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. 419 of 2118 indexed models fit at 64K 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
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 307vision language 42audio tts 17audio asr 31video 2embedding 20

What fits at 64K context

largest quantization that fits, per model · 419 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
LFM2.5-Audio-1.5B-JPF161.5B2.67 GiB0.00 GiB3.72 GiB0.00 GiB20±22%
Dolphin3.0-Qwen2.5-3bQ5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Qwen2.5-Coder-3B-Instruct-abliteratedI1-Q5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
GRM-Kerlin-3b-AbliteratedI1-Q5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Mythos-nanoI1-Q5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
MATE-3BI1-Q5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Mythos-nano-OBLITERATEDI1-Q5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Qwen2.5-3B-Instruct-UncensoredI1-Q5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Nanonets-OCR-sQ5_K_M3.8B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Qwen2.5-Coder-3BQ5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
raspberry-3BQ5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
VibeThinker-3B-OBLITERATEDI1-Q5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
VibeThinker-3BQ5_K_M3.1B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Fourier-Qwen2.5-VL-3B-0.67I1-Q5_K_M3.8B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Qwen2.5-VL-3B-InstructQ5_K_M3.8B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
jina-embeddings-v4Q5_K_M3.8B2.07 GiB0.63 GiB3.72 GiB0.00 GiB20±22%
Felldude-Uncensored-Ministral3-3B-bf16I1-IQ1_S3.8B0.87 GiB1.83 GiB3.71 GiB0.01 GiB20±22%
Amaretto-3BI1-IQ1_S4.3B0.87 GiB1.83 GiB3.71 GiB0.01 GiB20±22%
OpenClaude-1.7B-MergedQ2_K1.7B0.75 GiB1.97 GiB3.71 GiB0.01 GiB20±22%
umt5-xxlQ3_K_S5.7B2.66 GiB0.00 GiB3.71 GiB0.01 GiB21±22%
Vikhr-Gemma-2B-instructQ4_K_L2.6B1.72 GiB0.98 GiB3.71 GiB0.01 GiB20±22%
gemma-2-2b-it-abliteratedQ4_K_L2.6B1.72 GiB0.98 GiB3.71 GiB0.01 GiB20±22%
Gemmasutra-Mini-2B-v1Q4_K_L2.6B1.72 GiB0.98 GiB3.71 GiB0.01 GiB20±22%
granite-3.1-3b-a800m-instructMoEQ3_K_L3.3B1.62 GiB1.13 GiB3.71 GiB0.01 GiB20±37%
bitnet_b1_58-largeTQ2_0729M0.20 GiB2.53 GiB3.71 GiB0.01 GiB20±22%
MiniCPM-V-4Q4_14.1B2.14 GiB0.56 GiB3.71 GiB0.01 GiB20±22%
Qwen2.5-Omni-7BUD-IQ2_M10.7B2.66 GiB0.00 GiB3.70 GiB0.02 GiB21±22%
Parable-Granite-4.1-3B-Claude-Fable-5I1-IQ3_XXS3.4B1.29 GiB1.41 GiB3.70 GiB0.02 GiB20±22%
granite-4.0-microIQ3_XXS3.4B1.29 GiB1.41 GiB3.70 GiB0.02 GiB20±22%
Darwin-4B-ChimeraI1-IQ4_XS4.0B2.11 GiB0.57 GiB3.70 GiB0.02 GiB20±22%
Fara1.5-4BIQ3_XS4.5B2.12 GiB0.56 GiB3.69 GiB0.03 GiB20±22%
AREX-TurboIQ3_XS4.5B2.12 GiB0.56 GiB3.69 GiB0.03 GiB20±22%
granite-3.3-2b-instructIQ4_XS2.5B1.29 GiB1.41 GiB3.69 GiB0.03 GiB20±22%
granite-3.1-2b-instructIQ4_XS2.5B1.29 GiB1.41 GiB3.69 GiB0.03 GiB20±22%
granite-3.2-2b-instructIQ4_XS2.5B1.29 GiB1.41 GiB3.69 GiB0.03 GiB20±22%
granite-vision-3.2-2bIQ4_XS3.0B1.29 GiB1.41 GiB3.69 GiB0.03 GiB20±22%
Qwen3-0.6BQ8_0752M0.75 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Qwen3-0.6B-BaseQ8_0596M0.75 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Qwen3-0.6B-Distilled-30B-A3B-Thinking-SFTQ8_0752M0.75 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
moondream2F161.9B2.64 GiB0.00 GiB3.69 GiB0.03 GiB21±22%
granite-4.0-7B-A1B-Creative-v0.1MoEI1-IQ3_XS6.7B2.58 GiB0.14 GiB3.69 GiB0.03 GiB58±37%
granite-4.1-3bQ2_K3.4B1.28 GiB1.41 GiB3.69 GiB0.03 GiB20±22%
granite-4.0-micro-baseQ2_K3.4B1.28 GiB1.41 GiB3.69 GiB0.03 GiB20±22%
Supertron2-Reranker-2BI1-Q2_K2.1B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Uni-MuMER-Qwen3-VL-2BI1-Q2_K2.1B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Qwen3-VL-2B-ThinkingQ2_K2.1B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Qwen3-VL-Reranker-2BI1-Q2_K2.1B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Qwen3-VL-2B-InstructQ2_K2.1B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Qwen3-VL-Embedding-2BQ2_K2.1B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
OpenCaption-2B-VL-SFT-v1.0I1-Q2_K2.1B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Atomight-V2.5-1.7BI1-Q2_K1.7B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
gaon-1.7b-v2-translateI1-Q2_K1.7B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
gaon-1.7b-v2-instructI1-Q2_K1.7B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Lightning-1.7BQ2_K1.7B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
DorsetHeatwaveLLM2I1-Q2_K1.7B0.72 GiB1.97 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-3BQ4_13.1B2.04 GiB0.63 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-3B-Instruct-abliteratedI1-Q4_13.1B2.04 GiB0.63 GiB3.69 GiB0.03 GiB20±22%
GRM-Kerlin-3bI1-Q4_13.4B2.04 GiB0.63 GiB3.69 GiB0.03 GiB20±22%
Garnet-OCR-3B-0422I1-Q4_14.1B2.04 GiB0.63 GiB3.68 GiB0.04 GiB20±22%
Qwen3.5-4B-NSFW-ARA-Heretic-LiteroticaI1-Q3_K_M4.2B2.11 GiB0.56 GiB3.68 GiB0.04 GiB21±22%
●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 RTX A400 run?
419 of 2118 indexed open-weight models fit a RTX A400 at 65,536 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.