NVIDIA · workstation

RTX A1000

RTX A1000 has 8 GB of VRAM at 192 GB/s — about 7.44 GiB usable after driver and compositor overhead. 427 of 2118 indexed models fit at 128K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
8 GB
GDDR6
Bandwidth
192 GB/s
128-bit bus
Tensor FP16
27 TF
dense
TDP
50 W
$365 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 316vision language 43audio tts 18video 7image 1audio asr 29embedding 13

What fits at 128K context

largest quantization that fits, per model · 427 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
granite-3.1-1b-a400m-instructMoEQ2_K1.3B0.48 GiB6.00 GiB7.44 GiB0.00 GiB10±37%
FrickFritz-4BIQ4_XS4.7B2.42 GiB4.00 GiB7.43 GiB0.01 GiB17±22%
Newton-bot-3-VLM-mini-4BIQ4_XS4.7B2.42 GiB4.00 GiB7.43 GiB0.01 GiB17±22%
qwen3.5-4b-agentic-coder-v4IQ4_XS4.7B2.42 GiB4.00 GiB7.43 GiB0.01 GiB17±22%
Myth-4BIQ4_XS4.3B2.42 GiB4.00 GiB7.43 GiB0.01 GiB17±22%
Qwen3.5-4B-UncensoredIQ4_XS4.7B2.42 GiB4.00 GiB7.43 GiB0.01 GiB17±22%
JOSIE-2-4B-PreviewIQ4_XS4.7B2.42 GiB4.00 GiB7.43 GiB0.01 GiB17±22%
Surogate-3.5-4BIQ4_XS5.3B2.42 GiB4.00 GiB7.43 GiB0.01 GiB17±22%
Qwopus3.5-4B-v3IQ4_XS4.7B2.42 GiB4.00 GiB7.43 GiB0.01 GiB17±22%
Qwen3.5-4BQ3_K_M4.7B2.40 GiB4.00 GiB7.41 GiB0.03 GiB17±22%
Ace-Step1.5F32160M0.38 GiB6.04 GiB7.41 GiB0.03 GiB17±22%
Qwen3.5-4B-NSFW-ARA-Heretic-LiteroticaI1-Q4_K_S4.2B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Qwen3.5-4B-RpRMax-v1I1-Q4_K_S4.7B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Holo-3.1-4B-uncensored-hereticI1-Q4_K_S4.5B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
GRaPE-2-MiniI1-Q4_K_S4.7B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Qwen3.5-DPO-4B-2I1-Q4_K_S4.2B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Qwen3.5-4B-BaseQ4_K_S4.7B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliteratedI1-Q4_K_S4.7B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Qwopus3.5-4B-v3-hereticI1-Q4_K_S4.5B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Aureth-4B-Qwen3.5I1-Q4_K_S4.5B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Darwin-4B-ChimeraQ5_K_L4.0B2.86 GiB3.53 GiB7.40 GiB0.04 GiB17±22%
MiniCPM-V-4Q5_K_M4.1B2.39 GiB4.00 GiB7.40 GiB0.04 GiB17±22%
Gemma-4-E4B-LuchadorQ3_K_M8.0B4.56 GiB1.82 GiB7.40 GiB0.04 GiB17±22%
Wan2.2-Animate-14BQ2_K17.3B6.36 GiB0.00 GiB7.40 GiB0.04 GiB17±22%
Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKINGI1-Q4_K_S4.5B2.38 GiB4.00 GiB7.39 GiB0.05 GiB17±22%
Qwen3.5-4B-SOMPOA-heresy-v2I1-Q4_K_S4.5B2.38 GiB4.00 GiB7.39 GiB0.05 GiB17±22%
Qwen3.5-4B-SOMPOA-heresyI1-Q4_K_S4.5B2.38 GiB4.00 GiB7.39 GiB0.05 GiB17±22%
Qwen3.5-4B-Safety-ThinkingI1-Q4_K_S4.2B2.38 GiB4.00 GiB7.39 GiB0.05 GiB17±22%
Huihui-Qwen3.5-4B-abliteratedI1-Q4_K_S4.5B2.38 GiB4.00 GiB7.39 GiB0.05 GiB17±22%
Darkidol-Ballad-4BI1-Q4_K_S4.5B2.38 GiB4.00 GiB7.39 GiB0.05 GiB17±22%
Qwen3.5-4BQ4_K_S4.7B2.38 GiB4.00 GiB7.39 GiB0.05 GiB17±22%
LFM2-8B-A1BMoEQ4_18.3B4.89 GiB1.50 GiB7.38 GiB0.06 GiB26±37%
Qwen2.5-3BQ4_K_S3.1B1.87 GiB4.50 GiB7.38 GiB0.06 GiB17±22%
GRM-Kerlin-3bI1-Q4_K_S3.4B1.87 GiB4.50 GiB7.38 GiB0.06 GiB17±22%
Garnet-OCR-3B-0422I1-Q4_K_S4.1B1.87 GiB4.50 GiB7.38 GiB0.06 GiB17±22%
Dolphin3.0-Qwen2.5-1.5BF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2.5-1.5B-Instruct-abliteratedF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2.5-1.5B-VibeThinker-heretic-uncensored-abliteratedF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
NEXUS-Coder-OBLITERATEDF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
NEXUS-Coder-AbliteratedF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2.5-1.5B-hereticF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2.5-Math-1.5B-InstructBF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
ShellWhisperer-1.5BF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2.5-1.5BF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2.5-1.5B-InstructF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2.5-Coder-1.5B-InstructF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
PiCo-1BF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2-VL-2B-InstructF162.2B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
FableForge-1.5BF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2-1.5B-InstructBF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2-1.5BBF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
Qwen2.5-Coder-1.5BF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
NEXUS-MedicalF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
NEXUS-ScienceF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
NEXUS-LegalF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
NEXUS-CoderF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
NEXUS-FinanceF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
NEXUS-SecurityF161.5B2.88 GiB3.50 GiB7.38 GiB0.06 GiB17±22%
deepseek-coder-5.7bmqa-baseQ6_K5.7B4.36 GiB2.00 GiB7.38 GiB0.06 GiB17±22%
Fara1.5-4BIQ4_XS4.5B2.37 GiB4.00 GiB7.38 GiB0.06 GiB17±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.

Measured on this card

third-party benchmarks, aggregated
Workload◍MedianMiddle 50%Runs
Image generation3.75 it/s3.59–4.057
Benchmarked· n=7

Aggregated from community-submitted runs, so the spread is wide by nature — it covers different models, resolutions, step counts and settings, not one controlled configuration. Read the middle 50% rather than the median alone. These figures are reproduced with attribution from vladmandic-sd-data-benchmark, which publishes no licence — so we display and link rather than redistribute them.

Questions people ask

What AI models can a RTX A1000 run?
427 of 2118 indexed open-weight models fit a RTX A1000 at 131,072 context with f16 KV cache, the largest being granite-3.1-1b-a400m-instruct at Q2_K. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX A1000 actually have?
Its nameplate is 8 GB, but about 7.44 GiB is available to a model once driver and compositor overhead is accounted for.
Is a RTX A1000 fast for local AI?
Its memory bandwidth is 192 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.