NVIDIA · Ampere · 16 GB VRAM tier
Best AI Models for NVIDIA RTX A4000 (16 GB)
Runs 14B models well; 27–31B only just fits.
| # | Model | Speed | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 45.8 t/s | 45.8 tok/s | Q4_K_M · 6.6 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 89.3 t/s | 89.3 tok/s | Q4_K_M · 3.4 GB | |
| 3 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 97.9 t/s | 97.9 tok/s | Q4_K_M · 13.1 GB | |
| 4 | Phi 4Microsoft Q4_K_M · 12.5 GB · 30.6 t/s | 30.6 tok/s | Q4_K_M · 12.5 GB | |
| 5 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 35.4 t/s | 35.4 tok/s | Q4_K_M · 9.0 GB |
| # | Model | Speed | Coding Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | GPT OSS 20BOpenAIHow to run → Q4_K_M · 13.1 GB · 97.9 t/s | 97.9 tok/s | Q4_K_M · 13.1 GB | |
| 2 | Q4_K_M · 7.9 GB · 54.9 t/s | 54.9 tok/s | Q4_K_M · 7.9 GB | |
| 3 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 45.8 t/s | 45.8 tok/s | overall | Q4_K_M · 6.6 GB |
| 4 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 35.4 t/s | 35.4 tok/s | Q4_K_M · 9.0 GB | |
| 5 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB · 221 t/s | 221 tok/s | Q4_K_M · 5.5 GB |
| # | Model | Speed | Reasoning Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 45.8 t/s | 45.8 tok/s | Q4_K_M · 6.6 GB | |
| 2 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 97.9 t/s | 97.9 tok/s | Q4_K_M · 13.1 GB | |
| 3 | Qwen3 8BAlibaba Q4_K_M · 7.6 GB · 52.8 t/s | 52.8 tok/s | Q4_K_M · 7.6 GB | |
| 4 | Q4_K_M · 7.9 GB · 54.9 t/s | 54.9 tok/s | Q4_K_M · 7.9 GB | |
| 5 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 35.4 t/s | 35.4 tok/s | Q4_K_M · 9.0 GB |
| # | Model | Speed | Long context Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 45.8 t/s | 45.8 tok/s | Q4_K_M · 6.6 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 89.3 t/s | 89.3 tok/s | Q4_K_M · 3.4 GB | |
| 3 | Mistral Nemo Instruct 2407Mistral AI Q4_K_M · 11.0 GB · 36.1 t/s | 36.1 tok/s | Q4_K_M · 11.0 GB | |
| 4 | Q4_K_M · 7.9 GB · 54.9 t/s | 54.9 tok/s | Q4_K_M · 7.9 GB | |
| 5 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 35.4 t/s | 35.4 tok/s | Q4_K_M · 9.0 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | GPT OSS 20BOpenAIHow to run → Q4_K_M · 13.1 GB · 97.9 t/s | 97.9 tok/s | Q4_K_M · 13.1 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 89.3 t/s | 89.3 tok/s | Q4_K_M · 3.4 GB | |
| 3 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 45.8 t/s | 45.8 tok/s | Q4_K_M · 6.6 GB | |
| 4 | Gemma 4 12B ITGoogle Q4_K_M · 9.0 GB · 35.4 t/s | 35.4 tok/s | Q4_K_M · 9.0 GB | |
| 5 | Phi 4Microsoft Q4_K_M · 12.5 GB · 30.6 t/s | 30.6 tok/s | Q4_K_M · 12.5 GB |
| # | Model | Speed | Vision Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 45.8 t/s | 45.8 tok/s | Q4_K_M · 6.6 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 89.3 t/s | 89.3 tok/s | Q4_K_M · 3.4 GB | |
| 3 | Gemma 3 12B ITGoogle Q4_K_M · 10.9 GB · 36.2 t/s | 36.2 tok/s | Q4_K_M · 10.9 GB | |
| 4 | Gemma 3 4B ITGoogle Q4_K_M · 5.0 GB · 103 t/s | 103 tok/s | Q4_K_M · 5.0 GB | |
| 5 | Q4_K_M · 6.5 GB · 46.9 t/s | 46.9 tok/s | Q4_K_M · 6.5 GB |
The RTX A4000 runs 14B models well. Ampere generation, released 2021.
