NVIDIA · workstation

RTX A2000

RTX A2000 has 6 GB of VRAM at 288 GB/s — about 5.58 GiB usable after driver and compositor overhead. 1174 of 2118 indexed models fit at 4K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
6 GB
GDDR6
Bandwidth
288 GB/s
192-bit bus
Tensor FP16
32 TF
dense
TDP
70 W
$449 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1000vision language 88embedding 26audio asr 38image 1audio tts 19video 2

What fits at 4K context

largest quantization that fits, per model · 1174 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Gemma-4-E4B-LuchadorIQ3_M8.0B4.44 GiB0.12 GiB5.58 GiB0.00 GiB36±22%
Tini-Cybersec-8B-A1BMoEQ4_08.5B4.54 GiB0.05 GiB5.58 GiB0.00 GiB104±37%
MiMo-VL-7B-RLIQ4_XS8.3B3.99 GiB0.56 GiB5.58 GiB0.00 GiB36±22%
Mistral-NeMo-Minitron-8B-InstructQ3_K_M8.4B3.92 GiB0.63 GiB5.57 GiB0.01 GiB36±22%
LFM2.5-Queen-Opus-4.7-8B-A1BMoEI1-Q4_K_S8.5B4.53 GiB0.05 GiB5.57 GiB0.01 GiB104±37%
LFM2.5-8B-A1B-KO-SFTMoEI1-Q4_K_S8.5B4.53 GiB0.05 GiB5.57 GiB0.01 GiB104±37%
LFM2.5-8B-A1B-hereticMoEI1-Q4_K_S8.5B4.53 GiB0.05 GiB5.57 GiB0.01 GiB104±37%
LFM2.5-8B-A1B-SOMPOA-heresyMoEI1-Q4_K_S8.5B4.53 GiB0.05 GiB5.57 GiB0.01 GiB104±37%
Huihui-LFM2.5-8B-A1B-abliteratedMoEI1-Q4_K_S8.5B4.53 GiB0.05 GiB5.57 GiB0.01 GiB104±37%
Supertron2.1-8B-A1BMoEI1-Q4_K_S8.5B4.53 GiB0.05 GiB5.57 GiB0.01 GiB104±37%
Qwen3.5-9B-CoderI1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Qwopus3.5-9B-v3.5Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Qwythos-9B-Claude-Mythos-5-1M-MTPI1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliteratedI1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Qwen3.5-9B-Fable-5-v1I1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Qwythos-9B-v2I1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
PINQWEN-3.5-9B-1M-BF16I1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Openprose-2-FlashI1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Qwen3.5-9B-Nikusui-v1I1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Ornstein-3.5-9B-V1.5I1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Ornith-1.0-9B-heretic-MTPI1-Q3_K_M9.4B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Tess-4-9BI1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
dotwebs-1I1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
liftQ3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Hemlock-Qwopus3.5-9B-CoderI1-Q3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Qwen3.5-9B-DeepSeek-V4-FlashQ3_K_M9.7B4.41 GiB0.13 GiB5.57 GiB0.01 GiB36±22%
Assistant_Pepe_8BQ3_K_L—4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
DeepSeek-R1-Distill-Llama-8B-AbliteratedI1-IQ1_M8.0B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
Foundation-Sec-8B-InstructI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
Foundation-Sec-8B-Instruct-hereticI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
dolphin-2.9-llama3-8bQ3_K_L8.0B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
saiga_llama3_8bQ3_K_L8.0B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
Meta-Llama-3-8B-InstructQ3_K_L8.0B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
MiniCPM-Llama3-V-2_5Q3_K_L8.5B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
Llama3-ChatQA-1.5-8BQ3_K_L8.0B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
Qwen3-VL-Embedding-8BIQ4_XS8.1B3.97 GiB0.56 GiB5.57 GiB0.01 GiB36±22%
Meta-Llama-3-8BQ3_K_L8.0B4.03 GiB0.50 GiB5.57 GiB0.01 GiB36±22%
Llama-3.1-Tulu-3-8BQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Llama-3-Groq-8B-Tool-UseQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
llama3.1-heretic-unsensoredI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Dolphin3.0-Llama3.1-8BQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
dolphin-2.9.4-llama3.1-8bQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Dolphin3.0-Llama3.1-8B-abliteratedQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Anubis-Mini-8B-v1I1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Deepseek-R1-Distill-NSFW-RPv1Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
LLAMA-3_8B_Unaligned_BETAQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Llama-3.1-8BQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Meta-Llama-3-8BQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8B-hereticI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8BI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Gluon-8BI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-ReasoningI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Llama3.3-8B-Instruct-Thinking-Claude-4.5-Opus-High-ReasoningI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Hypnos-i1-8BQ3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
grok-oss-Apollyon-8B-hereticI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Llama-3.3-8B-Instruct-128K-JbliteratedI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
PsyCoPref-Llama3-8BI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
Llama3.1-GptDeluxe-8BI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
SelfCite-8B-CC-SFTI1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±22%
ai-girlfriend-v2I1-Q3_K_L8.0B4.03 GiB0.50 GiB5.56 GiB0.02 GiB36±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 A2000 run?
1174 of 2118 indexed open-weight models fit a RTX A2000 at 4,096 context with f16 KV cache, the largest being Gemma-4-E4B-Luchador at IQ3_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX A2000 actually have?
Its nameplate is 6 GB, but about 5.58 GiB is available to a model once driver and compositor overhead is accounted for.
Is a RTX A2000 fast for local AI?
Its memory bandwidth is 288 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.