text model · Flint · Linux
Can I run Flint 40B-A2.5B on AMD Ryzen AI Halo (128GB)?
Yes. Flint 40B-A2.5B runs on AMD Ryzen AI Halo (128GB) at Q4_K_M (~25.9 GB of ~96 GB usable).
Runs at Q4_K_M using ~25.9 GB of ~96 GB usable. You have room for FP16 for higher quality.
That figure is at a 4k context and moves about ±15% as context length changes. AMD Ryzen AI Halo (128GB) leaves ~70.1 GB of headroom, room to step up to FP16 for higher quality.
- Q4_K_M needed
- ~25.9 GB
- Usable on device
- ~96 GB
- Device memory
- 128 GB
- Best quant
- Q4_K_M
Which quant fits
Run it
Pick your tool. All 3 load the same Q4_K_M weights.
ollama run flint:40b llama-cli -hf bartowski/RecursiveSI_flint-GGUF:Q4_K_M lms get bartowski/RecursiveSI_flint-GGUF - Parameters
- 39.8B (MoE, 2.5B active)
- Q4_K_M size
- 23.61 GB
- Q8_0 size
- 39.44 GB
- Context
- 256k
- Ollama tag
- flint:40b
- Memory
- 128 GB unified
- Usable for weights
- ~96 GB
- Power draw
- ~120 W
- Best runtime
- llama.cpp (Vulkan/ROCm) / LM Studio
You could also run
Run Flint 40B-A2.5B on other hardware
FAQ
Can AMD Ryzen AI Halo (128GB) run Flint 40B-A2.5B?
Yes. Flint 40B-A2.5B runs on AMD Ryzen AI Halo (128GB) at Q4_K_M (~25.9 GB of ~96 GB usable).
How much memory does Flint 40B-A2.5B need?
AMD Ryzen AI Halo (128GB) has room to spare. At Q4_K_M the weights are ~23.61 GB; with KV cache and runtime overhead, budget ~25.9 GB at a 4k context. It is a Mixture-of-Experts model (39.8B total / 2.5B active), so all experts must stay in memory; memory tracks total params, not active params.
What is the best tool to run Flint 40B-A2.5B on Linux?
Ollama for a simple setup; vLLM for the most speed. vLLM shines for multi-user serving/throughput. For a single local chat, Ollama or llama.cpp is simpler and lighter.
Embed this
A live badge for your model card or README, updated as the data is. [](https://localmodel.run/can-i-run/flint-40b-a2.5b/amd-ryzen-ai-halo-128gb) Sources
-
amd.com · 2 sources
-
cnx-software.com · 1 source
-
frame.work · 1 source
-
github.com · 1 source
-
huggingface.co · 3 sources
-
microcenter.com · 1 source
-
notebookcheck.net · 1 source
-
ollama.com · 2 sources
Weights are measured from GGUF files; KV cache and overhead are computed, so totals can vary ~15% with context and runtime. Any tok/s is a bandwidth estimate. See methodology.