NVIDIA · Ampere · 8 GB VRAM tier
Best AI Models for NVIDIA GeForce RTX 3070 (8 GB)
Runs 7–9B models well; 14B doesn't fit in 8 GB.
| # | 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 | Gemma 2 9B ITGoogle Q4_K_M · 7.3 GB · 47.7 t/s | 47.7 tok/s | Q4_K_M · 7.3 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 | 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 | Gemma 3 4B ITGoogle Q4_K_M · 5.0 GB · 103 t/s | 103 tok/s | Q4_K_M · 5.0 GB | |
| 3 | Qwen2.5 7B InstructAlibaba Q4_K_M · 5.8 GB · 58.4 t/s | 58.4 tok/s | Q4_K_M · 5.8 GB | |
| 4 | Gemma 2 9B ITGoogle Q4_K_M · 7.3 GB · 47.7 t/s | 47.7 tok/s | overall | Q4_K_M · 7.3 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 | Long context Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 89.3 t/s | 89.3 tok/s | Q4_K_M · 3.4 GB | |
| 2 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 45.8 t/s | 45.8 tok/s | Q4_K_M · 6.6 GB | |
| 3 | Gemma 4 E4B ITGoogle Q4_K_M · 5.6 GB · 54.7 t/s | 54.7 tok/s | Q4_K_M · 5.6 GB | |
| 4 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB · 221 t/s | 221 tok/s | Q4_K_M · 5.5 GB | |
| 5 | MiniCPM5 2BOpenBMB Q4_K_M · 2.5 GB · 153 t/s | 153 tok/s | Q4_K_M · 2.5 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 89.3 t/s | 89.3 tok/s | Q4_K_M · 3.4 GB | |
| 2 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB · 221 t/s | 221 tok/s | Q4_K_M · 5.5 GB | |
| 3 | Gemma 4 E4B ITGoogle Q4_K_M · 5.6 GB · 54.7 t/s | 54.7 tok/s | Q4_K_M · 5.6 GB | |
| 4 | Q4_K_M · 6.5 GB · 46.9 t/s | 46.9 tok/s | Q4_K_M · 6.5 GB | |
| 5 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 45.8 t/s | 45.8 tok/s | Q4_K_M · 6.6 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 4B ITGoogle Q4_K_M · 5.0 GB · 103 t/s | 103 tok/s | Q4_K_M · 5.0 GB | |
| 4 | Q4_K_M · 6.5 GB · 46.9 t/s | 46.9 tok/s | Q4_K_M · 6.5 GB | |
| 5 | LFM3 Omni 2BLiquid AI Q4_K_M · 2.4 GB · 171 t/s | 171 tok/s | Q4_K_M · 2.4 GB |
The RTX 3070 is one of the fastest 8 GB cards: it runs 7–9B models well. It is power-hungry for its speed, and at its $499 MSRP it gave among the best speed per dollar of the cards compared. Ampere generation, released 2020.
RTX 3070 strengths and weaknesses
Strengths
- Runs 7–9B models: Qwen3.5 9B at 45.8 tok/s in 6.6 GB
Weaknesses
- 14B models don't fit: Qwen3 14B needs 11.9 GB at Q4 vs 8 GB here
- Little room for context: Qwen3.5 9B uses 6.6 of 8 GB, 1.4 GB left
- No FP8 or NVFP4: Ampere predates both; use GGUF, AWQ or GPTQ builds
Speed per dollar and per watt
On Qwen3.5 9B (9B params) at Q4_K_M. New prices are compared with new, used with used. The last two columns are the speed you get per $100 of price and per 100 W of board power; higher is better. The RTX 3070 has no current price, so it is shown at its MSRP.
| Card | New price | tok/s | tok/s per $100 | tok/s per 100 W |
|---|---|---|---|---|
RTX 50608 GBBest new value | about $460 | 45.8 | 10.0 | 31.6 |
RTX 30708 GB | about $499 | 45.8 | 9.2 | 20.8 |
RTX 507012 GB | about $840 | 68.7 | 8.2 | 27.5 |
RTX 40608 GB | about $520 | 27.8 | 5.3 | 24.2 |
| Card | Used price | tok/s | tok/s per $100 | tok/s per 100 W |
|---|---|---|---|---|
RTX 50608 GBBest used value | about $388 | 45.8 | 11.8 | 31.6 |
RTX 3060 12GB12 GB | about $329 | 36.8 | 11.2 | 21.6 |
RTX 507012 GB | about $736 | 68.7 | 9.3 | 27.5 |
RTX 407012 GB | about $575 | 51.5 | 9.0 | 25.8 |
RTX 40608 GB | about $335 | 27.8 | 8.3 | 24.2 |
RTX 4060 Ti 8GB8 GB | about $362 | 29.4 | 8.1 | 18.4 |
RTX 4070 SUPER12 GB | about $675 | 51.5 | 7.6 | 23.4 |
RTX 4070 Ti12 GB | about $722 | 51.5 | 7.1 | 18.1 |
Cost per token: the RTX 3070 against 8–12 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, MSRP
- $0.69
- Power
- $0.20
Renting costs as much as buying after ≈5,900 hours (≈4.0 years at 4 h a day).
