GPU memory size · 24 cards·307 featured models fit
Best LLMs for 16 GB VRAM
Runs 9B and 14B models well; 27–31B and 35B MoE only just fit in 16 GB.
| # | Model | Speed on RTX 5080 | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 98.1 t/s | 98.1 tok/s | Q4_K_M · 6.6 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 191 t/s | 191 tok/s | Q4_K_M · 3.4 GB | |
| 3 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 183 t/s | 183 tok/s | Q4_K_M · 13.1 GB | |
| 4 | Phi 4Microsoft Q4_K_M · 12.5 GB · 65.5 t/s | 65.5 tok/s | Q4_K_M · 12.5 GB | |
| 5 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 75.8 t/s | 75.8 tok/s | Q4_K_M · 9.0 GB |
| # | Model | Speed on RTX 5080 | Coding Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | GPT OSS 20BOpenAIHow to run → Q4_K_M · 13.1 GB · 183 t/s | 183 tok/s | Q4_K_M · 13.1 GB | |
| 2 | Q4_K_M · 7.9 GB · 118 t/s | 118 tok/s | Q4_K_M · 7.9 GB | |
| 3 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 98.1 t/s | 98.1 tok/s | overall | Q4_K_M · 6.6 GB |
| 4 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 75.8 t/s | 75.8 tok/s | Q4_K_M · 9.0 GB | |
| 5 | Q4_K_M · 6.5 GB · 101 t/s | 101 tok/s | Q4_K_M · 6.5 GB |
| # | Model | Speed on RTX 5080 | Reasoning Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 98.1 t/s | 98.1 tok/s | Q4_K_M · 6.6 GB | |
| 2 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 183 t/s | 183 tok/s | Q4_K_M · 13.1 GB | |
| 3 | Qwen3 8BAlibaba Q4_K_M · 7.6 GB · 113 t/s | 113 tok/s | Q4_K_M · 7.6 GB | |
| 4 | Q4_K_M · 7.9 GB · 118 t/s | 118 tok/s | Q4_K_M · 7.9 GB | |
| 5 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 75.8 t/s | 75.8 tok/s | Q4_K_M · 9.0 GB |
| # | Model | Speed on RTX 5080 | Long context Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 98.1 t/s | 98.1 tok/s | Q4_K_M · 6.6 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 191 t/s | 191 tok/s | Q4_K_M · 3.4 GB | |
| 3 | Mistral Nemo Instruct 2407Mistral AI Q4_K_M · 11.0 GB · 77.3 t/s | 77.3 tok/s | Q4_K_M · 11.0 GB | |
| 4 | Q4_K_M · 7.9 GB · 118 t/s | 118 tok/s | Q4_K_M · 7.9 GB | |
| 5 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 75.8 t/s | 75.8 tok/s | Q4_K_M · 9.0 GB |
| # | Model | Speed on RTX 5080 ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 191 t/s | 191 tok/s | Q4_K_M · 3.4 GB | |
| 2 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 183 t/s | 183 tok/s | Q4_K_M · 13.1 GB | |
| 3 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 98.1 t/s | 98.1 tok/s | Q4_K_M · 6.6 GB | |
| 4 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 75.8 t/s | 75.8 tok/s | Q4_K_M · 9.0 GB | |
| 5 | Phi 4Microsoft Q4_K_M · 12.5 GB · 65.5 t/s | 65.5 tok/s | Q4_K_M · 12.5 GB |
| # | Model | Speed on RTX 5080 | Vision Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 98.1 t/s | 98.1 tok/s | Q4_K_M · 6.6 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 191 t/s | 191 tok/s | Q4_K_M · 3.4 GB | |
| 3 | Gemma 3 12B ITGoogle Q4_K_M · 10.9 GB · 77.6 t/s | 77.6 tok/s | Q4_K_M · 10.9 GB | |
| 4 | Gemma 3 4B ITGoogle Q4_K_M · 5.0 GB · 220 t/s | 220 tok/s | Q4_K_M · 5.0 GB | |
| 5 | Q4_K_M · 6.5 GB · 101 t/s | 101 tok/s | Q4_K_M · 6.5 GB |
GPUs with 16 GB VRAM
Compare all GPUs →Popular cards come first. Picks need a widely used card and a reasonable price.
