Tesla V100 16GB Price & Used Prices: Which Local LLMs Can It Run?
Price alert
We'll email you when the eBay (US) price reaches your target.
NVIDIA's Tesla V100 16GB (2017) is a 900 GB/s data center card, still about 83 tokens/s on 12B at 4-bit. New CUDA and vLLM have dropped it.
eBay
Tesla V100 16GB 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 Tesla V100 16GB specifications
- Memory16 GB · HBM2
- Bandwidth900 GB/s
- FP16 / BF16112 TFLOPS
- INT856.1 TOPS
- Bus width
- 4,096-bit
- Architecture
- Volta (GV100)
- Cores
- 5,120
- Tensor cores
- 640
- Ecosystem
- CUDA
- Interface
- PCIe 3.0 x16
- NVLink
- Not supported
- Power
- 250 W
- Power connector
- 8-pin CPU (EPS-12V)
- Slots
- 2
- Length
- 267 mm
- Type
- Data center
- Launch date
- Jun 21, 2017
Best Tesla V100 16GB deals on eBay
Hand-picked deals from trusted eBay sellers. 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.
UsedNVIDIA Tesla V100 16GB PCIe GPU Accelerator - 900-2G500-0000-000$299Buy on eBay
UsedOpen Box 16GB nVIDIA Tesla V100 PCI Express 3.0 x16 GPU 699-2G500-0200-310Top Rated$320Buy on eBay
UsedOpen Box 16GB nVIDIA 900-2G500-0000-000 Tesla V100 PCI-E x16 Module GPUTop Rated$327Buy on eBay
UsedNVIDIA Tesla V100 16GB Graphic Card PCIE 3.0$340Buy on eBay
UsedNVIDIA Tesla V100 PCIe 16GB HBM2 Graphics Card GPU 699-2G500-0200-300/310$349Buy on eBay
UsedNvidia Tesla Volta V100 - 900-2G500-0000-000 - 16GB - GPU - Tested & WorkingTop Rated$350Buy on eBay
What LLMs can a single Tesla V100 16GB run?
One V100 16GB is best with up to 12B at 4-bit. 30B-class MoE at 2-bit fits on the card at about 150 tokens/s; 4-bit with system RAM keeps a third.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 142.5 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 132.7 | 256K |
| Gemma 4 12B12B | 4.7 GB | Runs | 117 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 137.7 | 256K |
| Qwen3.6 27B27.8B | 11.8 GB | Runs | 53 | 64K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 62.6 | 93K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 148.4 | 75K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 51.1 | 41K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 158.6 | 181K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 114.2 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 101.5 | 256K |
| Gemma 4 12B12B | 7.1 GB | Runs | 83.2 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Needs RAM offload | 38.7 | 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 | 36.5 | 198K |
| Gemma 4 31B30.7B | 18.3 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Needs RAM offload | 52 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 74.6 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 65.1 | 196K |
| Gemma 4 12B12B | 12.7 GB | Runs | 49.6 | 183K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Needs RAM offload | 26.2 | 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 | 22.8 | 198K |
| Gemma 4 31B30.7B | 32.6 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Needs RAM offload | 34.6 | 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.
What LLMs can multiple Tesla V100 16GB cards run?
Large MoE on the V100 16GB takes eight cards (128GB), running Qwen3.8 Flash Next at 4-bit at about 93 tokens/s. Stay on vLLM 0.19 or earlier.
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 78.9 GB | 8 cards | $2,960 | 128 GB | 109.1 | 2,000 W |
 DeepSeek V4 Flash284B (13B active) | 96.8 GB | 8 cards | $2,960 | 128 GB | 88.9 | 2,000 W |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | 8 cards | $2,960 | 128 GB | 72.3 | 2,000 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | 8 cards | $2,960 | 128 GB | 92.8 | 2,000 W |
At this quantization, even 8 × Tesla V100 16GB cannot hold any model over 50B.
Full calculation rules are on the Methodology page.
