GPU Cluster - Interconnect HPC Systems with GPU to Form a Cluster
A GPU cluster is not an average computer network — it is a computer cluster setup that can provide a maximum of computing power for all kinds of graphical computations and evaluations. Learn more about the versatility of our GPU clusters.
special highlight
High-Performance IoT SuperServer
48V DC Power Input
- 2U Rackmount Server, up to 40 Cores 270W TDP
- Dual Socket P+ Intel® Xeon Scalable Processors 3rd Gen.
- Up to 12TB RAM, 3200MHz ECC DDR4
- 6x 2.5 hot-swap Drive bays
- 4 PCI-E 4.0 x16 slots (optional)
- 6x Heavy duty Hot-swap fans with optimal speed control
- 1300W Redundant DC Power Supplies
special highlight
Up to 8TB RAM, 32 DIMM slots
Option for 3 Double-Slot GPUs
- 2U Rackmount Server, up to 40 Cores 270W TDP
- Dual Socket P+ Intel® Xeon Scalable Processors 3rd Gen.
- Up to 12TB RAM, 3200MHz ECC DDR4
- 6x 2.5 Hot-swap Hybrid Drive bays
- 4 PCI-E 4.0 x16 slots with GPU optional
- 6 heavy duty Hot-swap Fans with optimal speed control
- 2000W Redundant AC Power Supplies
special highlight
Multi-Node GPU Server
Up to 3 PCI-E GPUs per Node
- 2U Rackmount Server, up to 270W TDP
- Single Socket P+, 3rd Gen Intel Xeon Scalable Processors
- 8x DIMM slots, up to 2TB RAM DDR4-3200MHz
- 2x 2.5 Inch hot-swap NVMe drive bays
- 3x PCI-E 4.0 x16 FHFL DW slots
- 2x M.2 slots and AST2500 BMC
- 2x 2600W redundant power supplies (Titanium Level)
- 2U Rackmount Server, 280W cTDP
- Single AMD Ryzen Threadripper Pro 5000WX/3000WX Processors
- 8x DIMM slots, up to 2TB RAM DDR4-3200MHz
- 2x 2.5 Inch hot-swap drive bays
- 2x PCI-E 4.0 x16 External slots, 1x AIOM
- Integrated IPMI 2.0 + KVM with dedicated LAN
- 2x 2600W Redundant Power Supplies (Titanium Level)
special highlight
Up to 2 Add-on cards on 2U
- 2U Rackmount Server, up to 280W TDP
- Single Socket SP3, AMD EPYC 7003 Series Processors
- 8x DIMM slots, up to 2TB RAM DDR4-3200MHz
- 2x 2.5 hot-swap U.2 NVMe drive bays
- 6x PCI-E 4.0 x16 slots and 1x PCI-E AIOM
- Integrated IPMI 2.0 + KVM with dedicated LAN
- 2600W redundant power supplies with PMBus
- 4U Rackmount Server, up to 205W TDP
- Dual Socket P, 2nd Gen Intel Xeon Scalable Processors
- 24x DIMM slots, up to 6TB RAM DDR4-2933MHz ECC
- 8x GPU system for cloud computing
- 24x 2.5 Inch hot-swap SAS/SATA drive bays
- 8x PCI-E 3.0 x16 slots, 3x PCI-E 3.0 slots
- 4x 2000W redundant power supplies (Titanium Level)
special highlight
NVIDIA HGX A100 4-GPU 40GB/80GB, NVLink
- 2U Rackmount Server, up to 280W TDP
- Dual AMD EPYC Series Processors, 3rd Gen
- 32x DIMM slots, up to 8TB RAM DDR4 3200MHz
- 4x 2.5 hot-swap Drive bays
- Direct connect PCI-E Gen4 platform with NVIDIA® NVLink™ v3.0
- Onboard BMC supports integrated IPMI 2.0 + KVM
- 2x 2200W redundant Power supplies (Platinum Level)
special highlight
4U GPU Server up to 10 GPUs
- 4U Rackmount Server, up to 205W TDP
