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. 734 of 2118 indexed models fit at 16K 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 602audio asr 36audio tts 19vision language 50embedding 25video 2

What fits at 16K context

largest quantization that fits, per model · 734 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
rnj-1-instructUD-IQ1_M8.3B2.11 GiB0.56 GiB3.72 GiB0.00 GiB20±22%
granite-speech-4.1-2bQ8_02.3B2.38 GiB0.35 GiB3.72 GiB0.00 GiB20±22%
granite-4.0-1b-speechQ8_02.3B2.38 GiB0.35 GiB3.72 GiB0.00 GiB20±22%
zeta-2.1I1-IQ1_M8.3B2.12 GiB0.56 GiB3.72 GiB0.00 GiB20±22%
LFM2.5-Audio-1.5B-JPF161.5B2.67 GiB0.00 GiB3.72 GiB0.00 GiB20±22%
DeepHat-V1-7B-Heretic-AbliteratedI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
ShizhenGPT-7B-VLI1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
HuatuoGPT-o1-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
MathSmith-DS-Qwen-7B-LongCoTI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
AstraGPTCoder-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-Coder-7B-Instruct-Ghidra-v2I1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
EsDrac-v1-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Hemlock-Apothecary-7B-GRPO-e3I1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
openhands-lm-7b-v0.1I1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Hemlock2-Coder-7B-GRPOI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
shellwhiz-7bI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-Coder-7B-Instruct-abliteratedI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-Coder-7B-Instruct-OBLITERATED-advancedI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen-STEM-Specialist-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
VulnLLM-R-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Garnet-OCR-7B-0422I1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
UwU-7B-InstructI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Video-R1-7BI1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
HARC-Qwen2.5-7B-InstructI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-Coder-7B-AbliteratedI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Bozdogan-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Crazy-AI-ModelI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
turbo-ai-7bI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
DeepSeek-R1-Distill-Qwen-7B-abliterated-v2I1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Ghosty-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
SP-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-Coder-7B-Instruct-UncensoredI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-VL-7B-Instruct-abliteratedI1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
DeepSeek-R1-Distill-Qwen-8B-AbliteratedI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-7BIQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-VL-7B-Instruct-hereticI1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
AWARES-Qwen2.5-VL-7BI1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
olmOCR-2-7B-1025I1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2.5-7B-Instruct-UncensoredI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Med-RwRI1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
EVA-Qwen2.5-7B-v0.1I1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
SpatialThinker-7BI1-IQ2_S8.3B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Human-Like-Qwen2.5-7B-InstructI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Qwen2-7B-InstructIQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Hercules-5.0-Qwen2-7BIQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Darwin-4B-ChimeraQ4_14.0B2.45 GiB0.25 GiB3.72 GiB0.00 GiB20±22%
orpheus-3b-0.1-pretrainedQ3_K_L3.8B2.22 GiB0.49 GiB3.72 GiB0.00 GiB20±22%
Yi-6B-ChatI1-IQ3_XS6.1B2.41 GiB0.28 GiB3.72 GiB0.00 GiB20±22%
MQ-Coldbrew-BaseI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Kepler-Reasoning-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
DeepSeek-R1-Distill-Qwen-7B-Uncensored-ReasonerI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
DeepSeek-R1-Distill-Qwen-7B-UncensoredI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
DeepSeek-R1-STEM-Coder-7BI1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
Kepler-8B-Instruct-v2I1-IQ2_S7.6B2.42 GiB0.25 GiB3.72 GiB0.00 GiB21±22%
granite-4.0-7B-A1B-Creative-v0.1MoEI1-IQ3_S6.7B2.71 GiB0.04 GiB3.71 GiB0.01 GiB68±37%
G9v3-3BQ6_K_L3.0B2.49 GiB0.23 GiB3.71 GiB0.01 GiB20±22%
Hubble-4B-v1Q3_K_M4.5B2.14 GiB0.56 GiB3.71 GiB0.01 GiB20±22%
Aura-4BI1-Q3_K_M4.5B2.14 GiB0.56 GiB3.71 GiB0.01 GiB20±22%
magnum-v2-4bI1-Q3_K_M4.5B2.14 GiB0.56 GiB3.71 GiB0.01 GiB20±22%
Impish_LLAMA_4BQ3_K_M4.5B2.14 GiB0.56 GiB3.71 GiB0.01 GiB20±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?
734 of 2118 indexed open-weight models fit a RTX A400 at 16,384 context with q4_0 KV cache, the largest being rnj-1-instruct at UD-IQ1_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.