Skip to content
Run LLM Locally

Qwen2.5 72B Instruct on the RX 7600 XT

Qwen2.5 72B Instruct at Q4_K_M needs about 46.5 GB, more than the 16 GB of the RX 7600 XT: part of the model runs from system RAM, which slows generation. At Q3_K_M it does not fit in VRAM.

ShareFacebookXWhatsAppLinkedInEmail

Qwen2.5 72B Instruct

Q4_K_M · 8K context

Not recommended

The model needs about 46.5 GB but only 15.5 GB of VRAM is usable: 31 GB must be loaded into system RAM. It runs, 33% on the GPU, but more slowly.

Estimated speed is at most 1.2 tokens/s: too slow to be usable for chat.

Your hardware
RX 7600 XT 16 GB · 64 GB RAM
Estimated memory
~46.5 GB estimate
VRAM (usable memory)
15.5 GB
GPU only
No
CPU/GPU offload
Required (31 GB)
Generation speed
0.6–1.2 tok/s estimate
Largest context in fast memory
None
Confidence: low

Based on: memory bandwidth (288 GB/s) and model size · system RAM bandwidth (~51 GB/s)

Memory breakdown

~46.5 GB (45.3 GB–47.8 GB) needed · estimate

31 GB must be offloaded to system RAM

  • Model weights41 GB
  • KV cache2.5 GB
  • Runtime overhead0.8 GB
  • Safety margin2.2 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (Vulkan) · 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
  • split CPU/GPU runs vary a lot with the CPU and PCIe link
  • system RAM speed unknown (dual-channel DDR4-3200 assumed)

Other quantizations on the RX 7600 XT

Qwen2.5 72B Instruct on RX 7600 XT + 64 GB RAM · 8K context
Qwen2.5 72B Instruct Q8_08K~79.3 GB · No dataIncompatible
Qwen2.5 72B Instruct Q6_K8K~62 GB · 0.4–0.8 tok/s estimateNot recommended
Qwen2.5 72B Instruct Q5_K_M8K~54 GB · 0.5–1 tok/s estimateNot recommended
Qwen2.5 72B Instruct Q4_K_M8K~46.5 GB · 0.6–1.2 tok/s estimateNot recommended
Qwen2.5 72B Instruct IQ4_XS8K~42.3 GB · 0.7–1.4 tok/s estimateNot recommended
Qwen2.5 72B Instruct Q3_K_M8K~38.2 GB · 0.8–1.6 tok/s estimateNot recommended

Alternatives

GPUs that run Qwen2.5 72B Instruct Q4_K_M entirely in VRAM

Other models that fit the RX 7600 XT

Hardware that runs Qwen2.5 72B Instruct Q4_K_M entirely in memory

Prices and links follow the market you choose.

Radeon PRO W7900

AMD

CompatibleRuns Qwen2.5 72B Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

AMD Ryzen AI Max+ 395

Unified memory PC

CompatibleRuns Qwen2.5 72B Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

GeForce RTX 5090

Lowest memory tier

ExcellentRuns Qwen2.5 72B Instruct 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 7600 XT run Qwen2.5 72B Instruct?
The model needs about 46.5 GB but only 15.5 GB of VRAM is usable: 31 GB must be loaded into system RAM. It runs, 33% on the GPU, but more slowly. Estimated speed is at most 1.2 tokens/s: too slow to be usable for chat.
Which quantization should I use on the RX 7600 XT?
None fits entirely in VRAM at 8K context; the smallest, Q3_K_M, still needs offload.
What is the largest context on the RX 7600 XT?
At Q4_K_M, the model does not fit entirely in VRAM even at a short context.
How fast is Qwen2.5 72B Instruct (RX 7600 XT)?
We estimate 0.6–1.2 tok/s (estimate, confidence low). Based on: memory bandwidth (288 GB/s) and model size, system RAM bandwidth (~51 GB/s).