- Dual Socket P, 2nd Gen Intel Xeon Scalable Processors
- 24x DIMM slots, up to 6TB RAM DDR4-2933MHz ECC
- 24x 2.5 Hot-swap SAS/SATA drive bays
- 11x PCI-E 3.0 x16 slots and 1x PCI-E 3.0 x8
- Up to 10x NVIDIA Quadro/Tesla GPUs
- 4x 2000W redundant power supplies (Titanium Level)
special highlight
Up to 10 PCI-E GPUs double width FHFL
- 4U Rackmount Server, up to 270W TDP
- Dual Socket P+, 3rd Gen Intel Xeon Scalable Processors
- 32x DIMM slots, up to 12TB RAM DDR4-3200MHz ECC
- 24x 2.5 hot-swap NVMe/SATA/SAS drive bays
- 12x PCI-E 4.0 x16 slots
- 8x Heavy duty fans with optimal fan speed control
- 4x 2000W redundant power supplies (Titanium Level)
special highlight
GPU Server
8x NVIDIA V100 SXM2 GPUs on 4U
- 4U Rackmount Server, up to 205W TDP
- Dual Socket P, 2nd Gen Intel Xeon Scalable Processors
- 24x DIMM slots, up to 6TB RAM DDR4-2933MHz ECC
- Up to 8x NVIDIA V100 SXM2 GPUs
- 16x 2.5 hot-swap SAS/SATA drive bays
- 6x PCI-E 3.0 x16 slots
- 4x 2200W redundant power supplies (Titanium Level)
special highlight
A+ Rack Server
Up to 8x NVIDIA A100 Gen4, 6x NVLink Switch
- 4U Rackmount Server, up to 280W TDP
- Dual AMD EPYC 7003/7002 Series Processors
- 32x DIMM slots, up to 8TB RAM DDR4 3200MHz
- 6 Hot-swap U.2 NVMe 2.5 Drive bays
- Supports NVIDIA® HGX™ A100 8-GPU
- 8x PCI-E 4.0 x16 via PCI-E Switch
- 4x 2200W Redundant Power supplies Platinum Level
special highlight
8 GPUs NVIDIA® NVLink™ with NVSwitch™
- 4U Rackmount Server, up to 270W TDP
- Dual Socket P+, 3rd Gen Intel Xeon Scalable Processors
- 32x DIMM slots, up to 12TB RAM DDR4-3200MHz
- 6x 2.5 Hot-swap NVME/SATA/SAS Drive Bays
- 10x PCI-E 4.0 x16 LP slots
- 4x Heavy duty fans with optimal fan speed control
- 4x 2200W redundant Power supplies (Titanium Level)
special highlight
HGX A100 8-GPU 40GB/80GB SXM4 Multi-GPU
- 4U Rackmount Server, up to 280W TDP
- Dual Socket SP3, AMD EPYC 7003 Series Processors
- 32x DIMM slots, up to 8TB RAM DDR4-3200MHz
- 6x 2.5 Inch hot-swap U.2 NVME drive Bays
- 8x PCI-E 4.0 x16 slots and 2x PCI-E 4.0 LP slots
- 4x hot-swap heavy duty cooling fans
- 4x 3000W redundant power supplies (Titanium Level)
special highlight
Liquid Cooling GPU Server
- 2U Rackmount Server, up to 280W TDP
- Dual Socket SP3, AMD EPYC 7003 Series Processors
- 32x DIMM slots, up to 8TB RAM DDR4-3200MHz
- 4x NVIDIA GPU cards
- 4x 2.5 hot-swap SATA/NVMe/SAS drive bays
- 4x PCI-E 4.0 x16 LP slots
- 2x 2200W redundant power supplies (Platinum Level)
special highlight
Up to 4 GPUs in Edge Server
32 DIMMs, up to 8TB RAM
- 2U Rackmount Server, up to 270W TDP
- Dual Socket P+, 3rd Gen Intel Xeon Scalable Processors
- 32x DIMM slots, up to 12TB RAM DDR4-3200MHz
- 6x 2.5 hot-swap NVMe/SATA drive bays
- 4x PCI-E 4.0 x16 slots with GPU/Accelerator support
- 6x heavy duty hot-swap fans
- 2000W redundant AC Power Supplies with PMBus
Do you need help?
