DiggTop
Menu
⌕

VRAM Calculator

What can your GPU actually run? Pick a card's memory and a quantization — the table shows which local models fit, and by how much.

113 fit comfortably · 23 tight on 24 GB at Q4_K_M (4-bit)

ModelParamsEst. VRAMVerdict
MobileLLM-125M125M0.9 GBFits
SmolLM2-135M-Instruct135M0.9 GBFits
LFM2.5-230M230M0.9 GBFits
LFM2-350M350M1.0 GBFits
MobileLLM-350M350M1.0 GBFits
LFM2.5-350M350M1.0 GBFits
SmolLM2-360M-Instruct360M1.0 GBFits
Qwen2.5-0.5B-Instruct500M1.1 GBFits
Qwen3.5-0.8B-MTP800M1.3 GBFits
Qwen3.5-0.8B800M1.3 GBFits
gemma-3-1b-it1B1.4 GBFits
Llama-3.2-1B-Instruct1B1.4 GBFits
LFM2.5-1.2B-Instruct1B1.4 GBFits
Qwen3-0.6B1B1.4 GBFits
MiniCPM5-1B1B1.4 GBFits
Spark-X2.5-1.7B2B2.1 GBFits
Qwen3-1.7B2B2.1 GBFits
LFM2.5-2.6B3B2.7 GBFits
Llama-3.2-3B-Instruct3B2.7 GBFits
VibeThinker-3B3B2.7 GBFits
Ministral-3-3B-Instruct-25123B2.7 GBFits
OpenLLaMA-3B3B2.7 GBFits
Qwen3.5-4B-MTP4B3.4 GBFits
Qwen3.8-4B-Distill4B3.4 GBFits
Spark-X2.5-4B4B3.4 GBFits
NVIDIA-Nemotron-3-Nano-4B4B3.4 GBFits
Qwen3.5-4B4B3.4 GBFits
Qwen3-4B-Thinking-25074B3.4 GBFits
gemma-3n-E4B-it-text4B3.4 GBFits
Qwen3-4B4B3.4 GBFits
Phi-4-mini-reasoning4B3.4 GBFits
gemma-4-E2B-it5B4.0 GBFits
Falcon3-7B-Instruct7B5.3 GBFits
AgentTest-7B7B5.3 GBFits
Ling-3.0-tiny8B6.0 GBFits
Qwen3-8B8B6.0 GBFits
granite-4.2-8b8B6.0 GBFits
Qwen2.5-VL-7B-Instruct8B6.0 GBFits
DeepSeek-R1-0528-Qwen3-8B8B6.0 GBFits
Llama-3.1-8B-Instruct8B6.0 GBFits
LFM2.5-8B-A1B8B6.0 GBFits
OpenReasoning-Nemotron-7B8B6.0 GBFits
InternVL3-8B8B6.0 GBFits
gemma-4-E4B-it8B6.0 GBFits
Meta-Llama-3-8B-Instruct8B6.0 GBFits
DeepSeek-R1-Distill-Qwen-7B8B6.0 GBFits
AceReason-Nemotron-1.1-7B8B6.0 GBFits
LFM2-8B-A1B8B6.0 GBFits
WebWorld-14B8B6.0 GBFits
WebWorld-8B8B6.0 GBFits
Qwen3.5-9B-MTP9B6.6 GBFits
Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated-i19B6.6 GBFits
Qwen3.8-9B-Distill9B6.6 GBFits
Qwythos-9B-v29B6.6 GBFits
DeepSeek-V4-Pro-Qwen3.5-9B-MTP9B6.6 GBFits
Qwen3-VL-8B-Instruct9B6.6 GBFits
Ornith-1.0-9B9B6.6 GBFits
NVIDIA-Nemotron-Nano-9B-v29B6.6 GBFits
Carnice-9b9B6.6 GBFits
Qwen3.5-9B10B7.3 GBFits
Ornith-1.5-9B10B7.3 GBFits
GLM-4.6V-Flash10B7.3 GBFits
gemma-4-12B-it-qat12B8.6 GBFits
gemma-4-12B-coder-fable5-composer2.5-v112B8.6 GBFits
gemma-3-12b-it12B8.6 GBFits
gemma-4-12B-it12B8.6 GBFits
