8 (one run)llama.cpp1K context |
Estimated by the calculator
The quant the calculator recommends for each card and its estimated speed at a 32,768-token context with 32 GB of system RAM, from memory bandwidth and the model's shape. Where a model was also measured above, the measurement wins.
| Model | M5 Max 48GB | NVIDIA RTX 5090 |
|---|---|---|
| Qwen3.6 35B · 3B active | 88.4 t/sQ5_K_M | 163.0 t/sQ5_K_M |
| Gemma 4 31B | 19.4 t/sQ6_K | 46.2 t/sQ5_K_M |
| Gemma 4 26B · 4B active | 43.2 t/sQ8_0 | 114.5 t/sQ6_K |
| Qwen3 30B · 3B active | 104.1 t/sQ5_K_S | 262.8 t/sQ4_K_S |
| Qwen3.5 35B · 3B active | 88.4 t/sQ5_K_M | 163.0 t/sQ5_K_M |
| Qwen3.5 27B | 22.3 t/sQ6_K | 44.9 t/sQ6_K |
| Qwen3.8 27B | 22.3 t/sQ6_K | 44.9 t/sQ6_K |
| Qwen3.8 125B · 6B active | doesn't fit at these settings | 20.3 t/sQ2_K, experts in RAM |
| DeepSeek V4 Flash 284B · 13B active | doesn't fit at these settings | doesn't fit at these settings |
| Qwen3.6 27B | 22.3 t/sQ6_K | 44.9 t/sQ6_K |
| DeepSeek V4.1 Flash 552B · 16B active | doesn't fit at these settings | doesn't fit at these settings |
| GLM-5.3 320B · 18B active | doesn't fit at these settings | doesn't fit at these settings |
Change the context, RAM or model in the calculator: M5 Max 48GB · NVIDIA RTX 5090
Every report on each card, with its source: M5 Max 48GB · NVIDIA RTX 5090