NVIDIA GeForce RTX 3070
8 GB decides what fits. 448 GB/s decides how fast it runs once it does.
Open in the calculator →Best models for 8 GB, on every card that size →
Find one: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.
- Largest popular model that fitsOlmo-3-7B-Instruct7.3B params · Q4_K_M · needs 7.96 GB~44 tok/s Faster than you readOpen in the calculator →
- Best fast pickGLM-4.6V-Flash10B params · Q4_K_M · needs 7.45 GB~53 tok/s Faster than you readOpen in the calculator →
- Most downloaded that fitsQwen3.5-9B9.7B params · Q4_K_M · needs 7.05 GB~63 tok/s Faster than you readOpen in the calculator →
At 8,192 tokens of context with the whole model in device memory. Speeds are estimates from memory bandwidth, not benchmarks run on this card.
130 of 320 fit entirely
130 of the 320 models the engine can size fit entirely in device memory at Q4_K_M where it is published, otherwise the nearest published format, and 8,192 tokens. 87 run with some layers on system memory (32 GB assumed), and 103 do not run at all.
Featured models · Q4_K_M at 8,192 tokens, including offload
| Model | Needs | Verdict | Decode | Calculator |
|---|---|---|---|---|
| Qwen3.5-2B2.3B parameters | 2.32 GB | fits | ~247 tok/s | Open → |
| NVIDIA-Nemotron-3-Nano-4B-BF164.0B parameters | 3.46 GB | fits | ~114 tok/s | Open → |
| Qwen3-VL-8B-Instruct8.8B parameters | 7.04 GB | fits | ~52 tok/s | Open → |
| Qwen3.5-9B9.7B parameters | 7.05 GB | fits | ~63 tok/s | Open → |
| gemma-4-12B-it12B parameters | 9.54 GB | 10 of 48 layers on system RAM (1.64 GB) | ~21 tok/s with offload | Open → |
| gpt-oss-20b21B parameters | 13.8 GB | 11 of 24 layers on system RAM (5.84 GB) | ~45 tok/s with offload | Open → |
| Qwen3.8-27B28B parameters | 18.6 GB | 40 of 64 layers on system RAM (10.7 GB) | ~6.0 tok/s with offload | Open → |
| Qwen3.6-27B28B parameters | 18.6 GB | 40 of 64 layers on system RAM (10.7 GB) | ~6.0 tok/s with offload | Open → |
| Kimi-Linear-48B-A3B-Instruct49B parameters | 30.4 GB | 21 of 27 layers on system RAM (22.9 GB) | ~41 tok/s with offload | Open → |
| Qwen3.8-Flash-Next180B parameters | 111.6 GB | needs 105.9 GB of system RAM; 32 GB assumed | does not run | Open → |
| DeepSeek-V4-Flash-Vision-Exp305B parameters | 188.1 GB | needs 182.9 GB of system RAM; 32 GB assumed | does not run | Open → |
| GLM-5.3-Flash321B parameters | 198.3 GB | needs 192.9 GB of system RAM; 32 GB assumed | does not run | Open → |
These examples are selected from prominent labs using the catalogue’s latest Hugging Face 30-day downloads and repository-creation freshness signal, with newer releases guaranteed a place. Offloaded rows assume 32 GB of system RAM, and a speed is only shown for a row that runs. Fitting in memory is not the same as loading: whether the runtime and version you have supports each architecture and format on this machine has not been tested here. Each “Open” link carries the same model, format, context and RAM into the calculator.
Fits entirely in NVIDIA GeForce RTX 3070 memory — 130 of 320 sized models
The most demanding model that fits is Olmo-3-7B-Instruct at Q4_K_M: 7.96 GB of the 8 GB, leaving 0.04 GB spare.
Fully resident at 8,192 tokens (or the model’s own maximum, where that is shorter), offload off, at 448 GB/s. Each row is a run of the engine for this configuration; the rows start with current, prominent releases and “fits” is memory, not a tested runtime. Older or less prominent models remain available through this search and “Show all”.