Simply call us or use our inquiry form.
What is a GPU Cluster?
A GPU cluster is a GPU computing system provided in a loose coupling for specific tasks. Just like a conventional cluster, a GPU cluster consists of five different parts:
- Master Server Access to the GPU cluster is controlled via this server.
- GPU Computing Systeme Together, all systems form the individual cluster nodes.
- Common data area All cluster members connected in the GPU cluster have access to this common data area.
- Cluster network In a GPU cluster, the cluster network is also responsible for internal communication in the cluster.
- Operating system Each cluster member is managed by its own operating system.
When are GPU clusters used?
GPU clusters are an effective solution if a clearly defined task can be subdivided into several subtasks that are to be processed in parallel with a GPU computing system. Furthermore, a GPU cluster can be used to give several employees access to a specific application in order to complete a task simultaneously and as efficiently as possible.
GPU Cluster - Configuration Example with NVIDIA GPUs
Below you can find three versions per network connection:
GPU cluster with TCP/IP network connection
| GPU Cluster with TCP/IP network connection | |||
|---|---|---|---|
| GPU CUDA Cores | 32 x 5120 | 16 x 5120 | 8 x 5120 |
| NVIDIA NVLINK | 300 GB/s | 300 GB/s | 300 GB/s |
| GPU Tensor Cores | 32 x 640 | 16 x 640 | 8 x 640 |
| CPU Processors | 16 | 8 | 4 |
| Rack height | 8 | 4 | 2 |
| Memory GPU total | 1024 GB | 512 GB | 265 GB |
| Storage CPU total | 12 TB | 12 TB | 12 TB |
| Data Storage | 16 x M.2 NVMe 16 x 1920 GB | 8 x M.2 NVMe 8 x 1920 GB | 4 x M.2 NVMe 4 x 1920 GB |
| Network | 2 x 10Base-T | 2 x 10Base-T | 2 x 10Base-T |
GPU Cluster with FDR-Infiniband Network connection
| GPU Cluster with FDR-Infiniband Network connection | |||
|---|---|---|---|
| Infiniband ERB 100 GB/s | 8 x 100 GB/s | 4 x 100 GB/s | 2 x 100 GB/s |
| Infiniband FDR 56 GB/s | 8 x 56 GB/s | 4 x 56 GB/s | 2 x 56 GB/s |
GPU Cluster Prices - Cluster System Costs at HAPPYWARE
The costs for a full GPU cluster are difficult to estimate. This is mainly due to the fact that the prices for these types of GPU systems always depend on project prices and the number of cluster nodes.
If you want to buy a custom-built GPU cluster, please do not hesitate to contact us. You will receive a fully pre-configured GPU cluster based on your favorite Linux version, implemented at a reasonable price with GPU server systems from Supermicro, ASUS, or Gigabyte.
Information for educational and research institutes: For GPU clusters that are to be used in public institutions or for research or education purposes, we can offer special terms. If you are in the process of planning a GPU cluster project, please contact us.
From planning to implementation — our expert Jürgen Kabelitz knows what is required through the process. He will be able to answer all of your questions linked to your project.
Buy or rent GPU clusters - HAPPYWARE offers these options
- Individual configuration We design and offer GPU clusters that are optimized for your applications.
- Assistance with your decision: Alongside you, we can help determine if buying or renting a GPU cluster is more beneficial for your company. From experience, we know that a purchase is recommended for a planned operating time of more than 12 months.
- Systems with full scalability If you buy a GPU Cluster System from us, you can flexibly expand it later with additional cluster nodes. In this way, the cluster grows with your requirements.
- Financing options We are able to offer different financial solutions for our GPU cluster systems. Please contact us for leasing or hire-purchase options.
If you need more information about our GPU clusters please contact our HPC expert Jürgen Kabelitz and he will be happy to answer your questions.