Each dot is a card; hover or tap it for its numbers, click to open it. Cards without a price for the chosen basis are left out. How we calculate cost per token
Cards people weigh against the RTX 3070, each with the one reason it is worth a look. Speeds on Qwen3.5 9B.
Output speed depends on memory bandwidth (448 GB/s). Grey rows need more than 8 GB.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB45.8tok/s
- 14BQwen3 14B · not at 4-bitneeds 11.9 GB
- 27–31BQwen3.8 27B · not at 4-bitneeds 17.9 GB
- 35B MoEQwen3.6 35B A3B · not at 4-bitneeds 22.0 GB
- 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 822 models on the RTX 3070
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 |
|---|---|---|---|---|---|---|
LocoTrainer 4BChatCodeFunctions | Q4_K_M | 4.2 GB | 100.4 t/s | 262K | S91, Runs great | |
| BF16 | 1.6 GB | 493.6 t/s | — | S100, Runs great | ||
MobileLLM R1 360MChatReasoning | Q4_K_M | 1.3 GB | 1213.3 t/s | — | S100, Runs great | |
| Q4_K_M | 0.6 GB | 582.4 t/s | 8K | S100, Runs great | ||
Supra Mini V5 8MChat | Q4_K_M | 0.3 GB | 910.0 t/s | 1K | S100, Runs great | |
Qwen3.5 4B Safety ThinkingChatReasoning | Q4_K_M | 3.2 GB | 97.4 t/s | 262K | S91, Runs great | |
GPT X2 125MChat | BF16 | 0.6 GB | 455.0 t/s | 1K | S100, Runs great | |
Tiny Qwen3MoeForCausalLMChatFunctions | Q4_K_M | 0.6 GB | 970.7 t/s | 41K | S100, Runs great | |
| BF16 | 1.8 GB | 163.6 t/s | 2K | S97, Runs great | ||
Quark 135MChat | Q4_K_M | 0.4 GB | 677.2 t/s | 2K | S100, Runs great | |
| BF16 | 1.6 GB | 493.6 t/s | — | S100, Runs great | ||
AI21 Jamba Reasoning 3BChatReasoningFunctions | Q4_K_M | 2.3 GB | 129.4 t/s | 262K | S95, Runs great | |
| Q4_K_M | 5.3 GB | 57.9 t/s | 131K | A83, Runs well | ||
Qwen3.5 4B Claude 4.6 Opus Reasoning DistilledChatReasoning | Q4_K_M | 3.4 GB | 89.3 t/s | 262K | S89, Runs great | |
Nafie 473MChat | BF16 | 1.0 GB | 280.0 t/s | 1K | S100, Runs great |
Specs
- Architecture
- Ampere
- VRAM
- 8 GB GDDR6
- Memory bandwidth
- 448.0 GB/s
- FP16 (dense)
- 40.60 TFLOPS
- CUDA cores
- 5,888
- Tensor cores
- 184
- Compute capability
- 8.6
- TDP
- 220 W
- Released
- 2020-10-29
- MSRP
- $499
RTX 3070 FAQ
Is the RTX 3070 still worth it in 2026?
At its MSRP of $499 you get 9.2 tok/s per $100 on Qwen3.5 9B, which ranks 2nd of the 4 cards compared here; it is also the 2nd fastest out of 12 cards with 8 GB. It runs 7–9B models well; 14B doesn't fit in 8 GB.
What is the largest model the RTX 3070 can run?
Qwen3.5 9B (9.7B) is the largest at 4-bit or better: at Q4_K_M it uses 6.6 GB of 8 GB, and runs at about 45.8 tok/s.
Can the RTX 3070 run 14B models?
No. Qwen3 14B needs 11.9 GB at Q4, more than the 8 GB on this card.
How many tokens per second does the RTX 3070 give?
About 45.8 tok/s on Qwen3.5 9B at Q4_K_M, with the first token after 4.3 s on a 4K-token prompt.
What power supply does the RTX 3070 need?
Plan a 450 W PSU (2× the 220 W TDP).
Is renting cheaper than buying an RTX 3070?
Rent from $0.08/hr. Renting costs as much as buying after ≈5,900 hours (≈4.0 years at 4 h a day).