How much VRAM do I need for a 7B–8B model?

About 6.7 GB at Q4_K_M, 11.0 GB at Q8_0 and 19.8 GB at FP16, at a ~4k context. That is for Qwen3-8B (8.2B parameters): weights plus about 15% for the KV cache and 1 GB for the runtime.

How much VRAM does a 7B–8B model need for 32k context?

About 10.9 GB at Q4_K_M with an FP16 KV cache, or 8.5 GB with a Q8 cache, based on Qwen3-8B's published config. Models with different attention layouts need more or less.

What is the cheapest GPU for a 7B–8B model?

Of the GPUs we track, the Intel Arc B570 (10 GB, $219 list price) is the cheapest that runs it fully on the GPU at Q4.

Can a Mac run a 7B–8B model?

Yes. The cheapest Mac or unified-memory mini PC we track that runs it at Q4 is the Mac mini M4 16GB (16 GB, about 12.0 GB usable, $789).

Does a MoE model of the same size need less VRAM?

No. A mixture-of-experts model still has to hold all of its weights in memory, so it needs about the same VRAM as a dense model with the same total parameters. It runs faster, because only a few experts are read per token.

How much VRAM do you need for a 7B–8B model?

Short answer: about 6.7 GB at Q4_K_M, the usual quant for local use. Below: every quant, longer contexts, the cheapest hardware that fits, and every 7B–8B open model we track.

Q4_K_M: 6.7 GB 5.0 GB weights
Q8_0: 11.0 GB 8.7 GB weights
FP16: 19.8 GB 16.4 GB weights

Worked out for Qwen3-8B (8.2B parameters, from the Hugging Face API). Formula: weights × 1.15 + 1 GB at a ~4k context (how we calculate).

Cheapest GPU that fits at Q4: Intel Arc B570 (10 GB, $219). Cheapest Mac or mini PC: Mac mini M4 16GB ($789).

Cheapest hardware for each quant

Only devices with a sourced price. GPUs at list price (MSRP); street prices may be higher.

Cheapest GPU and Mac or mini PC for a 7B–8B model at each quant
Setting	Memory needed	Cheapest GPU	Cheapest Mac / mini PC
Q4_K_M, ~4k context	6.7 GB	Intel Arc B570 10 GB usable · $219	Mac mini M4 16GB 12 GB usable · $789
Q8_0, ~4k context	11.0 GB	Radeon RX 7600 XT 16 GB usable · $329	Mac mini M6 24GB 18 GB usable · $1,299
FP16, ~4k context	19.8 GB	Intel Arc Pro B70 32GB 32 GB usable · $949	Mac mini M6 32GB 24 GB usable · $1,799
Q4_K_M, 32k context	10.9 GB	Radeon RX 7600 XT 16 GB usable · $329	Mac mini M6 24GB 18 GB usable · $1,299

Memory by context length

Each extra token of context adds to the KV cache. Numbers for Qwen3-8B, from its published config; a Q8 KV cache halves the cache part.

Memory needed for a 7B–8B model by context length
Context	Q4_K_M, FP16 cache	Q4_K_M, Q8 cache	Q8_0, FP16 cache
4k tokens	6.7 GB	6.7 GB	11.0 GB
8k tokens	7.3 GB	6.7 GB	11.6 GB
16k tokens	8.5 GB	7.3 GB	12.8 GB
32k tokens	10.9 GB	8.5 GB	15.2 GB
64k tokens	15.8 GB	10.9 GB	20.1 GB
128k tokens	25.4 GB	15.8 GB	29.7 GB

Hardware that runs it

Bar: Q4 memory needed vs. usable memory. Speed is estimated decode at Q4 (how we estimate).FitsTight

