Key Specifications
We couldn't find any available K620 GPUs. Search for alternative GPUs.
Frequently Asked Questions
Why choose the K620?
2GB DDR3. Legacy Kepler professional GPU for basic display and old CAD workloads.
When is the K620 not a good fit?
Kepler architecture with no modern compute capability. Only viable as a basic display adapter for legacy systems.
Are K620 prices going up or down?
The median on-demand price across providers has fallen about 19% since May 2025, from $0.08 to $0.07/hr per GPU. This reflects the market-wide median, which moves both when providers change prices and when lower or higher priced offerings enter the market.
What size AI models can the K620 run?
With 2GB of VRAM, the K620 is mainly limited to small quantized models and lightweight GPU workloads.
How much VRAM does the K620 have?
The K620 has 2GB of VRAM. Multi-GPU setups increase total memory, but that memory is not automatically pooled across GPUs.
What is the K620's memory bandwidth?
The K620 has 29 GB/s of memory bandwidth. Higher bandwidth helps with faster data transfer between GPU memory and compute cores.
What data types does the K620 support?
The K620 supports 1 precision formats. Training: FP32.
Does the K620 support NVLink?
No. The K620 is a PCIe-only GPU with no NVLink, so it is better suited to single-GPU inference and smaller-scale workloads than large distributed training jobs.
Technical Specifications
| Feature | Specification |
|---|---|
| GPU Microarchitecture | Kepler |
| CUDA Cores | 384 |
| TDP | 45W |
| Memory Bus Width | 64 bit |
| Memory Clock Speed | 1600 MHz |
| Memory Bandwidth | 29 GB/s |
| Memory | 2GB GDDR3 |
| System Interface | PCIe 2.0 x16 |
| GPU Clock speed | 900 MHz, 1800 Mbps effective |
| CUDA Compute Capability | 5.0 |
| FP32 (float) | 863.2 GFLOPS |
| FP64 (double) | 26.98 GFLOPS (1:32) |
Source: official Nvidia K620 datasheet.
Last updated