← AI Hashrate

RX 7900 XTX 24GB

AMD · 24 GB GDDR6 · 960 GB/s · 355 W · MSRP $999 · ASRock Radeon RX 7900 XTX Phantom Gaming 24GB OC (affiliate) · More listings

The RX 7900 XTX 24GB is a AMD accelerator with 24 GB GDDR6 and about 960 GB/s peak memory bandwidth. AI Hashrate estimates local LLM decode (batch 1) at 8K context: fit counts weights + KV cache + 1 GB runtime overhead — not weights alone. At Q4, about 33 curated models fit fully on this card; at FP16, about 17 fit. Top Q4 speeds: Qwen3-0.6B ≈ 763.6 tok/s (estimated, fits); gemma-3-1b-it ≈ 610.9 tok/s (estimated, fits); Llama-3.2-1B-Instruct ≈ 509.1 tok/s (estimated, fits); Qwen3-1.7B ≈ 305.5 tok/s (estimated, fits); Qwen3-30B-A3B ≈ 203.6 tok/s (estimated, fits). Relative ranking is more reliable than absolute tok/s. See methodology for the bandwidth formula and measured-anchor policy. Methodology.

Table: decode speed estimates (Q4 / FP16). Measured rows override estimates. MSRP is list price, not live retail. Context default 8K.

ModelParams (B)Quanttok/sFits?
Qwen3-0.6B0.8Q4763.6 est.Yes
gemma-3-1b-it1.0Q4610.9 est.Yes
Llama-3.2-1B-Instruct1.2Q4509.1 est.Yes
Qwen3-1.7B2.0Q4305.5 est.Yes
Qwen3-0.6B0.8FP16210.0 est.Yes
Qwen3-30B-A3B30.5Q4203.6 est.Yes
Qwen3-Coder-30B-A3B-Instruct30.5Q4203.6 est.Yes
Qwen3-30B-A3B-Instruct-250730.5Q4203.6 est.Yes
Llama-3.2-3B-Instruct3.2Q4190.9 est.Yes
gpt-oss-20b21.5Q4169.7 est.Yes
gemma-3-1b-it1.0FP16168.0 est.Yes
Qwen3-Coder-Next79.7Q4162.3 est.No
Phi-4-mini-instruct3.8Q4160.8 est.Yes
Phi-4-mini-reasoning3.8Q4160.8 est.Yes
gemma-3-4b-it4.3Q4156.6 est.Yes
Qwen3-Next-80B-A3B-Instruct81.3Q4156.2 est.No
Qwen3-4B-Instruct-25074.0Q4152.7 est.Yes
Qwen3-4B-Thinking-25074.0Q4152.7 est.Yes
Llama-3.2-1B-Instruct1.2FP16140.0 est.Yes
Mixtral-8x7B-Instruct-v0.146.7Q491.0 est.No
Qwen3-1.7B2.0FP1684.0 est.Yes
Qwen2.5-7B-Instruct7.6Q480.4 est.Yes
Qwen2.5-Coder-7B-Instruct7.6Q480.4 est.Yes
DeepSeek-R1-Distill-Qwen-7B7.6Q480.4 est.Yes
Qwen3-8B8.2Q474.5 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2Q474.5 est.Yes
Mistral-7B-Instruct-v0.37.2Q468.0 measured (gh-chharithoeun-llm-amd-windows)Yes
Llama-3.1-8B-Instruct8.0Q465.0 measured (gh-chharithoeun-llm-amd-windows)Yes
Qwen2.5-7B-Instruct7.6FP1656.0 measured (xinkanglabs-rocm-local-ai-stack)Yes
Llama-3.2-3B-Instruct3.2FP1652.5 est.Yes
gemma-3-12b-it12.2Q450.9 est.Yes
GLM-4.7-Flash31.2Q450.9 est.Yes
Qwen3-32B32.8Q446.8 est.No
Qwen2.5-32B-Instruct32.8Q446.8 est.No
Qwen2.5-Coder-32B-Instruct32.8Q446.8 est.No
DeepSeek-R1-Distill-Qwen-32B32.8Q446.8 est.No
Phi-4-mini-instruct3.8FP1644.2 est.Yes
Phi-4-mini-reasoning3.8FP1644.2 est.Yes
gemma-3-4b-it4.3FP1643.1 est.Yes
Qwen3-4B-Instruct-25074.0FP1642.0 est.Yes
Qwen3-4B-Thinking-25074.0FP1642.0 est.Yes
Phi-414.7Q441.6 est.Yes
Phi-4-reasoning14.7Q441.6 est.Yes
Qwen3-14B14.8Q441.3 est.Yes
Qwen2.5-14B-Instruct14.8Q441.3 est.Yes
Qwen2.5-Coder-14B-Instruct14.8Q441.3 est.Yes
DeepSeek-R1-Distill-Qwen-14B14.8Q441.3 est.Yes
gpt-oss-20b21.5FP1638.4 est.No
gemma-3-12b-it12.2FP1629.8 est.No
gpt-oss-120b120.4Q426.5 est.No
Devstral-Small-250723.6Q425.9 est.Yes
Mistral-Small-3.2-24B-Instruct-250624.0Q425.5 est.Yes
Magistral-Small-250924.0Q425.5 est.Yes
Qwen3-30B-A3B30.5FP1623.5 est.No
Qwen3-Coder-30B-A3B-Instruct30.5FP1623.5 est.No
Qwen3-30B-A3B-Instruct-250730.5FP1623.5 est.No
Mistral-7B-Instruct-v0.37.2FP1623.3 est.Yes
gemma-3-27b-it27.4Q422.5 est.Yes
Qwen2.5-Coder-7B-Instruct7.6FP1622.1 est.Yes
DeepSeek-R1-Distill-Qwen-7B7.6FP1622.1 est.Yes
Llama-3.1-8B-Instruct8.0FP1621.0 est.Yes
Qwen3-8B8.2FP1620.5 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2FP1620.5 est.Yes
GLM-4.5-Air110.5Q420.3 est.No
Phi-414.7FP1616.9 est.No
Phi-4-reasoning14.7FP1616.9 est.No
Qwen3-14B14.8FP1616.6 est.No
Qwen2.5-14B-Instruct14.8FP1616.6 est.No
Qwen2.5-Coder-14B-Instruct14.8FP1616.6 est.No
DeepSeek-R1-Distill-Qwen-14B14.8FP1616.6 est.No
DeepSeek-V4-Flash158.1Q49.4 est.No
GLM-4.7-Flash31.2FP165.3 est.No
Llama-3.1-70B-Instruct70.6Q45.0 est.No
Llama-3.3-70B-Instruct70.6Q45.0 est.No
DeepSeek-R1-Distill-Llama-70B70.6Q44.9 est.No
Qwen2.5-72B-Instruct72.7Q44.5 est.No
Devstral-Small-250723.6FP164.2 est.No
Mistral-Small-3.2-24B-Instruct-250624.0FP164.0 est.No
Magistral-Small-250924.0FP164.0 est.No
gemma-3-27b-it27.4FP162.7 est.No
Qwen3-32B32.8FP161.6 est.No
Qwen2.5-32B-Instruct32.8FP161.6 est.No
Qwen2.5-Coder-32B-Instruct32.8FP161.6 est.No
DeepSeek-R1-Distill-Qwen-32B32.8FP161.6 est.No