Tracked hardware that runs a 7B–8B model at Q4
Hardware	Q4 need vs. usable	Q4 / Q8 / FP16	Est. tok/s
GeForce RTX 3050 8GBGPU · $249	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	27–38
Radeon RX 7600GPU · $269	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	35–49
Intel Arc A750GPU · $289	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	62–88
GeForce RTX 4060GPU · $299	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	33–47
GeForce RTX 5060GPU · $299	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	54–77
Radeon RX 9060 XT 8GBGPU · $299	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	39–55
GeForce RTX 5060 Ti 8GBGPU · $379	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	54–77
GeForce RTX 3060 TiGPU · $399	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	54–77
GeForce RTX 2060 SuperGPU · $399	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	54–77
GeForce RTX 4060 Ti 8GBGPU · $399	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	35–49
GeForce RTX 3070GPU · $499	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	54–77
GeForce RTX 2070GPU · $499	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	54–77
GeForce RTX 2070 SuperGPU · $499	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	54–77
GeForce RTX 3070 TiGPU · $599	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	73–104
GeForce RTX 2080GPU · $699	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	54–77
GeForce RTX 2080 SuperGPU · $699	6.7 / 8.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	60–85
Intel Arc B570GPU · $219	6.7 / 10.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	46–65
GeForce RTX 3080 10GBGPU · $699	6.7 / 10.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	92–130
GeForce RTX 2080 TiGPU · $999	6.7 / 11.0 GB	Q4FitsQ8Won’t fitFP16Won’t fit	74–105
GeForce RTX 2060 12GBGPU	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	41–58
Intel Arc B580GPU · $249	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	55–78
GeForce RTX 3060 12GBGPU · $329	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	43–62
Radeon RX 7700 XTGPU · $449	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	52–74
Radeon RX 6700 XTGPU · $479	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	46–66
GeForce RTX 5070GPU · $549	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	81–115
Radeon RX 6750 XTGPU · $549	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	52–74
GeForce RTX 4070GPU · $599	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	61–86
GeForce RTX 4070 SuperGPU · $599	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	61–86
Mac mini M4 16GBunified memory · $789	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	11–19
GeForce RTX 3080 12GBGPU · $799	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	110–156
GeForce RTX 4070 TiGPU · $799	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	61–86
Mac mini M6 16GBunified memory · $899	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	14–25
GeForce RTX 3080 TiGPU · $1,199	6.7 / 12.0 GB	Q4FitsQ8TightFP16Won’t fit	110–156
Radeon RX 7600 XTGPU · $329	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	35–49
Radeon RX 9060 XT 16GBGPU · $349	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	39–55
Intel Arc A770 16GBGPU · $349	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	68–96
GeForce RTX 5060 Ti 16GBGPU · $429	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	54–77
GeForce RTX 4060 Ti 16GBGPU · $499	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	35–49
Radeon RX 7800 XTGPU · $499	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	75–107
