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. 339 of 2118 indexed models fit at 64K context with q8_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 249vision language 29audio tts 17video 2embedding 13audio asr 29

What fits at 64K context

largest quantization that fits, per model · 339 of 2118 indexed
ModelBest quantParamsWeightsKVTotal in memoryHeadroomtok/s
GRM-Kerlin-3bI1-IQ3_M3.4B1.51 GiB1.20 GiB3.72 GiB0.00 GiB20±22%
LFM2.5-Audio-1.5B-JPF161.5B2.67 GiB0.00 GiB3.72 GiB0.00 GiB20±22%
Garnet-OCR-3B-0422I1-IQ3_M4.1B1.51 GiB1.20 GiB3.72 GiB0.00 GiB20±22%
umt5-xxlQ3_K_S5.7B2.66 GiB0.00 GiB3.71 GiB0.01 GiB21±22%
granite-4.0-h-microQ6_K3.2B2.44 GiB0.27 GiB3.71 GiB0.01 GiB20±22%
gemma-4-E2B-it-abliteratedI1-IQ3_XXS5.1B2.47 GiB0.25 GiB3.71 GiB0.01 GiB20±22%
gemma-4-E2B-it-qat-q4_0-unquantized-hereticI1-IQ3_XXS5.1B2.47 GiB0.25 GiB3.71 GiB0.01 GiB20±22%
Huihui-gemma-4-E2B-it-qat-q4_0-unquantized-abliteratedI1-IQ3_XXS5.1B2.47 GiB0.25 GiB3.71 GiB0.01 GiB20±22%
Gemma4_E2B_Abliterated_Baked_HF_ReadyI1-IQ3_XXS5.1B2.47 GiB0.25 GiB3.71 GiB0.01 GiB20±22%
gemma-4-E2B-itIQ3_XXS5.1B2.47 GiB0.25 GiB3.71 GiB0.01 GiB20±22%
EXAONE-4.0-1.2B-abliteratedI1-Q3_K_M1.5B0.73 GiB1.99 GiB3.70 GiB0.02 GiB20±22%
Qwen2.5-Omni-7BUD-IQ2_M10.7B2.66 GiB0.00 GiB3.70 GiB0.02 GiB21±22%
Holo-3.1-4BI1-IQ2_XXS5.2B1.63 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
AfriqueQwen3.5-4BI1-IQ2_XXS5.2B1.63 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
TimeOmni-1-4BI1-IQ2_XXS5.2B1.63 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
GLM-OCRI1-Q5_K_M1.3B0.59 GiB2.13 GiB3.70 GiB0.02 GiB20±22%
Qwen3.5-4B-NSFW-ARA-Heretic-LiteroticaI1-IQ2_M4.2B1.62 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
Qwen3.5-4B-RpRMax-v1I1-IQ2_M4.7B1.62 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
Holo-3.1-4B-uncensored-hereticI1-IQ2_M4.5B1.62 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
GRaPE-2-MiniI1-IQ2_M4.7B1.62 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
Qwen3.5-DPO-4B-2I1-IQ2_M4.2B1.62 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliteratedI1-IQ2_M4.7B1.62 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
Qwopus3.5-4B-v3-hereticI1-IQ2_M4.5B1.62 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
Aureth-4B-Qwen3.5I1-IQ2_M4.5B1.62 GiB1.06 GiB3.70 GiB0.02 GiB20±22%
DeepScaleR-1.5B-PreviewQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-Coder-1.5B-Instruct-abliteratedQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-1.5B-Instruct-uncensoredQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
VibeThinker-1.5BQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
Nemotron-Research-Reasoning-Qwen-1.5BQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
DeepSeek-R1-Distill-Qwen-1.5B-uncensoredQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
DeepSeek-R1-Distill-Qwen-1.5BQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-Coder-1.5B-InstructQ8_01.5B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-1.5B-InstructQ8_01.5B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
deepseek-r1-distill-qwen-1.5b-unsloth-bnb-4bitQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
dots.ocrQ8_03.0B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
DeepSeek-R1-Distill-Qwen-1.5B-Fully-UncensoredQ8_01.8B1.76 GiB0.93 GiB3.69 GiB0.03 GiB20±22%
granite-4.0-h-tinyMoEQ2_K6.9B2.46 GiB0.27 GiB3.69 GiB0.03 GiB49±37%
Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKINGI1-Q2_K_S4.5B1.62 GiB1.06 GiB3.69 GiB0.03 GiB20±22%
Qwen3.5-4B-SOMPOA-heresy-v2I1-Q2_K_S4.5B1.62 GiB1.06 GiB3.69 GiB0.03 GiB20±22%
Qwen3.5-4B-SOMPOA-heresyI1-Q2_K_S4.5B1.62 GiB1.06 GiB3.69 GiB0.03 GiB20±22%
Qwen3.5-4B-Safety-ThinkingI1-Q2_K_S4.2B1.62 GiB1.06 GiB3.69 GiB0.03 GiB20±22%
Huihui-Qwen3.5-4B-abliteratedI1-Q2_K_S4.5B1.62 GiB1.06 GiB3.69 GiB0.03 GiB20±22%
Darkidol-Ballad-4BI1-Q2_K_S4.5B1.62 GiB1.06 GiB3.69 GiB0.03 GiB20±22%
moondream2F161.9B2.64 GiB0.00 GiB3.69 GiB0.03 GiB21±22%
Dolphin3.0-Qwen2.5-3bQ3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-Coder-3B-Instruct-abliteratedI1-Q3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
GRM-Kerlin-3b-AbliteratedI1-Q3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Mythos-nanoI1-Q3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
MATE-3BI1-Q3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Mythos-nano-OBLITERATEDI1-Q3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-3B-Instruct-UncensoredI1-Q3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Nanonets-OCR-sQ3_K_M3.8B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-Coder-3BQ3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
VibeThinker-3B-OBLITERATEDI1-Q3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
raspberry-3BQ3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
VibeThinker-3BQ3_K_M3.1B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Fourier-Qwen2.5-VL-3B-0.67I1-Q3_K_M3.8B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-VL-3B-InstructQ3_K_M3.8B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
jina-embeddings-v4Q3_K_M3.8B1.48 GiB1.20 GiB3.69 GiB0.03 GiB20±22%
Qwen2.5-3BQ3_K_S3.1B1.48 GiB1.20 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?
339 of 2118 indexed open-weight models fit a RTX A400 at 65,536 context with q8_0 KV cache, the largest being GRM-Kerlin-3b at I1-IQ3_M. 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.