NVIDIA Tesla T4, PCIe®, Datacenter GPU / Graphics Card

16GB GDDR6, ECC, Turing 2560 Cores, 320 Tensor, PCIe® 3.0 (x16)

Byte code:GPU-0061-0001

Manufacturer code:T4 / HCKJM

NVIDIA Turing architecture provides dedicated processing for supported GPU applications.

16GB of GDDR6 holds the accelerator's active application data and working set.

PCIe 3.0 provides the host connection in a deployment matched to the exact card.

Dell HCKJM identifies this T4 assembly for host, bracket and software qualification.

A one-year warranty supports this Dell Tesla T4 GPU accelerator purchase.

Configuration

DELL

HPE

£595.00

£714.00 inc VAT

REQUEST A QUOTE

Tell us what you need. We aim to send an initial reply within one working day. Our team will confirm pricing and availability before providing your quote.

Any price shown for an unavailable item is a guide. We’ll confirm its price and availability in your quote.

We’re here to help

Speak to our team about compatibility, configuration, volume pricing or sourcing the quantities you need. We can also help you check lead times and suggest alternatives based on your business needs.

Offline
Overview

IDENTITY
The Dell NVIDIA Tesla T4 HCKJM is a Turing-based GPU accelerator with 16GB of GDDR6 memory. It supplies a dedicated processing resource for compatible inference and other supported GPU applications, with the exact Dell assembly matched to the host and its software environment.

OVERVIEW
A GPU-enabled service divides work between the server and an accelerator according to the application's processing path. The T4 gives supported operations their own compute and memory resource while the CPU handles surrounding tasks. Its 16GB local allocation provides a concrete working-set size for planning. The card is relevant when the application supports the Turing platform and the deployment needs an accelerator that fits this model's capabilities and host requirements.

PERFORMANCE
Turing architecture forms the processing foundation, with GDDR6 memory holding active application data and PCIe 3.0 linking the card to the host. Software controls which operations use the GPU and how data is transferred. Model structure, batch size, numerical requirements and driver support influence the results. The 16GB allocation addresses local data capacity, while the complete server and application determine practical execution speed and the service that can be delivered.

Use Cases
Compatible inference applications can use the card to process new input data with trained models. GPU-enabled software can also place supported compute stages on the accelerator while the host manages preparation and output. Development environments can use an appropriately configured T4 system to evaluate application deployment and resource needs. These activities give the card a clear purpose when the software and its working data fit a 16GB Turing accelerator.

CONFIGURATION
This offer identifies Dell HCKJM with 16GB of GDDR6 memory. Match the exact card assembly and supplied bracket to the host's PCIe provision, mechanical clearance, power arrangement and cooling requirements. Use supported operating-system, driver and application versions, with virtualisation software qualified where applicable. Cables, adapters, software licences and other deployment components follow the agreed order. GPU memory is separate from system RAM. Cooling and bracket details must follow the actual supplied assembly rather than a differently packaged T4 offer.

POSITIONING
The Tesla T4 is a Turing accelerator, distinct from the Pascal-based P40 and Ampere-based A2. Although the A2 also has a 16GB allocation, a shared memory capacity does not make their platforms or host requirements interchangeable. The T4 also serves a different deployment role from a Turing workstation graphics card. Keeping HCKJM and the full T4 identity together helps specify the intended resource within an established server and software configuration.

PURCHASE VALUE
A one-year warranty supports this Dell GPU accelerator. It provides a targeted option for a compatible host running applications suited to the T4 platform. Selecting the card around its actual processing role and memory requirement gives the purchase a practical purpose in deploying supported inference and compute work, while the existing server continues to supply the processor, storage, networking and system-memory resources around that workload.

Techspecs

Identity

Form / application
Datacenter, PCIe® Slot
Manufacturer
NVIDIA
OEM
DELL
Configuration Difference
DELL, 16GB, HCKJM
Family / Brand
Tesla
Model
T4
Model Alt A
HCKJM

Condition & Compliance

Grade
A
Condition
Recertifiable
Refurbisher
Manufacturer
Materials
Silicon, Copper, Epoxy Resin, Fibreglass, Aluminium
Colour
Black, Green, Nickel, Silver
Warranty (yrs)
1
Life Status
Discontinued
Launch Date
2018-09-12
Launch Price ($)
10631.0
ECCN
4A994.L
Certification
RoHS, RoHS 2, REACH, WEEE, CE, UKCA, FCC Class A, ICES-003, BSMI
Region
Global

Platform & Architecture

Architecture
Turing
Die
Compute
Interconnect
PCIe® Gen3

Performance

TDP (W)
70.0

Graphics & Processing

AI Engine
NVIDIA Turing Tensor Cores
GPU Type
NVIDIA Tesla
GPU cores
2560
Cores (rt)
40
Cores (tensor)
320
FP32 (TFLOPS)
8.1
FP64 (TFLOPS)
0.2544
Tensor (TFLOPS)
65.0
VRAM (GB)
16.0
VRAM Type
GDDR6
NVLINK
No

Memory

Memory Speed (MT/s)
5001.0
Memory Bandwidth (GB/s)
300.0
Memory ECC
Yes
Memory Bus Width (bit)
256

Expansion

PCIe Gen
3.0
PCIe Lanes
16
Slot Height
Full-Height / Half-Length
Slot Width
Single-Slot

