AMD · workstation

Radeon Pro W6400

Radeon Pro W6400 has 4 GB of VRAM at 128 GB/s — about 3.72 GiB usable after driver and compositor overhead. 824 of 2118 indexed models fit at 8K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
4 GB
GDDR6
Bandwidth
128 GB/s
64-bit bus
Tensor FP16
—
dense
TDP
50 W
$229 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
embedding 25vision language 58text 682image 1audio tts 19audio asr 37video 2

What fits at 8K context

largest quantization that fits, per model · 824 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
nomic-embed-codeQ2_K7.1B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
OvisOCR2F32853M2.81 GiB0.03 GiB3.72 GiB0.00 GiB28±26.5%
NVIDIA-Nemotron-3-Nano-4B-BF16Q4_04.0B2.42 GiB0.37 GiB3.72 GiB0.00 GiB29±26.5%
DeepHat-V1-7B-Heretic-AbliteratedI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
ShizhenGPT-7B-VLI1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
HuatuoGPT-o1-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
MathSmith-DS-Qwen-7B-LongCoTI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
AstraGPTCoder-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Qwen2.5-Coder-7B-Instruct-Ghidra-v2I1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
EsDrac-v1-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Hemlock-Apothecary-7B-GRPO-e3I1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
openhands-lm-7b-v0.1I1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Hemlock2-Coder-7B-GRPOI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
shellwhiz-7bI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Qwen2.5-Coder-7B-Instruct-OBLITERATED-advancedI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Qwen-STEM-Specialist-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
VulnLLM-R-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Garnet-OCR-7B-0422I1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
UwU-7B-InstructI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Video-R1-7BI1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
HARC-Qwen2.5-7B-InstructI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Qwen2.5-Coder-7B-AbliteratedI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Bozdogan-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Crazy-AI-ModelI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
turbo-ai-7bI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
DeepSeek-R1-Distill-Qwen-7B-abliterated-v2I1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Ghosty-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
SP-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Qwen2.5-Coder-7B-Instruct-UncensoredI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Qwen2.5-VL-7B-Instruct-abliteratedI1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
DeepSeek-R1-Distill-Qwen-8B-AbliteratedI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Qwen2.5-7BQ2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Qwen2.5-VL-7B-Instruct-hereticI1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
AWARES-Qwen2.5-VL-7BI1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
olmOCR-2-7B-1025I1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Med-RwRI1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
SpatialThinker-7BI1-Q2_K_S8.3B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Nexa-AI-4x4B-InstructMoEI1-IQ1_S12.1B2.49 GiB0.32 GiB3.72 GiB0.00 GiB28±37%
MQ-Coldbrew-BaseI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Kepler-Reasoning-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
DeepSeek-R1-Distill-Qwen-7B-Uncensored-ReasonerI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
DeepSeek-R1-Distill-Qwen-7B-UncensoredI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
DeepSeek-R1-STEM-Coder-7BI1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
Kepler-8B-Instruct-v2I1-Q2_K_S7.6B2.64 GiB0.12 GiB3.72 GiB0.00 GiB29±26.5%
gemma-3n-E2B-itQ4_K_S5.4B2.77 GiB0.04 GiB3.71 GiB0.01 GiB28±26.5%
Fimbulvetr-11B-v2I1-IQ1_S10.7B2.35 GiB0.42 GiB3.71 GiB0.01 GiB29±26.5%
Qwen2.5-3B-Instruct-abliteratedI1-IQ3_XS3.1B2.72 GiB0.08 GiB3.71 GiB0.01 GiB29±26.5%
Dolphin3.0-Llama3.2-3BQ6_K_L3.2B2.55 GiB0.25 GiB3.71 GiB0.01 GiB29±26.5%
Llama-Song-Stream-3B-InstructQ6_K_L3.2B2.55 GiB0.25 GiB3.71 GiB0.01 GiB29±26.5%
Llama-Doctor-3.2-3B-InstructQ6_K_L3.2B2.55 GiB0.25 GiB3.71 GiB0.01 GiB29±26.5%
llama-3.2-Korean-Bllossom-3BQ6_K_L3.2B2.55 GiB0.25 GiB3.71 GiB0.01 GiB29±26.5%
Llama-3.2-3B-InstructQ6_K_L3.2B2.55 GiB0.25 GiB3.71 GiB0.01 GiB29±26.5%
Hermes-3-Llama-3.2-3BQ6_K_L3.2B2.55 GiB0.25 GiB3.71 GiB0.01 GiB29±26.5%
Holo-3.1-4BI1-Q4_K_S5.2B2.72 GiB0.07 GiB3.70 GiB0.02 GiB29±26.5%
AfriqueQwen3.5-4BI1-Q4_K_S5.2B2.72 GiB0.07 GiB3.70 GiB0.02 GiB29±26.5%
TimeOmni-1-4BI1-Q4_K_S5.2B2.72 GiB0.07 GiB3.70 GiB0.02 GiB29±26.5%
ToriiGate-0.5Q4_K_S5.2B2.72 GiB0.07 GiB3.70 GiB0.02 GiB29±26.5%
chandra-ocr-2Q4_K_S5.3B2.72 GiB0.07 GiB3.70 GiB0.02 GiB29±26.5%
Janus-Pro-7BI1-IQ1_M7.4B1.72 GiB1.05 GiB3.70 GiB0.02 GiB29±26.5%
deepseek-coder-7b-instruct-v1.5I1-IQ1_M6.9B1.72 GiB1.05 GiB3.70 GiB0.02 GiB29±26.5%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a Radeon Pro W6400 run?
824 of 2118 indexed open-weight models fit a Radeon Pro W6400 at 8,192 context with q4_0 KV cache, the largest being nomic-embed-code at Q2_K. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon Pro W6400 actually have?
Its nameplate is 4 GB, but about 3.72 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Radeon Pro W6400 fast for local AI?
Its memory bandwidth is 128 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.