Can a AMD Ryzen AI Max+ 395 “Strix Halo” (64GB) run Llama 3.3 70B Instruct?

Just about — Llama 3.3 70B Instruct runs at Q2_K (26.4 GB) at about 6–9 tok/s on a AMD Ryzen AI Max+ 395 “Strix Halo” (64GB).

How much memory does Llama 3.3 70B Instruct need?

Llama 3.3 70B Instruct needs about 28.5 GB at 4K context at its smallest quantization. A AMD Ryzen AI Max+ 395 “Strix Halo” (64GB) has 64.0 GB.

Tight fit: Llama 3.3 70B Instruct on a AMD Ryzen AI Max+ 395 “Strix Halo” (64GB) at Q2_K

Yes, but it's tight.

Tight / slow

Best fit: Llama 3.3 70B Instruct at Q2_K · 26.4 GB weights · 6–9 tok/s on a AMD Ryzen AI Max+ 395 “Strix Halo” (64GB).

Arena Text Overall · style-controlled

1,318Q2_K untested

llama-3.3-70b-instruct · published configuration reference · Q2_K not tested

Arena Text Overall (style-controlled), exact published entry llama-3.3-70b-instruct, 2026-10-08: reference rating 1,318. Published confidence bounds 1,314.5–1,321.5. 54,504 votes. Published llama-3.3-70b-instruct configuration reference. Reasoning mode is not disclosed by the Arena dataset. Tested precision, runtime and sampling settings are not disclosed in this dataset. Selected Q2_K has not been measured by this Arena entry. No quantization quality adjustment is applied.

Snapshot published 2026-10-08 · snapshot imported 2026-10-11 UTC · validated last-good data · manual refresh

Arena leaderboard dataset · Arena / lmarena-ai, Arena Leaderboard Dataset (CC BY 4.0). Selected Text Overall style-controlled rows; field names adapted, ratings unchanged. · dataset CC BY 4.0 · revision e16f90232986

Total memory at 4K context: 28.5 GB; Weights + KV cache + runtime overhead
Estimated generation: 6–9 tok/s
Estimated first token: ~37s

Every quantization on this machine

On small screens, each quant lists every metric below its name. On wider screens, scroll sideways for total memory, speed and fit. Keyboard: focus the table, then use the arrow keys. Quant names stay visible.

Llama 3.3 70B Instruct quantizations on AMD Ryzen AI Max+ 395 “Strix Halo” (64GB), estimated at 4K context. Weights and total memory are shown separately.
Quant	Weights	Total memory	Speed	Fit
Q2_K	26.4 GB	28.5 GB	6–9 tok/s	Tight / slow
Q3_K_M	34.3 GB	36.4 GB	5–7 tok/s	Tight / slow
Q4_K_M	42.5 GB	44.7 GB	4–6 tok/s	Tight / slow
Q5_K_M	50.0 GB	52.1 GB	— not resident	Won't fit
Q6_K	57.9 GB	60.0 GB	— not resident	Won't fit
Q8_0	75.0 GB	77.1 GB	— not resident	Won't fit
FP16	141.1 GB	143.3 GB	— not resident	Won't fit

Open the full calculator →Everything a AMD Ryzen AI Max+ 395 “Strix Halo” (64GB) runs Llama 3.3 70B Instruct on other machines What to buy to run it →

Ran Llama 3.3 70B Instruct on a AMD Ryzen AI Max+ 395 “Strix Halo” (64GB)? Add your real numbers →