Speed, price and value
Speed on Qwen3.5 9B (9B params) at Q4_K_M. New prices are compared with new, used with used; a card without a price for the view is shown at its MSRP. The last two columns are the speed you get per $100 of price and per 100 W of board power; higher is better.
| Card | GB/s | New price | tok/s | tok/s per $100 | tok/s per 100 W |
|---|---|---|---|---|---|
| 960 | about $1,690 | 98.1 | 5.8 | 27.2 | |
| 896 | about $1,170 | 91.6 | 7.8 | 30.5 | |
| 736 | about $999 | 75.2 | – | 23.5 | |
| 672 | about $799 | 68.7 | – | 24.1 | |
| 717 | about $1,199 | 73.3 | – | 22.9 | |
| 640 | about $815 | 60.4 | 7.4 | 19.9 | |
| 640 | about $800 | 60.4 | 7.5 | 27.5 | |
| 320 | about $590 | 30.2 | 5.1 | 18.9 | |
16 more cards with 16 GB · less common or unpriced cards, not picked | |||||
| Card | GB/s | Used price | tok/s | tok/s per $100 | tok/s per 100 W |
|---|---|---|---|---|---|
| 960 | about $1,640 | 98.1 | 6.0 | 27.2 | |
| 896 | about $1,100 | 91.6 | 8.3 | 30.5 | |
| 736 | about $1,155 | 75.2 | 6.5 | 23.5 | |
| 672 | about $900 | 68.7 | 7.6 | 24.1 | |
| 717 | about $1,100 | 73.3 | 6.7 | 22.9 | |
| 640 | about $790 | 60.4 | 7.6 | 19.9 | |
| 640 | about $650 | 60.4 | 9.3 | 27.5 | |
| 320 | about $480 | 30.2 | 6.3 | 18.9 | |
16 more cards with 16 GB · less common or unpriced cards, not picked | |||||
Cost per token: 16 GB against 12 and 24 GB cards
On Qwen3.5 9B (9B params, Q4_K_M, 4K context). Speed, cost per token and power per token depend on the model.
- Card, new price
- $0.81
- Power
- $0.14
Renting costs as much as buying after ≈6,900 hours (≈4.7 years at 4 h a day).
Each dot is a card; hover or tap it for its numbers, click to open it. 16 GB cards without a price for the view are drawn hollow at their MSRP; other cards without one are left out. How we calculate cost per token
Next sizes down and up, plus why they're worth checking out. Speeds on Qwen3.5 9B.
Models that fit in 16 GB VRAM
Featured only1057 models. Best quant that fits with a 16K context, ordered by llmrun Score (100 = average model). Fit grades how well it runs, S best. How ranking works
| Model | Quant | VRAM | Speed | Context | Score | Fit |
|---|---|---|---|---|---|---|
Qwen3.5 9BVisionFunctions | Q4_K_M | 6.6 GB | 73.3 t/s | 262K | S86, Runs great | |
Qwen3.8 27BVisionFunctions | Q3_K_L | 15.4 GB | 31.1 t/s | 262K | B45, Decent | |
Gemma 4 12B ITChatFunctions | Q4_K_M | 9.0 GB | 56.6 t/s | 262K | A82, Runs well | |
Gemma 4 26B A4B ITVisionFunctions | Q4_K_S | 15.5 GB | 139.0 t/s | 262K | B62, Decent | |
GPT OSS 20BChatFunctions | Q4_K_M | 13.1 GB | 145.4 t/s | 131K | S94, Runs great | |
Muse Glimmer 30BVision | IQ3_M | 15.1 GB | 33.6 t/s | 131K | B49, Decent | |
Ling 3.0 TinyChatFunctions | Q4_K_M | 5.5 GB | 300.6 t/s | 131K | S100, Runs great | |
