RX 6900 XT

AMD Radeon RX 6000 (RDNA 2)

VRAM
16 GB
Memory bandwidth
512 GB/s
Class
Consumer

Figures assume this GPU plus 32 GB of system RAM (a typical desktop pairing). "What runs on it" is judged at a 8,192-token context. Speeds are estimates, not measurements.

What runs on it

Model Sweet-spot quant Est. speed Community
DeepSeek-R1-Distill-Llama-8B deepseek-ai Q8_0 Runs fully on GPU @ 8K ctx 32–43 tok/s (estimate) no community data
DeepSeek-R1-Distill-Qwen-1.5B deepseek-ai Q8_0 Runs fully on GPU @ 8K ctx 144–192 tok/s (estimate) no community data
DeepSeek-R1-Distill-Qwen-14B deepseek-ai Q4_K_M Runs fully on GPU @ 8K ctx 29–39 tok/s (estimate) no community data
DeepSeek-R1-Distill-Qwen-7B deepseek-ai Q8_0 Runs fully on GPU @ 8K ctx 36–48 tok/s (estimate) no community data
Devstral-Small-2-24B-Instruct-2512 mistralai IQ4_XS Runs fully on GPU @ 8K ctx 22–29 tok/s (estimate) no community data
Kimi-VL-A3B-Instruct moonshotai Q4_K_M Runs fully on GPU @ 8K ctx no community data
Llama-3.1-8B-Instruct meta-llama Q8_0 Runs fully on GPU @ 8K ctx 32–43 tok/s (estimate) no community data
Mistral-Small-3.2-24B-Instruct-2506 mistralai IQ4_XS Runs fully on GPU @ 8K ctx 22–29 tok/s (estimate) no community data
Phi-4-mini-instruct microsoft Q8_0 Runs fully on GPU @ 8K ctx 60–79 tok/s (estimate) no community data
Phi-4-reasoning microsoft Q4_K_M Runs fully on GPU @ 8K ctx 29–38 tok/s (estimate) no community data
Qwen2.5-7B-Instruct Qwen Q8_0 Runs fully on GPU @ 8K ctx 36–48 tok/s (estimate) no community data
Qwen3-14B Qwen Q4_K_M Runs fully on GPU @ 8K ctx 30–40 tok/s (estimate) no community data
Qwen3-8B Qwen Q8_0 Runs fully on GPU @ 8K ctx 31–41 tok/s (estimate) no community data
Qwen3-Embedding-0.6B Qwen Q8_0 Runs fully on GPU @ 8K ctx 195–259 tok/s (estimate) no community data
Qwen3-Embedding-4B Qwen Q4_K_M Runs fully on GPU @ 8K ctx 83–111 tok/s (estimate) no community data
Qwen3-Embedding-8B Qwen Q4_K_M Runs fully on GPU @ 8K ctx 52–70 tok/s (estimate) no community data
Qwen3-Reranker-0.6B Qwen BF16 Runs fully on GPU @ 8K ctx 144–192 tok/s (estimate) no community data
Qwen3-Reranker-4B Qwen BF16 Runs fully on GPU @ 8K ctx 33–44 tok/s (estimate) no community data
SmolLM3-3B HuggingFaceTB BF16 Runs fully on GPU @ 8K ctx 45–61 tok/s (estimate) no community data
dots.ocr rednote-hilab BF16 Runs fully on GPU @ 8K ctx 49–65 tok/s (estimate) no community data
gemma-4-12B-it google Q8_0 Runs fully on GPU @ 8K ctx 23–30 tok/s (estimate) no community data
gemma-4-26B-A4B-it google QAT-Q4_0 Runs fully on GPU @ 8K ctx no community data
gemma-4-E2B-it google Q8_0 Runs fully on GPU @ 8K ctx 60–80 tok/s (estimate) no community data
gemma-4-E4B-it google Q8_0 Runs fully on GPU @ 8K ctx 37–49 tok/s (estimate) no community data
gpt-oss-20b openai F16 Runs fully on GPU @ 8K ctx no community data
phi-4 microsoft Q4_K_M Runs fully on GPU @ 8K ctx 29–39 tok/s (estimate) no community data
DeepSeek-R1-Distill-Qwen-32B deepseek-ai Q8_0 CPU offload @ 8K ctx · ~22 GPU layers 2 tok/s (estimate) no community data
GLM-4.7-Flash zai-org Q8_0 CPU offload @ 8K ctx · ~20 GPU layers no community data
Hunyuan-A13B-Instruct tencent IQ4_XS CPU offload @ 8K ctx · ~9 GPU layers no community data
Kimi-Dev-72B moonshotai IQ4_XS CPU offload @ 8K ctx · ~22 GPU layers 1–2 tok/s (estimate) no community data
Llama-3.3-70B-Instruct meta-llama Q4_K_M CPU offload @ 8K ctx · ~21 GPU layers 1–2 tok/s (estimate) no community data
Qwen2.5-Omni-7B Qwen BF16 CPU offload @ 8K ctx · ~17 GPU layers 4–5 tok/s (estimate) no community data
Qwen2.5-VL-32B-Instruct Qwen Q8_0 CPU offload @ 8K ctx · ~22 GPU layers 2 tok/s (estimate) no community data
Qwen3-30B-A3B-Instruct-2507 Qwen Q8_0 CPU offload @ 8K ctx · ~19 GPU layers no community data
Qwen3-32B Qwen Q8_0 CPU offload @ 8K ctx · ~22 GPU layers 2 tok/s (estimate) no community data
Qwen3-Coder-30B-A3B-Instruct Qwen Q8_0 CPU offload @ 8K ctx · ~19 GPU layers no community data
Qwen3-Omni-30B-A3B-Instruct Qwen Q4_K_M CPU offload @ 8K ctx · ~36 GPU layers no community data
Qwen3-Reranker-8B Qwen BF16 CPU offload @ 8K ctx · ~30 GPU layers 9–11 tok/s (estimate) no community data
Qwen3.6-27B Qwen Q8_0 CPU offload @ 8K ctx · ~27 GPU layers 2–3 tok/s (estimate) no community data
Qwen3.6-35B-A3B Qwen Q8_0 CPU offload @ 8K ctx · ~14 GPU layers no community data
gemma-4-31B-it google Q8_0 CPU offload @ 8K ctx · ~22 GPU layers 2 tok/s (estimate) no community data