Radeon RX 7900 GREGPU · $549	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	70–99
Radeon RX 9070GPU · $549	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	77–110
Radeon RX 6800GPU · $579	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	62–88
Radeon RX 9070 XTGPU · $599	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	77–110
Radeon RX 6800 XTGPU · $649	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	62–88
NVIDIA RTX 2000 Ada 16GBGPU · $649	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	27–38
GeForce RTX 5070 TiGPU · $749	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	108–153
GeForce RTX 4070 Ti SuperGPU · $799	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	81–115
GeForce RTX 4080 SuperGPU · $999	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	89–126
GeForce RTX 5080GPU · $999	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	116–164
Radeon RX 6900 XTGPU · $999	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	62–88
NVIDIA RTX A4000GPU · $1,000	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	54–77
Radeon RX 6950 XTGPU · $1,099	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	70–99
GeForce RTX 4080GPU · $1,199	6.7 / 16.0 GB	Q4FitsQ8FitsFP16Won’t fit	87–123
Mac mini M6 24GBunified memory · $1,299	6.7 / 18.0 GB	Q4FitsQ8FitsFP16Won’t fit	15–27
Radeon RX 7900 XTGPU · $899	6.7 / 20.0 GB	Q4FitsQ8FitsFP16Tight	97–137
NVIDIA RTX 4000 Ada 20GBGPU · $1,250	6.7 / 20.0 GB	Q4FitsQ8FitsFP16Tight	43–62
NVIDIA Tesla P40 24GBGPU	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	42–59
Radeon RX 7900 XTXGPU · $999	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	116–164
GeForce RTX 3090GPU · $1,499	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	113–160
GeForce RTX 4090GPU · $1,599	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	122–173
Mac mini M6 32GBunified memory · $1,799	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	15–27
GeForce RTX 3090 TiGPU · $1,999	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	122–173
NVIDIA RTX A5000GPU · $2,250	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	93–131
MacBook Pro M6 32GBunified memory · $2,399	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	15–27
NVIDIA Titan RTXGPU · $2,499	6.7 / 24.0 GB	Q4FitsQ8FitsFP16Fits	81–115
NVIDIA Tesla V100 32GB (PCIe)GPU	6.7 / 32.0 GB	Q4FitsQ8FitsFP16Fits	109–154
AMD Radeon Instinct MI50 32GBGPU	6.7 / 32.0 GB	Q4FitsQ8FitsFP16Fits	124–175
AMD Radeon Instinct MI60 32GBGPU	6.7 / 32.0 GB	Q4FitsQ8FitsFP16Fits	124–175
Intel Arc Pro B70 32GBGPU · $949	6.7 / 32.0 GB	Q4FitsQ8FitsFP16Fits	73–104
GeForce RTX 5090GPU · $1,999	6.7 / 32.0 GB	Q4FitsQ8FitsFP16Fits	217–307
Mac mini M5 Pro 64GBunified memory · $3,199	6.7 / 48.0 GB	Q4FitsQ8FitsFP16Fits	28–49
NVIDIA RTX A6000GPU · $4,649	6.7 / 48.0 GB	Q4FitsQ8FitsFP16Fits	93–131
NVIDIA DGX Spark 64GBunified memory · $4,999	6.7 / 48.0 GB	Q4FitsQ8FitsFP16Fits	25–44
NVIDIA RTX 6000 Ada 48GBGPU · $6,799	6.7 / 48.0 GB	Q4FitsQ8FitsFP16Fits	116–164
Framework Desktop (Ryzen AI Max+ 395) 128GBunified memory · $3,449	6.7 / 96.0 GB	Q4FitsQ8FitsFP16Fits	23–41
MacBook Pro M4 Max 128GBunified memory · $4,999	6.7 / 96.0 GB	Q4FitsQ8FitsFP16Fits	49–88
MacBook Pro M5 Max 128GBunified memory · $6,699	6.7 / 96.0 GB	Q4FitsQ8FitsFP16Fits	56–99
NVIDIA DGX Spark 128GBunified memory · $6,950	6.7 / 96.0 GB	Q4FitsQ8FitsFP16Fits	25–44
NVIDIA RTX Pro 6000 Blackwell 96GBGPU · $8,565	6.7 / 96.0 GB	Q4FitsQ8FitsFP16Fits	217–307
Mac Studio M5 Ultra 256GBunified memory · $9,499	6.7 / 192.0 GB	Q4FitsQ8FitsFP16Fits	109–193
Mac Studio M5 Ultra 512GBunified memory · $22,299	6.7 / 384.0 GB	Q4FitsQ8FitsFP16Fits	109–193