RTX A4000 strengths and weaknesses
Strengths
- Runs 35B MoE models: Qwen3.6 35B A3B at 108 tok/s in 15.7 GB
- Reads long prompts fast: A 4K-token prompt on Qwen3.5 9B starts answering in 2.3 s
- Low power draw: 140 W board power; a 300 W PSU is enough
Weaknesses
- 70B models don't fit: Llama 3.3 70B Instruct needs 51.3 GB at Q4 vs 16 GB here
- No FP8 or NVFP4: Ampere predates both; use GGUF, AWQ or GPTQ builds
Output speed depends on memory bandwidth (448 GB/s). Grey rows need more than 16 GB.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB45.8tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB30.7tok/s
- 19.4tok/s
- 108tok/s
- 70BLlama 3.3 70B Instruct · not at 4-bitneeds 51.3 GB
- 120B MoEGPT OSS 120B · not at 4-bitneeds 70.8 GB
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
All 1057 models on the RTX A4000
Featured onlyBest quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works
| Model | Quant | VRAM | Speed | Context | Score | Fit |
|---|---|---|---|---|---|---|
Qwen3.5 4B BaseVision | Q4_K_M | 3.4 GB | 89.3 t/s | 262K | S89, Runs great | |
Qwen3.5 2B BaseVision | Q4_K_M | 1.9 GB | 164.5 t/s | 262K | S97, Runs great | |
| Q4_K_M | 1.7 GB | 334.7 t/s | 33K | S100, Runs great | ||
Agents A1ChatVisionFunctions | Q3_K_S | 15.8 GB | 121.8 t/s | 262K | C44, Usable | |
Tiny Random Phi3Chat | Q4_K_M | 0.3 GB | 970.7 t/s | 4K | S100, Runs great | |
Gemma 2 Mitra EChat | Q4_K_M | 7.7 GB | 45.1 t/s | 8K | A77, Runs well | |
Tiny MixtralChat | Q4_K_M | 0.5 GB | 1206.9 t/s | 131K | S100, Runs great | |
LFM2.5 1.2B InstructChatFunctions | Q4_K_M | 1.2 GB | 272.1 t/s | 128K | S100, Runs great | |
Supra 50M BaseChat | Q4_K_M | 0.3 GB | 808.9 t/s | 1K | S100, Runs great | |
Olmo 3.1 32B InstructChatFunctions | IQ3_S | 15.9 GB | 20.0 t/s | 66K | C30, Usable | |
Distilgpt2Chat | Q4_K_M | 0.1 GB | 4853.3 t/s | 1K | S100, Runs great | |
KernelLLMChat | Q4_K_M | 7.3 GB | 54.0 t/s | 131K | A81, Runs well | |
NuExtract 1.5Chat | Q4_K_M | 9.0 GB | 85.6 t/s | 131K | S88, Runs great | |
| Q4_K_M | 5.7 GB | 59.3 t/s | 8K | A84, Runs well | ||
Chatglm3 6BChat | Q4_K_M | 5.2 GB | 70.7 t/s | 8K | S86, Runs great |
Rent NVIDIA RTX A4000 in the cloud
Hourly rate on demand for one RTX A4000.
Vast.aiMarketplace
from$0.088/hr
Buying one ($1,000) costs as much as ~11,300 hours of rental, about 47 months at 8 hours a day.
Cost per million tokens
Output-token cost on an RTX A4000 rented at $0.088/hr, one request at a time. Batch several users on the same card and the cost per token drops several-fold.
| Model | Quant | Speed | Per 1M output |
|---|---|---|---|
| GPT OSS 20B | 98 tok/s | $0.25$0.0059 input | |
| Gemma 4 26B A4B IT | 95 tok/s | $0.26$0.0062 input | |
| Qwen3.5 9B | 46 tok/s | $0.53$0.014 input | |
| Gemma 4 12B IT | 35 tok/s | $0.69$0.017 input | |
| Qwen3.8 27B | 19 tok/s | $1.26$0.039 input |
Specs
- Architecture
- Ampere
- VRAM
- 16 GB GDDR6
- Memory bandwidth
- 448.0 GB/s
- FP16 (dense)
- 76.70 TFLOPS
- CUDA cores
- 6,144
- Tensor cores
- 192
- Compute capability
- 8.6
- TDP
- 140 W
- Released
- 2021-04-12
- MSRP
- $1,000
RTX A4000 FAQ
What is the largest model the RTX A4000 can run?
Gemma 4 26B A4B IT (25.8B) is the largest at 4-bit or better: at Q4_K_S it uses 15.5 GB of 16 GB, and runs at about 95.1 tok/s. At 3-bit, a bigger model fits too: Gemma 2 27B IT (27.2B) at IQ3_M, 15.0 GB.
Can the RTX A4000 run Qwen3.5 9B?
Yes. Qwen3.5 9B fits at Q4_K_M, using 6.6 GB of 16 GB, and it is the best model that runs well on this card.
Can the RTX A4000 run 70B models?
No. Llama 3.3 70B Instruct needs 51.3 GB at Q4, more than the 16 GB on this card.
How many tokens per second does the RTX A4000 give?
About 45.8 tok/s on Qwen3.5 9B at Q4_K_M, with the first token after 2.3 s on a 4K-token prompt.
What power supply does the RTX A4000 need?
Plan a 300 W PSU (2× the 140 W TDP).
Is 16 GB enough for coding and agents?
Yes for coding: GPT OSS 20B, the best of the 2 coding models that fit, runs at about 98 tok/s. For agents, not yet: no agent-scored model fits in less than 96 GB.