"Est. speed" is a modelled range labelled estimate (D8) for generation (decode) throughput. "Community" shows the median of approved user-submitted reports on this GPU class only where enough exist — never an estimate. "pp" is measured prompt-processing (ingestion) throughput from approved community reports; rows without a measurement show none.

What can I run on a RX 6900 XT?

On this GPU, 12 catalog models run fully on the GPU at an 8,192-token context. The most capable is gemma-4-26B-A4B-it at QAT-Q4_0 (needs ~15.8 GiB). Pick a smaller model or a lower quant for more headroom.

Biggest model: gemma-4-26B-A4B-it at QAT-Q4_0 

llama-server -m gemma-4-26B_q4_0-it.gguf -c 8192 -ngl 999

Derived from the fit engine at an 8,192-token context. See more answer packs.

Add a second RX 6900 XT?

A second RX 6900 XT pools VRAM: 2 × 16 GB = 32 GB combined. Bigger models can then load because their weights split across both cards — but a second card does not make generation proportionally faster (see the reality check below).

A second RX 6900 XT adds 16 GB of VRAM for about $400 — ≈$25.00/GB of added VRAM (used price as of 2026-07-18 — BestValueGPU price tracker).

12 more catalog models could newly fit fully in the combined 32 GB at a 8,192-token context — for example:

Estimate — assumes the model's weights split across both cards (a layer split, as llama.cpp does by default). This is a combined-VRAM projection, not a measured or verdict-chipped result: the fit engine treats two cards as one summed memory pool and does not model the link between them. Check your exact model and context in the calculator.

The honest reality of a second card

  • Bandwidth doesn't add. Two cards give more VRAM, not more memory bandwidth per token. Token generation is bandwidth-bound, so a layer-split model decodes at roughly one card's speed — not double.
  • Layer-split vs tensor-parallel. The common desktop setup (llama.cpp) splits layers across cards and runs them in sequence, so one GPU works at a time. True tensor-parallel serving (e.g. vLLM) can use both at once, but wants matched cards and a fast interconnect.
  • pp vs tg. Prompt processing (pp) can gain more from a second card than token generation (tg); raw decode throughput barely moves. Don't expect a 2× tok/s jump.
  • PCIe / NUMA. Cards talk over PCIe (or NVLink where supported), far slower than on-card VRAM. A layer split crosses it about once per token so the hit is small; tensor-parallel crosses it constantly, and cards on different CPU sockets (NUMA) add latency.
  • Power & PSU. A second card roughly doubles GPU power draw — check PSU headroom, connectors, slot spacing and airflow before buying.

Read: Multi-GPU for local LLMs — when a second card is worth it →

Derived from the fit engine at a 8,192-token context, comparing a single 16 GB card against a summed 32 GB two-card pool.

Price history

Prices are point-in-time observations, not live quotes.

new

  • Dec 2020 · $999 new — The FPS Review launch coverage

Not enough history yet — trends appear once at least three dated observations are recorded.

used

  • Jul 2026 · $400 used — BestValueGPU price tracker (medium confidence)

Not enough history yet — trends appear once at least three dated observations are recorded.

selected to compare · pick at least 2

Compare ()