7B–8B models we track

Memory needed by each 7B–8B model we track
Model	Q4_K_M	Q8_0	FP16
granite-4.0-tiny-preview6.7B params · ibm-granite · MoE	5.7 GB	9.2 GB	16.3 GB
OLMo-7B-0724-hf6.9B params · allenai	5.8 GB	9.4 GB	16.8 GB
OLMoE-1B-7B-0924-Instruct6.9B params · allenai · MoE	5.8 GB	9.5 GB	16.9 GB
OLMoE-1B-7B-09246.9B params · allenai · MoE	5.8 GB	9.5 GB	16.9 GB
OLMoE-1B-7B-0125-Instruct6.9B params · allenai · MoE	5.8 GB	9.5 GB	16.9 GB
OLMoE-1B-7B-01256.9B params · allenai · MoE	5.8 GB	9.5 GB	16.9 GB
granite-4.0-h-tiny6.9B params · ibm-granite · MoE	5.8 GB	9.5 GB	17.0 GB
Mistral-7B-Instruct-v0.37.2B params · mistralai	6.1 GB	9.9 GB	17.7 GB
Olmo-3-7B-Think7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Olmo-3-7B-Instruct7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Olmo-3-1025-7B7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Olmo-3-7B-Instruct-SFT7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Olmo-3-7B-Think-SFT7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Olmo-3-7B-Instruct-DPO7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Olmo-3-7B-RL-Zero-Math7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Olmo-3.1-7B-RL-Zero-Math7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Olmo-3-7B-Think-DPO7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
OLMo-2-1124-7B-Instruct7.3B params · allenai	6.1 GB	9.9 GB	17.8 GB
Zamba2-7B-Instruct7.4B params · Zyphra	6.1 GB	10.0 GB	17.9 GB
LLaDA-MoE-7B-A1B-Base7.4B params · inclusionAI · MoE	6.1 GB	10.0 GB	17.9 GB
Olmo-Hybrid-7B7.4B params · allenai	6.2 GB	10.1 GB	18.1 GB
Falcon3-7B-Instruct7.5B params · tiiuae	6.2 GB	10.1 GB	18.1 GB
starvector-8b-im2svg7.5B params · starvector	6.2 GB	10.2 GB	18.3 GB
Falcon-H1R-7B7.6B params · tiiuae	6.3 GB	10.3 GB	18.4 GB
Falcon-H1-7B-Instruct7.6B params · tiiuae	6.3 GB	10.3 GB	18.4 GB
Qwen2.5-Coder-7B-Instruct7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
DeepSeek-R1-Distill-Qwen-7B7.6B params · deepseek-ai	6.3 GB	10.3 GB	18.5 GB
Qwen2.5-7B-Instruct7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
Qwen2.5-7B7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
Qwen2-7B-Instruct7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
Qwen2.5-Coder-7B7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
Qwen2-7B7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
Qwen2.5-7B-Instruct-1M7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
Qwen2.5-Math-7B7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
Qwen2.5-Math-7B-Instruct7.6B params · Qwen	6.3 GB	10.3 GB	18.5 GB
Marco-o17.6B params · ATH-MaaS	6.3 GB	10.3 GB	18.5 GB
SeaLLMs-v3-7B-Chat7.6B params · SeaLLMs	6.3 GB	10.3 GB	18.5 GB
Arcee-Spark7.6B params · arcee-ai	6.3 GB	10.3 GB	18.5 GB
Phi-mini-MoE-instruct7.6B params · microsoft · MoE	6.3 GB	10.3 GB	18.6 GB
internlm2_5-7b-chat7.7B params · internlm	6.4 GB	10.5 GB	18.8 GB
internlm2_5-7b-chat-1m7.7B params · internlm	6.4 GB	10.5 GB	18.8 GB