MiMo V2.6 Distill Qwen 9BVisionCodeFunctions | Q4_K_M | 6.5 GB | 75.0 t/s | 262K | S86, Runs great | |
Qwen3.5 4BVisionFunctions | Q4_K_M | 3.4 GB | 142.9 t/s | 262K | S96, Runs great | |
MiniCPM5 2BChatFunctions | Q4_K_M | 2.5 GB | 245.2 t/s | 131K | S100, Runs great | |
Spark X2.6 4BChatFunctions | Q4_K_M | 4.3 GB | 157.4 t/s | 1049K | S97, Runs great | |
Ornith 1.5 9BChatFunctions | Q4_K_M | 6.6 GB | 73.3 t/s | 262K | S86, Runs great | |
LFM2.5 24B A2BChatFunctions | Q4_K_M | 15.0 GB | 183.0 t/s | 131K | B62, Decent | |
ZAYA1 VL 8BVision | Q4_K_M | 7.5 GB | 315.1 t/s | 33K | S100, Runs great | |
NVIDIA Nemotron 3.5 Bolt E4B BF16ChatFunctions | Q4_K_M | 5.6 GB | 87.6 t/s | 66K | S89, Runs great |
Tokens per second by model size. Grey rows need more than 16 GB.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB98.1tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB65.7tok/s
- 41.6tok/s
- 181tok/s
- 70BLlama 3.3 70B Instruct · not at 4-bitneeds 51.3 GB
- 120B MoEGPT OSS 120B · not at 4-bitneeds 70.8 GB
Speed estimates for the RTX 5080: Q4_K_M, 4K context, single user. How we estimate
Macs keep 3.5 GB for the system, so a Mac needs 24 GB to match a 16 GB card (20.5 GB usable). They run the same models slower because memory bandwidth is lower.
| Device | Usable | Bandwidth | Speed on Qwen3.5 9B |
|---|---|---|---|
Mac Mini M5 Pro (24 GB)M5 Pro · 24 GB memory | 20.5 GB | 307 GB/s | 33.8 tok/s |
Mac Mini M4 Pro (24 GB)M4 Pro · 24 GB memory | 20.5 GB | 273 GB/s | 30.0 tok/s |
MacBook Air 13" M5 (24 GB)M5 · 24 GB memory | 20.5 GB | 153.6 GB/s | 16.9 tok/s |
iMac 24" M4 (24 GB)M4 · 24 GB memory | 20.5 GB | 120 GB/s | 13.2 tok/s |
16 GB VRAM FAQ
What models can I run with 16 GB of VRAM?
Models up to the 35B MoE class. 225 of 307 featured models run great. The top pick is Qwen3.5 9B at Q4_K_M: it needs 6.6 GB and runs at about 98.1 tok/s on the RTX 5080. Gemma 4 12B IT and GPT OSS 20B also run well.
Is 16 GB enough for 70B models?
No. Llama 3.3 70B Instruct needs 51.3 GB at Q4, more than 16 GB. Rent a bigger card by the hour, or look at 48 GB VRAM.
What is the best 16 GB GPU for local LLMs?
It depends on budget. Best value new: the RTX 5070 Ti (91.6 tok/s); Best value on the used market: the RX 9070 (60.4 tok/s); Fastest: the RTX 5080 (98.1 tok/s). Speeds are on Qwen3.5 9B.
How fast is a 16 GB card?
On the RTX 5080, Qwen3.5 9B runs at about 98.1 tok/s and Qwen3.6 35B A3B at about 181 tok/s. Older and cheaper cards are slower: speed follows memory bandwidth, not memory size.
Is renting a 16 GB GPU cheaper than buying one?
An RTX 5070 Ti rents from $0.17/hr. Renting costs as much as buying after ≈6,900 hours (≈4.7 years at 4 h a day).
Is a 16 GB Mac the same as a 16 GB GPU?
No. A Mac keeps about 3.5 GB for the system, so you need a 24 GB Mac to match. It is also slower: the Mac Mini M5 Pro (24 GB) runs Qwen3.5 9B at about 33.8 tok/s.