中文/English
Contact Sales
AI COMPUTE INFRASTRUCTURE

Dedicated GPU servers
for AI training
and inference.

Explore GPU models, server configurations and pricing. Contact us to confirm server quantities, rental duration and delivery arrangements.

View PricingExplore GPUs

Availability and exact configurations are confirmed before delivery.

Dedicated resourcesFind a server for your workload
01
Rental optionsPlan around your actual requirements
02
Delivery coordinationConfirm configuration and arrangements
03
GPU Servers

Compute for your next breakthrough.

Single-GPU configurations for training, inference and visual computing.

Discuss configuration
Availability on request

NVIDIA Hopper · Single GPU

H100 PCIe

For AI inference, model development and HPC.

GPU count
1–8 GPUs
GPU memory
80 GB HBM2e
System RAM
180 GB
Memory Bandwidth
2 TB/s
FP64
26 TFLOPS
Tensor Core TF32
756 TFLOPS*
Tensor Core FP16
1,513 TFLOPS*
View Specs+

Reference configurations include 1–8 H100 PCIe GPUs (80 GB HBM2e each); confirm system RAM and other server specifications for the selected configuration. * Tensor Core TF32 / FP16 figures are theoretical GPU peaks with structural sparsity, not application benchmarks. Confirm CPU, storage, networking and availability.

Discuss configuration
Availability on request

NVIDIA Hopper · Single GPU

H100 SXM

For model training, fine-tuning and high-throughput inference.

GPU count
1–8 GPUs
GPU memory
80 GB HBM3
System RAM
200 GB
Memory Bandwidth
3.35 TB/s
FP64
34 TFLOPS
Tensor Core TF32
989 TFLOPS*
Tensor Core FP16
1,979 TFLOPS*
View Specs+

Reference configurations include 1–8 H100 SXM GPUs (80 GB HBM3 each); confirm system RAM and other server specifications for the selected configuration. * Tensor Core TF32 / FP16 figures are theoretical GPU peaks with structural sparsity, not application benchmarks. Confirm CPU, storage, networking and availability.

Discuss configuration
Availability on request

NVIDIA Hopper · Single GPU

H200 SXM

Larger GPU memory for LLMs and memory-intensive workloads.

GPU count
1–8 GPUs
GPU memory
141 GB HBM3e
System RAM
300 GB
Memory Bandwidth
4.8 TB/s
FP64
34 TFLOPS
Tensor Core TF32
989 TFLOPS*
Tensor Core FP16
1,979 TFLOPS*
View Specs+

Reference configurations include 1–8 H200 SXM GPUs (141 GB HBM3e each); confirm system RAM and other server specifications for the selected configuration. * Tensor Core TF32 / FP16 figures are theoretical GPU peaks with structural sparsity, not application benchmarks. Confirm CPU, storage, networking and availability.

Discuss configuration
Availability on request

Blackwell Server Edition · Single GPU

RTX PRO 6000

Blackwell Server Edition for AI inference and rendering.

GPU count
1–8 GPUs
GPU memory
96 GB GDDR7
System RAM
218 GiB
Memory Bandwidth
1,597 GB/s
Tensor Core TF32
234 TFLOPS
Tensor Core FP16
1 PFLOP
View Specs+

Reference configurations include 1–8 RTX PRO 6000 Blackwell Server Edition GPUs (96 GB GDDR7 each); confirm system RAM for the selected configuration. Tensor performance follows the values published on NVIDIA’s product page. GPU figures are theoretical peaks, not application benchmarks. Confirm CPU, storage, networking and availability.

Discuss configuration
Availability on request

NVIDIA Blackwell · Multi-GPU

B200 SXM

For large-scale AI training, inference and HPC.

GPU count
1–8 GPUs
GPU memory
180 GB HBM3e
Memory Bandwidth
Up to 8 TB/s
FP64
37 TFLOPS
Tensor Core TF32
2.25 PFLOPS*
Tensor Core FP16
4.5 PFLOPS*
View Specs+

Per-GPU B200 SXM specifications. FP64, TF32 and FP16 performance are derived from NVIDIA's published eight-GPU HGX B200 totals; * TF32 / FP16 figures are theoretical peaks with structural sparsity. Confirm server configuration and availability.

Discuss configuration
Availability on request

NVIDIA Blackwell Ultra · Multi-GPU

B300 SXM

Blackwell Ultra accelerator for large-model training and inference.

GPU count
1–8 GPUs
GPU memory
288 GB HBM3e
Memory Bandwidth
Up to 8 TB/s
FP64
1.2 TFLOPS
Tensor Core TF32
2.25 PFLOPS*
Tensor Core FP16
4.5 PFLOPS*
View Specs+

Per-GPU B300 SXM specifications. NVIDIA lists 10 TFLOPS FP64 for the eight-GPU HGX B300, converted here to a per-GPU value; * TF32 / FP16 figures are theoretical peaks with structural sparsity. Confirm server configuration and availability.

Discuss configuration

System RAM represents a reference configuration; confirm availability. GPU figures are manufacturer-published theoretical peaks. * H100 / H200 Tensor figures use structural sparsity; RTX PRO 6000 figures follow NVIDIA’s product page.

How we work

Focus on your models.
Start with clear requirements.

From one server to multiple independent servers, move your compute plans forward with clear configurations and communication.

01

Dedicated Servers

Discuss a dedicated server configuration for your training, inference or development workload. For multiple servers, we confirm resources and conditions individually.

02

Rental Options

Discuss a rental plan based on your project stage, expected duration and server count. Confirm pricing units and included services with your quote.

03

Delivery Coordination

Coordinate availability, region, networking and access requirements. Once both parties confirm the configuration and quote, discuss a delivery date and arrangements.

Use cases

Compute that fits your work.

From model exploration to application delivery, start with your actual workload.

01 / Start with your data and model

Model training

Assess GPU options and server configurations around model size, memory needs and training duration.

Discuss configurations
02 / Build around your application

Model inference

Discuss server options based on model size, concurrency requirements and your deployment approach.

Discuss configurations
03 / Make room for the next idea

Experimentation

Discuss server quantities and rental periods for model validation, algorithm experiments and development.

Discuss configurations
The rental process

From requirements to a plan, in four steps.

01

Choose a configuration

Explore GPU models and define your workload and configuration needs.

02

Share your requirements

Specify server count, use case, rental duration and region.

03

Confirm the quote

Agree on configurations, pricing units and rental terms together.

04

Arrange delivery

Confirm availability and access, then coordinate delivery arrangements.

Contact Sales

A clear brief starts a better conversation.

Copy the template below to outline your configuration and rental requirements.

This is a homepage preview. Sales contact details have not been provided. Copying the template does not send or submit an enquiry.