NVIDIA-Nemotron-Nano-12B-v212B8.6 GBFits
Ministral-3-14B-Reasoning-251214B9.9 GBFits
InternVL3-14B14B9.9 GBFits
Qwen3-14B14B9.9 GBFits
phi-415B10.5 GBFits
Phi-4-reasoning-plus15B10.5 GBFits
Qwen2.5-Coder-14B-Instruct15B10.5 GBFits
Phi-4-reasoning15B10.5 GBFits
DeepSeek-R1-Distill-Qwen-14B15B10.5 GBFits
Nemotron-Cascade-14B-Thinking15B10.5 GBFits
DeepSeek-Coder-V2-Lite-Instruct16B11.2 GBFits
gpt-oss-20b21B14.4 GBFits
reka-flash-321B14.4 GBFits
mistralai_Mistral-Small-3.2-24B-Instruct-250624B16.3 GBFits
LFM2-24B-A2B24B16.3 GBFits
DeepHermes-3-Mistral-24B-Preview24B16.3 GBFits
Trinity-Mini26B17.6 GBFits
Qwen3.8-27B-MTP27B18.3 GBFits
gemma-4-26B-A4B-it-ultra-uncensored-heretic27B18.3 GBFits
gemma-4-26B-A4B-it27B18.3 GBFits
Gemma4-26B-A4B-QAT-Uncensored-HauhauCS-Balanced-MTP27B18.3 GBFits
gemma-4-26B-A4B-it-qat27B18.3 GBFits
Ornstein3.6-27B-MTP-NSC-ACE-SABER27B18.3 GBFits
Underdog-Saluki-27B27B18.3 GBFits
Ternary Bonsai 2 27B27B18.3 GBFits
Qwen3.8-27B28B18.9 GBFits
Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP28B18.9 GBFits
Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-YMQ28B18.9 GBFits
Qwen3.6-27B28B18.9 GBFits
Qwen3.8-27B-Uncensored28B18.9 GBFits
Qwen3.8-27B-Fable-Distill28B18.9 GBFits
Huihui-Qwen3.8-27B-abliterated28B18.9 GBFits
Bonsai-27B28B18.9 GBFits
Qwen3.8-27B-5090-goldilocks28B18.9 GBFits
Qwen3.8-27B-Ridge28B18.9 GBFits
Godoter-27B28B18.9 GBFits
Qwen3.6-27B-i128B18.9 GBFits
Qwen3.6-27B-MTP28B18.9 GBFits
ThinkingCap-Qwen3.6-27B28B18.9 GBFits
Qwen3.5-27B28B18.9 GBFits
Tess-4-27B28B18.9 GBFits
granite-4.2-30b29B19.6 GBFits
Hy-MT2-30B-A3B30B20.2 GBFits
Nemotron-3-Nano-Omni-30B-A3B-Reasoning30B20.2 GBFits
Qwen3-Coder-30B-A3B-Instruct30B20.2 GBFits
Muse-Glimmer-30B30B20.2 GBFits
North-Mini-Code-1.030B20.2 GBFits
Qwen3-30B-A3B-Thinking-250731B20.9 GBTight
QUEST-30B-RL-B70-Turbo31B20.9 GBTight
Qwen3-30B-A3B-Instruct-250731B20.9 GBTight
NVIDIA-Nemotron-3.5-Lightning-30B-A3B32B21.5 GBTight
NVIDIA-Nemotron-3-Nano-30B-A3B32B21.5 GBTight
Olmo-3-32B-Think32B21.5 GBTight
Nemotron-Cascade-2-30B-A3B32B21.5 GBTight
Laguna-XS-2.133B22.1 GBTight
gemma-4-31B-it33B22.1 GBTight
GLM-Z1-32B-041433B22.1 GBTight
Apodex-1.1-mini35B23.4 GBTight
Qwen3.6-35B-A3B-MTP35B23.4 GBTight
Qwen3.5-35B-A3B35B23.4 GBTight
