NVIDIA · workstation

RTX A400

RTX A400 has 4 GB of VRAM at 96 GB/s — about 3.72 GiB usable after driver and compositor overhead. 573 of 2118 indexed models fit at 32K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
4 GB
GDDR6
Bandwidth
96 GB/s
64-bit bus
Tensor FP16
11 TF
dense
TDP
50 W
$135 MSRP
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 449vision language 47audio tts 19audio asr 36video 2embedding 20

What fits at 32K context

largest quantization that fits, per model · 573 of 2118 indexed
ModelBest quantParamsWeightsKVTotal in memoryHeadroomtok/s
LFM2.5-Audio-1.5B-JPF161.5B2.67 GiB0.00 GiB3.72 GiB0.00 GiB20±22%
umt5-xxlQ3_K_S5.7B2.66 GiB0.00 GiB3.71 GiB0.01 GiB21±22%
medgemma-4b-itQ4_K_L4.3B2.47 GiB0.22 GiB3.71 GiB0.01 GiB20±22%
amoral-gemma3-4B-v1Q4_K_L4.3B2.47 GiB0.22 GiB3.71 GiB0.01 GiB20±22%
gemma-3-4b-it-abliteratedQ4_K_L4.3B2.47 GiB0.22 GiB3.71 GiB0.01 GiB20±22%
gemma-3-4b-itQ4_K_L4.3B2.47 GiB0.22 GiB3.71 GiB0.01 GiB20±22%
FrickFritz-4BIQ4_XS4.7B2.42 GiB0.28 GiB3.71 GiB0.01 GiB20±22%
Newton-bot-3-VLM-mini-4BIQ4_XS4.7B2.42 GiB0.28 GiB3.71 GiB0.01 GiB20±22%
qwen3.5-4b-agentic-coder-v4IQ4_XS4.7B2.42 GiB0.28 GiB3.71 GiB0.01 GiB20±22%
Myth-4BIQ4_XS4.3B2.42 GiB0.28 GiB3.71 GiB0.01 GiB20±22%
Qwen3.5-4B-UncensoredIQ4_XS4.7B2.42 GiB0.28 GiB3.71 GiB0.01 GiB20±22%
JOSIE-2-4B-PreviewIQ4_XS4.7B2.42 GiB0.28 GiB3.71 GiB0.01 GiB20±22%
Surogate-3.5-4BIQ4_XS5.3B2.42 GiB0.28 GiB3.71 GiB0.01 GiB20±22%
Qwopus3.5-4B-v3IQ4_XS4.7B2.42 GiB0.28 GiB3.71 GiB0.01 GiB20±22%
Llama-3.2-3B-Instruct-roleplay-tunedIQ4_XS3.2B1.71 GiB0.98 GiB3.71 GiB0.01 GiB20±22%
Llama-3.2-3B-Instruct-heretic-ablitered-uncensoredIQ4_XS3.2B1.71 GiB0.98 GiB3.71 GiB0.01 GiB20±22%
Llama3.2-3B-creative-writer-v0.1IQ4_XS3.2B1.71 GiB0.98 GiB3.71 GiB0.01 GiB20±22%
Firefly-V3.2IQ4_XS3.2B1.71 GiB0.98 GiB3.71 GiB0.01 GiB20±22%
Firefly-V3IQ4_XS3.2B1.71 GiB0.98 GiB3.71 GiB0.01 GiB20±22%
SmolLM3-3BQ5_K_M3.1B2.06 GiB0.63 GiB3.71 GiB0.01 GiB20±22%
Qwen2.5-Omni-7BUD-IQ2_M10.7B2.66 GiB0.00 GiB3.70 GiB0.02 GiB21±22%
Qwen3-4BUD-IQ2_M4.0B1.43 GiB1.27 GiB3.70 GiB0.02 GiB20±22%
Jan-nanoUD-IQ2_M4.0B1.43 GiB1.27 GiB3.70 GiB0.02 GiB20±22%
Phi-4-mini-instruct-abliteratedQ2_K3.8B1.57 GiB1.13 GiB3.70 GiB0.02 GiB20±22%
Phi-4-mini-reasoningQ2_K3.8B1.57 GiB1.13 GiB3.70 GiB0.02 GiB20±22%
Phi-4-mini-instructQ2_K3.8B1.57 GiB1.13 GiB3.70 GiB0.02 GiB20±22%
orpheus-3b-0.1-pretrainedIQ3_XS3.8B1.71 GiB0.98 GiB3.70 GiB0.02 GiB20±22%
Qwen3-VL-4B-ThinkingUD-IQ2_M4.4B1.43 GiB1.27 GiB3.70 GiB0.02 GiB20±22%
Qwen3-VL-4B-InstructUD-IQ2_M4.4B1.43 GiB1.27 GiB3.70 GiB0.02 GiB20±22%
Qwen3-4B-Thinking-2507UD-IQ2_M4.0B1.43 GiB1.27 GiB3.70 GiB0.02 GiB20±22%