Molmo2-O-7B7.8B params · allenai	6.4 GB	10.5 GB	18.8 GB
FastVLM-7B7.8B params · apple	6.4 GB	10.5 GB	18.9 GB
salamandra-7b-instruct7.8B params · BSC-LT	6.4 GB	10.5 GB	18.9 GB
salamandra-7b7.8B params · BSC-LT	6.4 GB	10.5 GB	18.9 GB
salamandra-7b-instruct-26067.8B params · BSC-LT	6.4 GB	10.5 GB	18.9 GB
EXAONE-3.5-7.8B-Instruct7.8B params · LGAI-EXAONE	6.5 GB	10.6 GB	19.0 GB
EXAONE-Deep-7.8B7.8B params · LGAI-EXAONE	6.5 GB	10.6 GB	19.0 GB
MiMo-7B-RL7.8B params · XiaomiMiMo	6.5 GB	10.6 GB	19.0 GB
MiMo-7B-Base7.8B params · XiaomiMiMo	6.5 GB	10.6 GB	19.0 GB
MiMo-7B-RL-05307.8B params · XiaomiMiMo	6.5 GB	10.6 GB	19.0 GB
MiMo-7B-SFT7.8B params · XiaomiMiMo	6.5 GB	10.6 GB	19.0 GB
Ling-3.0-tiny7.9B params · inclusionAI · MoE	6.5 GB	10.6 GB	19.2 GB
Nemotron 3.5 Bolt E4B8.0B params · nvidia	6.6 GB	10.8 GB	19.4 GB
Gemma 4 E4B8.0B params · google	6.6 GB	10.8 GB	19.4 GB
Hy-MT2-7B8.0B params · tencent	6.6 GB	10.8 GB	19.5 GB
Llama-3.1-8B-Instruct8.0B params · meta-llama	6.6 GB	10.8 GB	19.5 GB
Llama-3.1-8B8.0B params · meta-llama	6.6 GB	10.8 GB	19.5 GB
DeepSeek-R1-Distill-Llama-8B8.0B params · deepseek-ai	6.6 GB	10.8 GB	19.5 GB
Meta-Llama-3.1-8B-Instruct8.0B params · NousResearch	6.6 GB	10.8 GB	19.5 GB
L3-8B-Stheno-v3.28.0B params · Sao10K	6.6 GB	10.8 GB	19.5 GB
Llama-3.1-Nemotron-Nano-8B-v18.0B params · nvidia	6.6 GB	10.8 GB	19.5 GB
Meta-Llama-3.1-8B8.0B params · NousResearch	6.6 GB	10.8 GB	19.5 GB
LongWriter-llama3.1-8b8.0B params · zai-org	6.6 GB	10.8 GB	19.5 GB
kanana-1.5-8b-instruct-25058.0B params · kakaocorp	6.6 GB	10.8 GB	19.5 GB
Llama-3.1-Tulu-3-8B8.0B params · allenai	6.6 GB	10.8 GB	19.5 GB
Llama-3.1-Tulu-3-8B-DPO8.0B params · allenai	6.6 GB	10.8 GB	19.5 GB
Llama-3.1-Tulu-3.1-8B8.0B params · allenai	6.6 GB	10.8 GB	19.5 GB
Apertus-8B-Instruct-25098.1B params · swiss-ai	6.6 GB	10.8 GB	19.5 GB
Apertus-8B-25098.1B params · swiss-ai	6.6 GB	10.8 GB	19.5 GB
granite-3.1-8b-instruct8.2B params · ibm-granite	6.7 GB	11.0 GB	19.8 GB
granite-3.2-8b-instruct8.2B params · ibm-granite	6.7 GB	11.0 GB	19.8 GB
granite-3.0-8b-instruct8.2B params · ibm-granite	6.7 GB	11.0 GB	19.8 GB
granite-3.3-8b-instruct8.2B params · ibm-granite	6.7 GB	11.0 GB	19.8 GB
MiniCPM4-8B8.2B params · openbmb	6.7 GB	11.0 GB	19.8 GB
MiniCPM4.1-8B8.2B params · openbmb	6.7 GB	11.0 GB	19.8 GB
AgentCPM-Report8.2B params · openbmb	6.7 GB	11.0 GB	19.8 GB
MiniCPM4-MCP8.2B params · openbmb	6.7 GB	11.0 GB	19.8 GB
Qwen3-8B8.2B params · Qwen	6.7 GB	11.0 GB	19.8 GB
DeepSeek-R1-0528-Qwen3-8B8.2B params · deepseek-ai	6.7 GB	11.0 GB	19.8 GB
Qwen3-8B-Base8.2B params · Qwen	6.7 GB	11.0 GB	19.8 GB
WeDLM-8B-Instruct8.2B params · tencent	6.7 GB	11.0 GB	19.8 GB
PaCoRe-8B8.2B params · stepfun-ai	6.7 GB	11.0 GB	19.8 GB
Seed-Coder-8B-Instruct8.3B params · ByteDance-Seed	6.8 GB	11.1 GB	20.0 GB
Seed-Coder-8B-Reasoning8.3B params · ByteDance-Seed	6.8 GB	11.1 GB	20.0 GB
Stable-DiffCoder-8B-Instruct8.3B params · ByteDance-Seed	6.8 GB	11.1 GB	20.0 GB
