Key takeaways
- 5 listings match this collection right now.
- The T4 connects over PCIe 3.0 x16 and needs above 4G decoding and large BAR support enabled in host firmware, a setting GPU-ready 1U and 2U servers expose in BIOS.
- Its Turing-generation NVENC engine adds hardware HEVC encoding alongside H.264, the video engine behind its transcoding role rather than raw compute throughput.
- Turing's compute capability keeps the T4 on current CUDA toolkits and driver branches well past the Pascal and Maxwell generations it replaced for inference work.
The Tesla T4 is a single-slot, 70 W Turing accelerator with 16GB of GDDR6 and no external power connector, which is why it fits almost any 1U or 2U server with a free PCIe slot. It is the standard card for inference, video transcoding and VDI where an 80GB card is not needed.
5 listings in this collection, filtered from Graphics Cards. The part number and manufacturer are stated on every listing, and the condition wherever the catalogue records one. For quantity, request a volume quote.
How to choose gpus and accelerators
Datacenter cards (Tesla, A-series) are passively cooled and expect server airflow; workstation cards (Quadro, RTX A) carry their own fans and display outputs. Match the slot (PCIe generation and width), the auxiliary power the card needs, and the memory the workload needs: inference and video work fits in 16 GB to 24 GB, training and large models want 32 GB to 80 GB.
Other gpus and accelerators collections
Browse every collection by specification
Account