Dimensions & Mounting

Dimension Height (mm)
68.9
Dimension Width (mm)
20.0
Dimension Length (mm)
168.0

Cooling

Cooling Type
Passive
Heatsink
Yes

Power

Power (max W)
70.0
Power Idle (W)
8.0
Power External
No

Features

Firmware
Controller, GPU, Microcode, Security, UEFI, UEFI Capsule, VBIOS, Vendor-Specific
Security Feature
Hardware Root of Trust, IOMMU, Measured Boot, Proprietary, Rowhammer Mitigation, Secure Boot, Secure Firmware Update, Side-Channel Attack Mitigations, UEFI Secure Boot, VBS (Virtualization-Based Security)
Workload
AI Inference, AI Training, GPU Accelerated, HPC (High Performance Computing), Machine Learning
Supported Tech
CUDA, ECC Memory, PCI Express 3.0, Tensor Cores

Documentation

Datasheet
Warranty
1 year warranty badge

Byteguard 1 year warranty

This option includes a 1-year warranty. Warranty length depends on the item and configuration.

About warranty support

The warranty stated for this option is subject to our Terms & Conditions, including the separate treatment of consumable components in clause 19.7. Consumer legal rights are unaffected. Where no warranty period is stated, see clause 19.4.

Standard on-site support

On-site warranty support is included as standard. Contact the support team first so that we can assess the fault and arrange the appropriate support. Hardware should be returned only where the team confirms this is needed. Any next-business-day service requires separate written agreement.

How to claim

Warranty claims should be raised through the support portal with the serial number, order details and a clear description of the suspected fault.

What is covered

Warranty is for faults under normal use. Coverage does not extend to misuse, unauthorised modification, operation outside specification, or software and media not supplied by Bytestock.

Full warranty terms · Warranty support portal

Selected option condition

Recertifiable · Refurbisher: Manufacturer

About ByteStock refurbishment

This describes ByteStock's refurbishment process. It does not mean a Recertifiable or manufacturer-supplied option has been refurbished or recertified by ByteStock.

This is where our engineering team shines. Every Bytestock Refurbished product undergoes a multi-point inspection, deep clean, full testing, and restoration process. We replace worn parts, update firmware, and ensure performance is as close to new as physically possible.

  • Professionally refurbished by Bytestock technicians
  • Multi-stage quality control and component testing
  • Backed by our industry-leading warranty options
  • Sustainable, reliable, and performance-guaranteed

Buying refurbished from Bytestock means getting enterprise-level performance with measurable savings and zero compromise on quality.

FAQs

Refurbished Tech Questions

Refurbished hardware is pre owned equipment that has been inspected, cleaned, tested and restored to full working order by a trained technician. Recertified hardware goes one step further, it has been processed, tested or validated by the original manufacturer or an authorised partner and is returned to a condition that meets a documented standard. Both routes deliver reliable hardware, but recertified stock carries the additional reassurance of manufacturer level checks.
Yes, when sourced from a specialist refurbisher with proper test procedures. Quality refurbished kit is fully diagnostics tested, professionally cleaned, securely wiped, updated and burn in validated. Components that fail inspection are replaced, so only hardware that passes all checks reaches sale. For business users this means performance and reliability that is very close to new, often with the added benefit of an extended warranty and a reduced purchase price.
Refurbished and recertified hardware offers significant cost savings without sacrificing performance. It also reduces electronic waste and keeps high quality enterprise grade components in circulation for longer. Many organisations choose refurbished gear because it delivers excellent value, predictable performance and longer term sustainability benefits. It is a smart way to access professional hardware without the full new equipment price tag.

Speak to an expert.

Get fast, honest guidance from our in house subject matter experts, specialists and engineers. Whether you need technical clarity or buying confidence, we’ll help you make the right choice for your business.

Bytestock Expert on a call
Articles
New

Introducing the new Bytestock website: one site for enterprise IT hardware.

Our company site and online shop are now one website. Research, configure and buy refurbished enterprise IT hardware in one place, with a new server configurator, spare-parts finder and the...

Introducing the new Bytestock website: one site for enterprise IT hardware

AI infrastructure explained, how artificial intelligence workloads are reshaping enterprise IT hardware.

AI workloads demand faster, more specialised infrastructure. Learn how artificial intelligence is reshaping enterprise hardware and what to look for when building AI-ready systems.

AI infrastructure explained, how artificial intelligence workloads are reshaping enterprise IT hardware

Office relocation IT checklist, common IT mistakes when moving offices, and how to avoid downtime.

Moving offices soon? Discover the five most common IT mistakes businesses make during office relocations and how a simple checklist can help you avoid downtime.

Office relocation IT checklist, common IT mistakes when moving offices, and how to avoid downtime

NVIDIA A100 vs H100 GPUs explained, architecture differences, AI performance, and choosing the right data centre GPU.

NVIDIA A100 vs H100 explained. Learn the key differences in architecture, AI performance, and real-world use cases to choose the right data centre GPU for your workloads.

NVIDIA A100 vs H100 GPUs explained, architecture differences, AI performance, and choosing the right data centre GPU

IT procurement strategy explained, improving efficiency, controlling costs, and reducing operational risk.

Is your IT procurement strategy costing more time and money than it should? Learn how a procurement productivity scorecard helps optimise spend, improve efficiency, and reduce operational risk.

IT procurement strategy explained, improving efficiency, controlling costs, and reducing operational risk

Server capacity planning explained, forecasting demand, optimising performance, and scaling infrastructure effectively.

Server capacity planning helps businesses forecast demand, optimise performance, and scale infrastructure without disruption. Learn how monitoring, forecasting tools, and refurbished hardware support reliable, cost-effective growth.

Server capacity planning explained, forecasting demand, optimising performance, and scaling infrastructure effectively