GeForce RTX 5070 Price & Used Prices: Which Local LLMs Can It Run?
Price alert
We'll email you when the eBay (US) price reaches your target.
The RTX 5070 is NVIDIA's 2025 12GB gaming card, fast on models up to 12B: Qwen3.5 9B runs about 79 tokens/s. The laptop version has only 8GB.
eBay
GeForce RTX 5070 price history
eBay
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
Full GeForce RTX 5070 specifications
- Memory12 GB · GDDR7
- Bandwidth672 GB/s
- FP16 / BF16123.5 TFLOPS
- FP8 / INT8247 / 247 TOPS
- Bus width
- 192-bit
- Architecture
- Blackwell (GB205)
- Cores
- 6,144
- Tensor cores
- 192
- Ecosystem
- CUDA
- Interface
- PCIe 5.0 x16
- NVLink
- Not supported
- Power
- 250 W
- Power connector
- 16-pin 12V-2x6
- Type
- Consumer
- Launch date
- Mar 5, 2025
- MSRP
- $549
Best GeForce RTX 5070 deals on eBay
Deals from trusted eBay sellers, picked by our rules. Shop with confidence.
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
UsedPNY Overclocked Triple Fan GeForce RTX 5070 12GB - Graphics Card - ExcellentTop Rated$770Buy on eBay
UsedNVIDIA GeForce RTX 5070 Windforce OC 12GB GDDR7 Graphics Card GV-N5070WF3OC-12GDTop Rated$770Buy on eBay
UsedMSI GeForce RTX 5070 OC 12GB GDDR7 Graphics Card GB205-300Top Rated$770Buy on eBay
UsedGigabyte GeForce RTX 5070 Windforce OC 12GB GDDR7 Graphics Card N5070WF3OC-12GDTop Rated$770Buy on eBay
UsedGigabyte Windforce OC SFF RTX 5070 12GB Excellent Condition$750+$20 shippingBuy on eBay
UsedGigabyte GeForce RTX 5070 12GB GDDR7 Graphics Card GV-N5070WF30C-12GD$785Buy on eBay
What LLMs can a single GeForce RTX 5070 run?
One RTX 5070 is best with up to 12B at 4-bit, at 60 to 90 tokens/s. Bigger models need 2-bit or system RAM, and context gets tight either way.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 112 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 104 | 229K |
| Gemma 4 12B12B | 4.7 GB | Runs | 91.2 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 123 | 52K |
| Qwen3.6 27B27.8B | 11.8 GB | Too big | — | — |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 47.8 | 29K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Needs RAM offload | 50.2 | 185K |
| Gemma 4 31B30.7B | 11.8 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Needs RAM offload | 75.7 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 88.9 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 78.6 | 182K |
| Gemma 4 12B12B | 7.1 GB | Runs | 64 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Needs RAM offload | 37.4 | 256K |
| Qwen3.6 27B27.8B | 16.8 GB | Too big | — | — |
| Qwen3.8 27B27.8B | 16.5 GB | Too big | — | — |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Needs RAM offload | 35.9 | 180K |
| Gemma 4 31B30.7B | 18.3 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Needs RAM offload | 50.3 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 57.3 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 49.8 | 68K |
| Gemma 4 12B12B | 12.7 GB | Too big | — | — |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Needs RAM offload | 25.3 | 256K |
| Qwen3.6 27B27.8B | 28.6 GB | Too big | — | — |
| Qwen3.8 27B27.8B | 29 GB | Too big | — | — |
| GLM 4.7 Flash30B (3B active) | 31.8 GB | Needs RAM offload | 22.4 | 170K |
| Gemma 4 31B30.7B | 32.6 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Needs RAM offload | 33.4 | 256K |
RunsFits fully in VRAM; speed is estimated from VRAM bandwidth.
Needs RAM offloadMoE only: expert weights sit in system RAM, so speed is estimated from 70 GB/s RAM bandwidth.
Too bigWon’t fit on one card at this quantization, even with RAM offload.
Full calculation rules are on the Methodology page.