KAT-Coder-V2.5-Dev-AWQ-W4A16-ASYM35B23.4 GBTight
Kwaipilot_KAT-Coder-V2.5-Dev35B23.4 GBTight
BTL-4-Compact35B23.4 GBTight
badtheorylabs_BTL-435B23.4 GBTight
Ornith-1.0-35B-PrismaAURA-4.75bit-vllm-MTP35B23.4 GBTight
Ornith-1.0-35B-B70-Turbo35B23.4 GBTight
Nex-N2-mini-B70-Turbo35B23.4 GBTight
Ornith-1.0-35B35B23.4 GBTight
Qwen-AgentWorld-35B-A3B-B70-Turbo35B23.4 GBTight
MiniCPM-V 4.7-35B-A3B35B23.4 GBTight
Ornith-1.5-35B-A3B-MTP36B24.1 GBToo big
Ornith-1.5-35B-A3B36B24.1 GBToo big
Unsloth-Ornith-1.5-35B-A3B36B24.1 GBToo big
Ornith-1.5-35B-A3B-GPTQ-Int4-sym-G128-MTP-BF16-MixedCal-v236B24.1 GBToo big
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved36B24.1 GBToo big
Qwen3.6-35B-A3B36B24.1 GBToo big
SIQ-1-35B-B70-Turbo36B24.1 GBToo big
Qwen3.6-35B-A3B-abliterated-v436B24.1 GBToo big
Kimi-Linear-48B-A3B-Instruct49B32.5 GBToo big
Llama-3.3-70B-Instruct71B46.7 GBToo big
Qwen3-Coder-Next80B52.5 GBToo big
Llama-4-Scout-17B-16E-Instruct109B71.3 GBToo big
gpt-oss-120b117B76.5 GBToo big
Laguna-S-2.1118B77.1 GBToo big
NVIDIA-Nemotron-3.5-Super-VL-120B-A12B120B78.4 GBToo big
Qwen 3.8 Flash125B81.7 GBToo big
Qwen 3.8 Flash-Next Base125B81.7 GBToo big
Ling-3.0-flash127B83.0 GBToo big
DeepSeek-V4-Flash-180B158B103.0 GBToo big
Qwen3.8-Flash-Next180B117.2 GBToo big
Qwen3.8-Flash-Next-REAP-256-duo180B117.2 GBToo big
Qwen3.8-Flash-Next-hibrid48180B117.2 GBToo big
CYBER-FROST-3.8180B117.2 GBToo big
MiniMax-M2.7229B148.9 GBToo big
Baichuan-M3-235B235B152.8 GBToo big
DeepSeek-V4-Flash291B189.0 GBToo big
Hy3299B194.2 GBToo big
DeepSeek-V4-Flash-0731304B197.4 GBToo big
DeepSeek-V4-Flash-0731-ROCmFP3304B197.4 GBToo big
DeepSeek-V4-Flash-Vision-Exp305B198.1 GBToo big
keys-DeepSeekV4Flash-Vision-EXP-ablit305B198.1 GBToo big
GLM-5.3-Flash321B208.4 GBToo big
DeepSeek-V4.1-Flash552B357.9 GBToo big
DeepSeek-V4.1-Flash-1017552B357.9 GBToo big
GLM-5.2753B487.9 GBToo big
How this is estimated: weights = parameters × bits per weight; we add ~15% for KV cache at moderate context and runtime overhead, plus 0.8 GB engine floor. Long contexts, larger batch sizes and vision towers raise the real number — treat a "Tight" as "possible with reduced context". Measured throughput lives on the benchmark board.

Related: benchmark board · quantization guide · the VRAM cliff · VRAM matrix

↑