Jan-nano-128kUD-IQ2_M4.0B1.43 GiB1.27 GiB3.70 GiB0.02 GiB20±22%
Qwen3-4B-Instruct-2507UD-IQ2_M4.0B1.43 GiB1.27 GiB3.70 GiB0.02 GiB20±22%
Yi-Coder-1.5B-ChatQ5_K_M1.5B1.02 GiB1.69 GiB3.70 GiB0.02 GiB20±22%
Yi-Coder-1.5BQ5_K_M1.5B1.02 GiB1.69 GiB3.70 GiB0.02 GiB20±22%
Dolphin3.0-Llama3.2-3BIQ4_XS3.2B1.70 GiB0.98 GiB3.70 GiB0.02 GiB20±22%
Llama-Doctor-3.2-3B-InstructI1-IQ4_XS3.2B1.70 GiB0.98 GiB3.70 GiB0.02 GiB20±22%
Llama-Song-Stream-3B-InstructIQ4_XS3.2B1.70 GiB0.98 GiB3.70 GiB0.02 GiB20±22%
llama-3.2-Korean-Bllossom-3BIQ4_XS3.2B1.70 GiB0.98 GiB3.70 GiB0.02 GiB20±22%
Llama-3.2-3B-InstructIQ4_XS3.2B1.70 GiB0.98 GiB3.70 GiB0.02 GiB20±22%
llama-3.2-3b-instructIQ4_XS3.2B1.70 GiB0.98 GiB3.70 GiB0.02 GiB20±22%
Hermes-3-Llama-3.2-3BIQ4_XS3.2B1.70 GiB0.98 GiB3.70 GiB0.02 GiB20±22%
Darwin-4B-ChimeraI1-Q4_K_M4.0B2.33 GiB0.36 GiB3.70 GiB0.02 GiB20±22%
LFM2-2.6BQ8_02.6B2.55 GiB0.14 GiB3.69 GiB0.03 GiB20±22%
LFM2-2.6B-TranscriptQ8_02.6B2.55 GiB0.14 GiB3.69 GiB0.03 GiB20±22%
LFM2-VL-3BQ8_03.0B2.55 GiB0.14 GiB3.69 GiB0.03 GiB20±22%
Gemma-3-4b-it-Uncensored-DBL-XI1-Q4_K_S4.7B2.41 GiB0.26 GiB3.69 GiB0.03 GiB20±22%
Dolphin3.0-Qwen2.5-3bQ6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-Coder-3B-Instruct-abliteratedI1-Q6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
GRM-Kerlin-3b-AbliteratedI1-Q6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
Mythos-nanoI1-Q6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
MATE-3BI1-Q6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
Mythos-nano-OBLITERATEDI1-Q6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-3B-Instruct-UncensoredI1-Q6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
Nanonets-OCR-sQ6_K3.8B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-Coder-3BQ6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
raspberry-3BQ6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
VibeThinker-3B-OBLITERATEDI1-Q6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
VibeThinker-3BQ6_K3.1B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
Fourier-Qwen2.5-VL-3B-0.67I1-Q6_K3.8B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-VL-3B-InstructQ6_K3.8B2.36 GiB0.32 GiB3.69 GiB0.03 GiB20±22%
From the filePredictedwhat 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 RTX A400 run?
573 of 2118 indexed open-weight models fit a RTX A400 at 32,768 context with q4_0 KV cache, the largest being LFM2.5-Audio-1.5B-JP at F16. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX A400 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 RTX A400 fast for local AI?
Its memory bandwidth is 96 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.