Who it suits
Good fit
- Fast speeds on models up to 12B
Qwen3.5 9B at 4-bit runs at about 79 tokens/s, and Gemma 4 12B at about 64
- Running Qwen3.8 Flash Next with Strata
On a 5070 with 64GB of RAM, the author reports about 94 tokens/s with Q2_0 and about 53 with IQ3_S (the author's numbers, not retested by us)
- Small cases
The Founders Edition is dual-slot and 242 mm long, the card draws 250W, and NVIDIA recommends a 650W system power supply
Not a good fit
- VRAM is the ceiling
Gemma 4 12B at 8-bit (12.7 GB) doesn't fit, and 27B-class dense models only run at 2-bit, where Qwen3.8 27B gets just 29K of context
- Using expert offloading for speed
Qwen3.6 35B A3B at 4-bit with its experts in system RAM runs at about 50 tokens/s, slower than Gemma 4 12B running from VRAM
Before you buy
Dual-slot Founders Edition, modest power needs
NVIDIA makes a Founders Edition: dual-slot and 242 mm long. Partner card sizes vary by brand. Power comes through a 12V-2x6 connector, and the box includes an adapter for two 8-pin cables; you can also use a single native PCIe 5.0 cable rated for 300W or more.
Match Strata's RAM setup
For Qwen3.8 Flash Next, system RAM matters more than the GPU. The Strata author's test machine is a Ryzen 5 7600 with 64GB of RAM. According to the author's notes, 32GB of RAM only runs the Coder version, 64GB runs every version, and the NVIDIA driver must be 580 or newer.
The laptop 5070 is a different card
The RTX 5070 Laptop GPU has only 8GB of VRAM and a quarter fewer cores than the desktop card. If you see a “5070 laptop”, don't estimate from the numbers on this page.
Compared with similar GPUs
GeForce RTX 5070 vs GeForce RTX 5060 Ti 16GB, GeForce RTX 3060 12GB, and GeForce RTX 5070 Ti: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | GeForce RTX 5070 | GeForce RTX 5060 Ti 16GB | GeForce RTX 3060 12GB | GeForce RTX 5070 Ti |
|---|---|---|---|---|
| Memory | 12 GB | 16 GB33% more | 12 GBSame | 16 GB33% more |
| Bandwidth | 672 GB/s | 448 GB/s33% lower | 360 GB/s46% lower | 896 GB/s33% higher |
| BF16 / FP16 compute | 124 TFLOPS | 95 TFLOPS23% lower | 51 TFLOPS59% lower | 176 TFLOPS42% higher |
| INT8 compute | 247 TOPS | 190 TOPS23% lower | 102 TOPS59% lower | 352 TOPS42% higher |
| Qwen3.8 27B speed (2-bit) | 47.8 t/s | 32.6 t/s32% slower | 26.4 t/s45% slower | 62.3 t/s30% faster |
| Largest dense model (4-bit) | 16B | 23B44% more | 16BSame | 23B44% more |
| Power | 250 W | 180 W28% lower | 170 W32% lower | 300 W20% higher |
| Interface | PCIe 5.0 x16 | PCIe 5.0 x8 | PCIe 4.0 x16 | PCIe 5.0 x16 |
| Released | 2025-03 | 2025-04 | 2021-02 | 2025-02 |
| eBay median |
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
GeForce RTX 5070 rental price per hour
On Vast.ai, the GeForce RTX 5070 rents for $0.22 per hour on demand, about $54/month at 8 hours a day. Buying one (eBay median $899) pays off in about 3.5 years, 1.7 years or 7 months at 4, 8 or 24 hours a day.
| Option | Price | 4 h/day | 8 h/day | 24 h/day | View |
|---|---|---|---|---|---|
 Vast.aiRentUpdated 1 hr. ago | $0.22/hr | $27/mo | $54/mo | $163/mo | Rent |
 eBayUpdated 3 hr. ago | $899 | Payback ~3.5 yrsPower $5/mo | Payback ~1.7 yrsPower $11/mo | Payback ~7 moPower $33/mo | Buy |
Break-even uses the US residential average ($0.18/kWh) and 250 W rated power, excluding resale value and the rest of the build.
Rental links are referral links: we may receive a commission or credit if you sign up through them. As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
Amazon.jp
Xianyu