Showing 40 of 130 models that fit.
| Model | Format | Needs | Spare | Decode | Downloads / 30d | Calculator |
|---|---|---|---|---|---|---|
| Qwen3.5-9BQwen · 9.7B params | Q4_K_M | 7.05 GB | 0.95 GB | ~63 tok/s | 9M | Open → |
| Qwen3.5-4BQwen · 4.7B params | Q4_K_M | 3.98 GB | 4.02 GB | ~113 tok/s | 7.8M | Open → |
| Qwen3.5-2BQwen · 2.3B params | Q4_K_M | 2.32 GB | 5.68 GB | ~247 tok/s | 4.9M | Open → |
| gemma-4-E4B-itGoogle · 8.0B params | Q4_K_M | 5.86 GB | 2.14 GB | ~101 tok/s | 4.4M | Open → |
| NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA · 4.0B params | Q4_K_M | 3.46 GB | 4.54 GB | ~114 tok/s | 3.4M | Open → |
| gemma-4-E2B-itGoogle · 5.1B params | Q4_K_M | 4.02 GB | 3.98 GB | ~266 tok/s | 3M | Open → |
| Qwen3.5-0.8BQwen · 873M params | Q4_K_M | 1.46 GB | 6.54 GB | ~541 tok/s | 2.6M | Open → |
| Qwen3-VL-8B-InstructQwen · 8.8B params | Q4_K_M | 7.04 GB | 0.96 GB | ~52 tok/s | 14.6M | Open → |
| North-Micro-Vision-InstructCohereLabs · 2.5B params | Q4_K_M | 2.99 GB | 5.01 GB | ~162 tok/s | 180.7K | Open → |
| granite-4.2-8bIBM · 8.8B params | Q4_K_M | 7.49 GB | 0.51 GB | ~47 tok/s | 128.4K | Open → |
| granite-4.1-3bIBM · 3.4B params | Q4_K_M | 3.72 GB | 4.28 GB | ~110 tok/s | 521.8K | Open → |
| LFM2.5-2.6BLiquidAI · 2.7B params | Q4_K_M | 2.61 GB | 5.39 GB | ~176 tok/s | 108.1K | Open → |
| Qwen3-VL-4B-InstructQwen · 4.4B params | Q4_K_M | 4.72 GB | 3.28 GB | ~82 tok/s | 3.5M | Open → |
| Qwen3-VL-2B-InstructQwen · 2.1B params | Q4_K_M | 3.05 GB | 4.95 GB | ~149 tok/s | 2.8M | Open → |
| granite-4.2-3bIBM · 3.7B params | Q4_K_M | 3.72 GB | 4.28 GB | ~110 tok/s | 50.6K | Open → |
| LFM2.5-230MLiquidAI · 230M params | Q4_K_M | 1.05 GB | 6.95 GB | ~1230 tok/s | 88.6K | Open → |
| Qwen3-0.6BQwen · 752M params | Q4_K_M | 2.22 GB | 5.78 GB | ~229 tok/s | 29.7M | Open → |
| granite-4.1-8bIBM · 8.8B params | Q4_K_M | 7.49 GB | 0.51 GB | ~47 tok/s | 179.3K | Open → |
| LFM2.5-VL-3BLiquidAI · 3.1B params | Q4_K_M | 2.85 GB | 5.15 GB | ~176 tok/s | 26.7K | Open → |
| MiMo-V2.6-Distill-Qwen-9BNEWXiaomiMiMo · 9.4B params | Q4_K_M | 6.90 GB | 1.10 GB | ~63 tok/s | 14.6K | Open → |
| Qwen3-4B-Instruct-2507Qwen · 4.0B params | Q4_K_M | 4.72 GB | 3.28 GB | ~82 tok/s | 3.7M | Open → |
| Ling-3.0-tinyinclusionAI · 7.9B params | Q4_K_M | 5.85 GB | 2.15 GB | ~355 tok/s | 17.7K | Open → |
| Qwen3-8BQwen · 8.2B params | Q4_K_M | 7.04 GB | 0.96 GB | ~52 tok/s | 10.7M | Open → |
| Olmo-3-7B-Instructallenai · 7.3B params | Q4_K_M | 7.96 GB | 0.04 GB | ~44 tok/s | 481.7K | Open → |
| Qwen3-4BQwen · 4.0B params | Q4_K_M | 4.51 GB | 3.49 GB | ~82 tok/s | 7.8M | Open → |
| LFM2.5-8B-A1BLiquidAI · 8.5B params | Q4_K_M | 6.06 GB | 1.94 GB | ~275 tok/s | 32.4K | Open → |
| LFM2.5-350MLiquidAI · 354M params | Q4_K_M | 1.13 GB | 6.87 GB | ~794 tok/s | 69.9K | Open → |
| Hy-MT2-1.8BTencent · 2.0B params | Q4_K_M | 2.47 GB | 5.53 GB | ~183 tok/s | 28.9K | Open → |
| LFM2.5-1.2B-InstructLiquidAI · 1.2B params | Q4_K_M | 1.63 GB | 6.37 GB | ~293 tok/s | 119.2K | Open → |
| Qwen3-1.7BQwen · 2.0B params | Q4_K_M | 3.02 GB | 4.98 GB | ~149 tok/s | 3.1M | Open → |
| Nemotron-3.5-Content-SafetyNVIDIA · 4.3B params | Q4_K_M | 3.73 GB | 4.27 GB | ~110 tok/s | 14.1K | Open → |
| Hy-MT2-7BTencent · 8.0B params | Q4_K_M | 6.50 GB | 1.50 GB | ~53 tok/s | 15.4K | Open → |
| GLM-4.6V-Flashzai-org · 10B params | Q4_K_M | 7.45 GB | 0.55 GB | ~53 tok/s | 103.7K | Open → |
| Qwen2.5-VL-7B-InstructQwen · 8.3B params | Q4_K_M | 6.36 GB | 1.64 GB | ~63 tok/s | 5.8M | Open → |