Qwen2-VL-7B-Instruct8.3B params · Qwen	6.8 GB	11.1 GB	20.1 GB
UI-TARS-7B-DPO8.3B params · ByteDance-Seed	6.8 GB	11.1 GB	20.1 GB
UI-TARS-7B-SFT8.3B params · ByteDance-Seed	6.8 GB	11.1 GB	20.1 GB
Qwen2.5-VL-7B-Instruct8.3B params · Qwen	6.8 GB	11.1 GB	20.1 GB
UI-TARS-1.5-7B8.3B params · ByteDance-Seed	6.8 GB	11.1 GB	20.1 GB
Fara-7B8.3B params · microsoft	6.8 GB	11.1 GB	20.1 GB
UI-Venus-Ground-7B8.3B params · inclusionAI	6.8 GB	11.1 GB	20.1 GB
MiMo-VL-7B-RL-25088.3B params · XiaomiMiMo	6.8 GB	11.1 GB	20.1 GB
MiMo-VL-7B-RL8.3B params · XiaomiMiMo	6.8 GB	11.1 GB	20.1 GB
MiMo-Embodied-7B8.3B params · XiaomiMiMo	6.8 GB	11.1 GB	20.1 GB
MiMo-VL-7B-SFT8.3B params · XiaomiMiMo	6.8 GB	11.1 GB	20.1 GB
rnj-1-instruct8.3B params · EssentialAI	6.8 GB	11.2 GB	20.1 GB
LFM2-8B-A1B8.3B params · LiquidAI · MoE	6.8 GB	11.2 GB	20.2 GB
Mistral-NeMo-Minitron-8B-Base8.4B params · nvidia	6.9 GB	11.3 GB	20.4 GB
LFM2.5-8B-A1B8.5B params · LiquidAI · MoE	6.9 GB	11.3 GB	20.5 GB
Nemotron-Labs-Diffusion-8B-Base8.5B params · nvidia	6.9 GB	11.4 GB	20.5 GB
Intern-S1-mini8.5B params · internlm	7.0 GB	11.4 GB	20.6 GB
aya-vision-8b8.6B params · CohereLabs	7.0 GB	11.5 GB	20.9 GB
MiniCPM-V-4_58.7B params · openbmb	7.1 GB	11.6 GB	21.0 GB
Llama-3.1-Nemotron-Nano-VL-8B-V18.7B params · nvidia	7.1 GB	11.7 GB	21.1 GB
Qwen3-VL-8B-Instruct8.8B params · Qwen	7.1 GB	11.7 GB	21.2 GB
Qwen3-VL-8B-Thinking8.8B params · Qwen	7.1 GB	11.7 GB	21.2 GB
UI-Venus-1.5-8B8.8B params · inclusionAI	7.1 GB	11.7 GB	21.2 GB
granite-switch-4.1-8b8.8B params · ibm-granite	7.1 GB	11.7 GB	21.2 GB
granite-4.1-8b8.8B params · ibm-granite	7.1 GB	11.7 GB	21.2 GB
granite-4.2-8b8.8B params · ibm-granite	7.1 GB	11.7 GB	21.2 GB
internlm3-8b-instruct8.8B params · internlm	7.1 GB	11.8 GB	21.2 GB
Yi-Coder-9B-Chat8.8B params · 01-ai	7.2 GB	11.8 GB	21.3 GB
Yi-Coder-9B8.8B params · 01-ai	7.2 GB	11.8 GB	21.3 GB
ZAYA1-base8.8B params · Zyphra	7.2 GB	11.8 GB	21.3 GB
NVIDIA-Nemotron-Nano-9B-v28.9B params · nvidia	7.2 GB	11.9 GB	21.4 GB
EuroLLM-9B-Instruct-25129.2B params · utter-project	7.4 GB	12.2 GB	22.1 GB
gemma-2-9b-it9.2B params · google	7.4 GB	12.3 GB	22.3 GB
gemma-2-9b9.2B params · google	7.4 GB	12.3 GB	22.3 GB
Gemma-SEA-LION-v3-9B-IT9.2B params · aisingapore	7.4 GB	12.3 GB	22.3 GB
moondream3-preview9.3B params · moondream	7.5 GB	12.3 GB	22.3 GB
glm-4-9b-chat-hf9.4B params · zai-org	7.6 GB	12.5 GB	22.6 GB
codegeex4-all-9b9.4B params · zai-org	7.6 GB	12.5 GB	22.6 GB
GLM-4-9B-04149.4B params · zai-org	7.6 GB	12.5 GB	22.6 GB
GLM-Z1-9B-04149.4B params · zai-org	7.6 GB	12.5 GB	22.6 GB
NorthCUA-Nano9.4B params · beipo-ai	7.6 GB	12.5 GB	22.6 GB
MiniCPM-SALA9.5B params · openbmb	7.6 GB	12.6 GB	22.8 GB
Ornith-1.5-9B9.7B params · ornith-ai	7.7 GB	12.8 GB	23.2 GB
Qwen3.5-9B9.7B params · Qwen	7.7 GB	12.8 GB	23.2 GB
SenseNova-Loop-9B-Preview9.7B params · sensenova	7.7 GB	12.8 GB	23.2 GB
ZAYA1-VL-8B9.7B params · Zyphra	7.8 GB	12.9 GB	23.4 GB