Who it suits
Good fit
- Speed on small models
Qwen3.5 9B at 4-bit runs at about 102 tokens/s, and still about 65 at 8-bit
- Running 35B-class MoE models at 2-bit
Qwen3.6 35B A3B at 2-bit runs at about 159 tokens/s with 181K of context
- Building an 8-GPU setup in an SXM2 server
SXM2 modules are linked directly to each other by NVLink
Not a good fit
- Running 27B models at 4-bit
16GB falls well short, so you have to drop to 2-bit or get the 32GB version
- You need vLLM, BF16 or long-term driver updates
Recent vLLM no longer supports this generation, and 580 is the last driver branch that supports it
Before you buy
SXM2 module or PCIe card
The V100 comes in two forms: the PCIe version is a standard add-in card, 250W with one CPU 8-pin power connector; the SXM2 version is a server module, 300W, with NVLink between modules. Many SXM2 modules on the used market are sold with third-party adapter boards. The adapter board handles only power and the slot, so you have to supply your own cooling; some people use an 80mm fan with a 3D-printed shroud. Before buying an adapter setup, ask whether the cooler and power cables are included.
Temperature limits and power capping
The PCIe version has no fan. According to NVIDIA's product brief, the GPU halves its clocks at 87°C and shuts down at 90°C, and memory tops out at 85°C. Without server airflow you must add your own fan; you can also lower the power limit with nvidia-smi -pl, and NVIDIA's example uses 180W.
ECC takes no capacity; no video output
Its HBM2 memory has native ECC, and NVIDIA states that enabling ECC takes away neither capacity nor bandwidth, so all 16GB is usable. The card has no video output, so the system needs integrated graphics or another card to drive the display.
How long software support lasts
CUDA 13 no longer supports this generation, and NVIDIA's 580 is the last driver branch that supports it. llama.cpp and Ollama still work, but vLLM is stuck at version 0.19. It also lacks BF16, so models released in BF16 have to be loaded as FP16.
Compared with similar GPUs
Tesla V100 16GB vs Tesla V100 32GB, Tesla T10, and GeForce RTX 5060 Ti 16GB: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | Tesla V100 16GB | Tesla V100 32GB | Tesla T10 | GeForce RTX 5060 Ti 16GB |
|---|---|---|---|---|
| Memory | 16 GB | 32 GB100% more | 16 GBSame | 16 GBSame |
| Bandwidth | 900 GB/s | 900 GB/sSame | 403 GB/s55% lower | 448 GB/s50% lower |
| BF16 / FP16 compute | 112 TFLOPS | 112 TFLOPSSame | 80 TFLOPS29% lower | 95 TFLOPS15% lower |
| INT8 compute | 56 TOPS | 56 TOPSSame | 160 TOPS185% higher | 190 TOPS238% higher |
| Qwen3.8 27B speed (2-bit) | 62.6 t/s | 62.6 t/sSame | 29.4 t/s53% slower | 32.6 t/s48% slower |
| Largest dense model (4-bit) | 23B | 51B122% more | 23BSame | 23BSame |
| Power | 250 W | 250 WSame | 150 W40% lower | 180 W28% lower |
| Interface | PCIe 3.0 x16 | PCIe 3.0 x16 | PCIe 3.0 x16 | PCIe 5.0 x8 |
| Released | 2017-06 | 2018-03 | 2020-02 | 2025-04 |
| 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.
Tesla V100 16GB rental price per hour
On-demand rental of the Tesla V100 16GB starts at $0.12 per hour (Vast.ai), about $28/month at 8 hours a day. Buying one (eBay median $370) pays off in about 3.6 years, 1.8 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 50 min. ago | $0.12/hr | $14/mo | $28/mo | $85/mo | Rent |
 RunPodRentUpdated 50 min. ago | $0.19/hr | $23/mo | $46/mo | $139/mo | Rent |
 eBayUpdated 6 hr. ago | $370 | Payback ~3.6 yrsPower $5/mo | Payback ~1.8 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