| SmolLM3-3BHuggingFaceTB · 3.1B params | Q4_K_M | 3.46 GB | 4.54 GB | ~121 tok/s | 617K | Open → |
| Ministral-3-8B-Instruct-2512Mistral AI · 8.9B params | Q4_K_M | 7.14 GB | 0.86 GB | ~50 tok/s | 121.2K | Open → |
| Ministral-3-3B-Instruct-2512Mistral AI · 3.8B params | Q4_K_M | 3.82 GB | 4.18 GB | ~101 tok/s | 114.1K | Open → |
| NVIDIA-Nemotron-Nano-9B-v2NVIDIA · 8.9B params | Q4_K_M | 6.54 GB | 1.46 GB | ~53 tok/s | 333.6K | Open → |
| Qwen3-VL-8B-ThinkingQwen · 8.8B params | Q4_K_M | 7.04 GB | 0.96 GB | ~52 tok/s | 144.2K | Open → |
| DeepSeek-R1-0528-Qwen3-8BDeepSeek · 8.2B params | Q4_K_M | 7.04 GB | 0.96 GB | ~52 tok/s | 694.9K | Open → |
Sized at the format most people actually download, not at FP16. The catalogue ordered by downloads →
Rent an RTX 3070 by the hour
What the NVIDIA GeForce RTX 3070 costs on Vast.ai and RunPod right now, per machine, from one card to eight. Every price is read from the provider's own API.
$0.082per hour
Vast.ai · Marketplace · 99.1% reliableRunPod $0.13/h Community CloudRent on Vast.ai$0.22per hour
Vast.ai · Marketplace · 99.7% reliableRunPod $0.26/h Community CloudRent on Vast.ai$0.43per hour
Vast.ai · Marketplace · 99.7% reliableRunPod $0.52/h Community CloudRent on Vast.aiBuy one or rent one?
Your price, your hours, your electricity. Everything else is arithmetic.
Watts start at the NVIDIA GeForce RTX 3070’s published board power, an upper bound: decoding rarely holds a card at its limit. Rent starts at the cheapest live price (Vast.ai). The electricity price is a placeholder — put yours in.
At 4 h a day, renting costs $116.80 a year.
- That is $9.73 a month, and nothing when the machine is stopped.
- Owning would cost $127.75 a year in electricity, on top of the price.
- At this usage the electricity alone costs more than renting.
Enter the price you would pay to see the exact break-even.
On-demand prices for the whole machine. Vast hosts below 98% measured reliability are left out; RunPod’s Community Cloud is vetted third-party hosts and its Secure Cloud is data-centre capacity. Every rentable card compared.
Referral links Vast.ai, RunPod and Novita pay us a share of what you spend if you sign up through these buttons. It costs you nothing, and it never decides an order or a recommendation: both are computed from the live price and the speed, and options that pay us nothing are listed and recommended on the same terms. How we rank
Find the NVIDIA GeForce RTX 3070: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.
The specification behind every figure
What the manufacturer publishes for this device, and the pages it was read from.
Manufacturer specification
| Memory scope | dedicated |
|---|---|
| Capacity | 8 GB |
| Published options | 8 GB |
| Bandwidth | 448 GB/s |
| Memory type | GDDR6 |
| Bus width | 256-bit |
| FP32 peak | Not published |
| Dense matrix peak | Not published without sparsity |
| Power | Not published |
Source ledger
Caveats
- Memory bandwidth is quoted from NVIDIA's RTX Blackwell architecture whitepaper, whose appendix compares each 50-series card against its predecessors. NVIDIA's web specification pages no longer publish it for this generation.
- No peak throughput figure is entered for this card, so no compute roof is priced for it and time to first token is withheld rather than estimated.
Published capacity is a hardware ceiling, not guaranteed free runtime memory. The calculator shows the runtime reserve separately rather than folding it into a single number.
Run the diagnostic on the NVIDIA GeForce RTX 3070 →