Have a different amount of memory? Try the VRAM calculator or see best LLMs by VRAM size.

Quick answers

How much VRAM do I need for a 7B–8B model?: About 6.7 GB at Q4_K_M, 11.0 GB at Q8_0 and 19.8 GB at FP16, at a ~4k context. That is for Qwen3-8B (8.2B parameters): weights plus about 15% for the KV cache and 1 GB for the runtime.
How much VRAM does a 7B–8B model need for 32k context?: About 10.9 GB at Q4_K_M with an FP16 KV cache, or 8.5 GB with a Q8 cache, based on Qwen3-8B's published config. Models with different attention layouts need more or less.
What is the cheapest GPU for a 7B–8B model?: Of the GPUs we track, the Intel Arc B570 (10 GB, $219 list price) is the cheapest that runs it fully on the GPU at Q4.
Can a Mac run a 7B–8B model?: Yes. The cheapest Mac or unified-memory mini PC we track that runs it at Q4 is the Mac mini M4 16GB (16 GB, about 12.0 GB usable, $789).
Does a MoE model of the same size need less VRAM?: No. A mixture-of-experts model still has to hold all of its weights in memory, so it needs about the same VRAM as a dense model with the same total parameters. It runs faster, because only a few experts are read per token.

Other sizes: 1B–2B · 2B–3B · 4B–6B · 12B–14B · 16B–21B · 24B–27B · 30B–32B · 40B–50B · 70B · 120B · 200B–300B · 300B–480B · 500B–800B · 1T+ · All sizes · Ollama VRAM by tag · GGUF sizes by quant