Skip to content
Run LLM Locally

GPU · AMD

RX 7900 XT for local LLMs: which models run?

With 20 GB of VRAM and 800 GB/s of bandwidth, the RX 7900 XT runs models up to about Gemma 3 27B IT entirely on the GPU.

  • VRAM20 GB GDDR6
  • Bandwidth800 GB/s
  • Board power315 W
  • ArchitectureRDNA 3
ShareFacebookXWhatsAppLinkedInEmail

Specifications

VRAM
20 GB
Memory type
GDDR6
Memory bus
320-bit
Bandwidth
800 GB/s
Board power
315 W
Backends
ROCm, Vulkan
Interface
PCIe 4.0 x16
Released
December 2022
Segment
Consumer
FP16 compute
Not available

Official specData updated 24 days agoSource: AMD specifications

What the RX 7900 XT runs

Popular models on RX 7900 XT + 64 GB RAM · 8K context
gpt-oss-20bMXFP4 · 8K~13 GB · 138–152 tok/s measuredExcellent
Mistral Nemo 12B InstructQ8_0 · 8K~14.9 GB · 36–51 tok/s estimateExcellent
Gemma 3 12B ITQ8_0 · 8K~14.2 GB · 36–53 tok/s estimateExcellent
Gemma 2 9B ITQ8_0 · 8K~12.8 GB · 43–62 tok/s estimateExcellent
Qwen3 8BQ8_0 · 8K~10.6 GB · 50–73 tok/s estimateExcellent
Llama 3.1 8B InstructQ8_0 · 8K~10.2 GB · 52–75 tok/s estimateExcellent
DeepSeek R1 Distill Llama 8BQ8_0 · 8K~10.2 GB · 52–75 tok/s estimateExcellent
Qwen2.5 7B InstructQ8_0 · 8K~9.3 GB · 57–82 tok/s estimateExcellent
Mistral 7B Instruct v0.3Q8_0 · 8K~9.2 GB · 56–82 tok/s estimateExcellent
Gemma 3 4B ITQ8_0 · 8K~5.5 GB · 91–142 tok/s estimateExcellent
Qwen3 4BQ8_0 · 8K~6.2 GB · 82–127 tok/s estimateExcellent
Phi-4 Mini Instruct 3.8BQ8_0 · 8K~6 GB · 88–135 tok/s estimateExcellent
Llama 3.2 3B InstructQ8_0 · 8K~5.2 GB · 103–160 tok/s estimateExcellent
Qwen3 1.7BQ8_0 · 8K~3.7 GB · 146–240 tok/s estimateExcellent
Llama 3.2 1B InstructQ8_0 · 8K~2.6 GB · 240–390 tok/s estimateExcellent
Qwen3 0.6BQ8_0 · 8K~2.6 GB · 210–386 tok/s estimateExcellent
Qwen3 32BQ3_K_M · 8K~18.6 GB · 27–40 tok/s estimateCompatible
DeepSeek R1 Distill Qwen 32BQ3_K_M · 8K~18.6 GB · 27–40 tok/s estimateCompatible
Qwen2.5 32B InstructQ3_K_M · 8K~18.6 GB · 27–40 tok/s estimateCompatible
Qwen2.5 Coder 32B InstructQ3_K_M · 8K~18.6 GB · 27–40 tok/s estimateCompatible
QwQ 32BQ3_K_M · 8K~18.6 GB · 27–40 tok/s estimateCompatible
Qwen3 30B-A3BIQ4_XS · 8K~17.7 GB · 78–143 tok/s estimateCompatible
Qwen3 Coder 30B-A3B InstructIQ4_XS · 8K~17.7 GB · 78–143 tok/s estimateCompatible
Gemma 3 27B ITQ4_K_M · 8K~18.4 GB · 28–40 tok/s estimateCompatible
Gemma 2 27B ITIQ4_XS · 8K~17.9 GB · 31–44 tok/s estimateCompatible
Mistral Small 3.2 24B InstructQ5_K_M · 8K~18.6 GB · 29–41 tok/s estimateCompatible
Qwen3 14BQ8_0 · 8K~17.6 GB · 30–43 tok/s estimateCompatible
DeepSeek R1 Distill Qwen 14BQ8_0 · 8K~17.8 GB · 29–42 tok/s estimateCompatible
Qwen2.5 14B InstructQ8_0 · 8K~17.8 GB · 29–42 tok/s estimateCompatible
Phi-4 14BQ8_0 · 8K~17.7 GB · 30–43 tok/s estimateCompatible
Mixtral 8x7B InstructQ3_K_M · 8K~21.6 GB · 19–36 tok/s estimateCPU/GPU offload
gpt-oss-120bMXFP4 · 8K~63.3 GB · 9–18 tok/s estimatePossible but slow
Llama 4 Scout 17B-16E InstructQ3_K_M · 8K~55.7 GB · 3–6 tok/s estimatePossible but slow
Qwen2.5 72B InstructQ3_K_M · 8K~38.3 GB · 1–2 tok/s estimatePossible but slow
Llama 3.3 70B InstructQ3_K_M · 8K~37 GB · 1.1–2 tok/s estimatePossible but slow
DeepSeek R1 Distill Llama 70BQ3_K_M · 8K~37 GB · 1.1–2 tok/s estimatePossible but slow
DeepSeek R1 (671B)Q8_0 · 8K~699 GB · No dataIncompatible
Qwen3 235B-A22BQ8_0 · 8K~247 GB · No dataIncompatible

For each model, the highest-precision quantization that runs entirely in fast memory; otherwise the one with the best result. Speeds are ranges; "estimate" unless a benchmark exists for this exact hardware.

EstimateData updated 24 days agoSource: Run LLM Locally engine

Limits in plain words

20 GB of VRAM holds the weights and KV cache of models up to about Gemma 3 27B IT at Q4_K_M. Bigger models spill into system RAM and slow down sharply: generation speed is then limited by RAM bandwidth, not by the GPU.

Where to buy a RX 7900 XT

Prices and links follow the market you choose.

Radeon RX 7900 XT

20 GB GDDR6 · 800 GB/s · 315 W

RX 7900 XT 20 GB · 64 GB RAM → Gemma 3 27B IT

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

Which LLMs can the RX 7900 XT run?
Entirely on the GPU: Qwen3 32B (Q3_K_M), DeepSeek R1 Distill Qwen 32B (Q3_K_M), Qwen2.5 32B Instruct (Q3_K_M), Qwen2.5 Coder 32B Instruct (Q3_K_M), QwQ 32B (Q3_K_M), Qwen3 30B-A3B (IQ4_XS), Qwen3 Coder 30B-A3B Instruct (IQ4_XS), Gemma 3 27B IT (Q4_K_M). Larger models need CPU/GPU offload.
Can the RX 7900 XT run a 70B model?
Not recommended: a 70B model at Q4_K_M needs about 45.1 GB; the RX 7900 XT has 20 GB.
How is the speed estimated?
Token generation reads every active weight once per token, so it is limited by memory bandwidth (800 GB/s here). We apply an efficiency range per backend and label the result as an estimate until real benchmarks exist.