Can a Ryzen AI Max+ 395 128GB run Llama 3.3 70B Instruct?
Yes
At Q8_0 it needs 72.85 GB of the 96 GB available and generates around 2.0 tokens per second at an 8k context window.
In practice
Memory is not the problem here, bandwidth is. Llama 3.3 70B Instruct at Q8_0 fits with 23.15 GB to spare, and the Ryzen AI Max+ 395 128GB reads memory at 256 GB/s. Every token needs about 75.0 GB read out of it, which puts the ceiling near 2.0 tokens per second. That is slower than reading: a 500-token answer takes about 4 minutes.
It is fine for jobs you start and walk away from, and tiring for conversation. A smaller file raises the speed in proportion: Q6_K would run at about 2.6 tokens per second, at some cost in quality. A smaller model, or a mixture-of-experts model that reads only a fraction of its weights per token, is the other way out, and both are listed further down.
Reading a 4,000-token prompt, roughly a long email or a few pages of text, takes about 50 seconds before the first word appears. That part is limited by compute and not by bandwidth, so it follows a different spec from the writing speed.
The numbers
Llama 3.3 70B Instruct has 70.55 billion parameters across 80 layers. The Ryzen AI Max+ 395 128GB has 128 GB of unified memory at 256 GB/s, of which about 96 GB is left for a model once the operating system has taken its share.
| Quantization | Total needed | Fits | Speed |
|---|---|---|---|
| Q8_0 | 72.85 GB | yes | 2.0 t/s |
| Q6_K | 56.92 GB | yes | 2.6 t/s |
| Q5_K_M | 49.77 GB | yes | 3.0 t/s |
| Q4_K_M | 42.71 GB | yes | 3.5 t/s |
| IQ4_XS | 37.95 GB | yes | 3.9 t/s |
| Q3_K_M | 35.15 GB | yes | 4.3 t/s |
| IQ3_XXS degraded | 28.17 GB | yes | 5.3 t/s |
| Q2_K degraded | 30.56 GB | yes | 4.9 t/s |
| IQ2_XXS degraded | 19.96 GB | yes | 7.6 t/s |
9 of the 9 levels fit entirely. Levels marked degraded are listed for completeness, not as advice.
How long a conversation it holds
Memory rises with the length of the conversation because the attention cache keeps a key and a value for every token. At Q8_0, the longest window that still fits on this device is 64k tokens. Setting the window to what you need and no more is the cheapest way to buy back memory.
| Context | Cache | Total | Fits |
|---|---|---|---|
| 4k | 1.25 GB | 71.6 GB | yes |
| 8k | 2.5 GB | 72.85 GB | yes |
| 16k | 5 GB | 75.35 GB | yes |
| 32k | 10 GB | 80.35 GB | yes |
| 64k | 20 GB | 90.35 GB | yes |
| 128k | 40 GB | 110.35 GB | no |
The settings we would use
llama-server -m <model>.gguf -ngl all -c 8192 -fa on
In Ollama, set OLLAMA_CONTEXT_LENGTH to the window you want, run ollama ps while the model is loaded and check that
the processor column reads 100% GPU. If it shows a split on a model that fits, the Ollama guide shows how to fix it.
The llama.cpp flags guide explains each option above.
What else a Ryzen AI Max+ 395 128GB runs well
For comparison, these are the other popular models that fit this device at a sound quantization, best pairings first.
| Model | Quantization | Memory | Speed |
|---|---|---|---|
| GPT OSS 120B | MXFP4 | 59.68 GB | 26.8 t/s |
| Qwen3 30B A3B | Q8_0 | 31.3 GB | 21.0 t/s |
| GPT OSS 20B | MXFP4 | 11.84 GB | 35.3 t/s |
| Qwen3.5 9B | Q8_0 | 10.25 GB | 16.5 t/s |
The same model on nearby hardware
Memory decides whether Llama 3.3 70B Instruct loads and bandwidth decides how quickly it answers. Here is what a step down, a card with the same memory and a step up do with it.
| Device | Memory | Bandwidth | Result |
|---|---|---|---|
| Apple M4 Max 64GB | 64 GB | 546 GB/s | IQ4_XS, 10.1 t/s |
| Apple M4 Max 128GB | 128 GB | 546 GB/s | IQ4_XS, 10.1 t/s |
See how fast it feels
Ryzen AI Max+ 395 128GB running Llama 3.3 70B Instruct at Q8_0
YouWhy does my model use more memory when the conversation gets longer?
Reading your question
Simulated from our estimate at a 512-token question, not a recording. Assumes nothing else is competing for the GPU.
Questions people ask
How much VRAM does Llama 3.3 70B Instruct need on Ryzen AI Max+ 395 128GB?
At Q4_K_M with an 8k context window it needs 42.71 GB in total: 39.67 GB of weights, 2.5 GB of KV cache and 0.54 GB of runtime overhead. The Ryzen AI Max+ 395 128GB makes about 96 GB available once the operating system has taken its share.
What context length can Ryzen AI Max+ 395 128GB handle with Llama 3.3 70B Instruct?
At Q8_0 the longest window that still fits is about 64k tokens. Memory grows with the window, so a shorter one leaves room for a higher quantization.
How fast is Llama 3.3 70B Instruct on Ryzen AI Max+ 395 128GB?
We estimate about 2.0 tokens per second at Q8_0, which is slower than comfortable reading speed. A 500-token answer takes about 4 minutes. Memory bandwidth of 256 GB/s sets that ceiling, and the estimate is good to about 10 to 30 percent.
Keep going
Everything the Ryzen AI Max+ 395 128GB runs, the full breakdown for Llama 3.3 70B Instruct, how the numbers are made in the methodology, or a different pairing in the calculator.