Skip to content
Run LLM Locally

Phi-4 Mini Instruct 3.8B on the RX 7900 GRE

With 16 GB of VRAM, the RX 7900 GRE runs Phi-4 Mini Instruct 3.8B at Q4_K_M entirely on the GPU with an 8K context (about 4.5 GB needed). At Q8_0 it fits entirely.

ShareFacebookXWhatsAppLinkedInEmail

Phi-4 Mini Instruct 3.8B

Q4_K_M · 8K context

Excellent

The model needs about 4.5 GB and fits in the 15.5 GB available, with 11 GB to spare.

Your hardware
RX 7900 GRE 16 GB · 64 GB RAM
Estimated memory
~4.5 GB estimate
VRAM (usable memory)
15.5 GB
GPU only
Yes
CPU/GPU offload
Not required
Generation speed
94–146 tok/s estimate
Largest context in fast memory
92K tokens
Confidence: low

Based on: memory bandwidth (576 GB/s) and model size

Memory breakdown

~4.5 GB (4.1 GB–5 GB) needed · estimate

Fits entirely in VRAM · 11 GB left

  • Model weights2.3 GB
  • KV cache1 GB
  • Runtime overhead0.9 GB
  • Safety margin0.3 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (ROCm) · flash attention on · weights = real file size

EstimateData updated 24 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card

What makes this estimate less reliable

  • no benchmark for this hardware and model yet

Other quantizations on the RX 7900 GRE

Phi-4 Mini Instruct 3.8B on RX 7900 GRE + 64 GB RAM · 8K context
Phi-4 Mini Instruct 3.8B Q8_08K~6 GB · 69–102 tok/s estimateExcellent
Phi-4 Mini Instruct 3.8B Q6_K8K~5.2 GB · 82–124 tok/s estimateExcellent
Phi-4 Mini Instruct 3.8B Q5_K_M8K~4.9 GB · 87–133 tok/s estimateExcellent
Phi-4 Mini Instruct 3.8B Q4_K_M8K~4.5 GB · 94–146 tok/s estimateExcellent
Phi-4 Mini Instruct 3.8B IQ4_XS8K~4.3 GB · 100–157 tok/s estimateExcellent
Phi-4 Mini Instruct 3.8B Q3_K_M8K~4.2 GB · 103–162 tok/s estimateExcellent

Alternatives

GPUs that run Phi-4 Mini Instruct 3.8B Q4_K_M entirely in VRAM

Other models that fit the RX 7900 GRE

Where to buy a RX 7900 GRE

Prices and links follow the market you choose.

Radeon RX 7900 GRE

16 GB GDDR6 · 576 GB/s · 260 W

ExcellentRuns Phi-4 Mini Instruct 3.8B Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

No current offer imported for this market: the link opens a marketplace search.

Frequently asked questions

Can the RX 7900 GRE run Phi-4 Mini Instruct 3.8B?
The model needs about 4.5 GB and fits in the 15.5 GB available, with 11 GB to spare.
Which quantization should I use on the RX 7900 GRE?
Q8_0 is the highest-precision quantization that fits entirely in VRAM at 8K context. ~6 GB / 15.5 GB.
What is the largest context on the RX 7900 GRE?
At Q4_K_M, about 92K tokens fit entirely in VRAM with an f16 KV cache. A q8_0 KV cache roughly doubles that.
How fast is Phi-4 Mini Instruct 3.8B (RX 7900 GRE)?
We estimate 94–146 tok/s (estimate, confidence low). Based on: memory bandwidth (576 GB/s) and model size.