Building a GPU Private Cloud can take a lot of design work and a lot of consideration based on your use case. Software can be challenging or provide underperformance if not assessed and implemented correctly.
The correct GPU alone is quite the task. You can see our guide to datacenter GPUs here: https://www.icc-usa.com/understanding-nvidia-gpus/
We’ve spent a long time integrating GPUs into private GPU clouds and there is some important building blocks we consider:
Compute – 1U Supermicro Ultra
For the compute element of the GPU cloud, we recommend using SuperMicros Ultra SuperServer. We have extensive experience building this out as the compute node. It has 32 DIMM Slots, which can fit up to 8TB DRAM. For the CPU element, it supports Dual Socket P+ (LGA-4189) 3rd Gen Intel® Xeon® Scalable Processors.
Hyper-Converged Infrastructure – 2U 4 Node Supermicro Twin Pro
For the Hyper-Converged Infrastructure, this particular server supports: Dual Socket P+ (LGA-4189) 3rd Generation Intel® Xeon® Scalable Processors, 16 DIMMs which can have up to 4TB RAM. With Intel® X710 Dual port 10GBase-T LAN onboard, the networking element is well equipped to handle many heavy workloads.
Storage – 4U 90 Bay Storage Server
For storage, there is typically a lot of deployment scenarios, this server supports up to 90 bays with the following disk configuration: 90 3.5″/2.5″ Hot-swap SAS3/SATA3 drives, 2x Fixed slim SATA SSD, 2x NVMe M.2 (form factor: 2280 and 22110).
GPU – 4U 8 x H100 GPU Server
This GPU system has NVIDIA® NVLink™ with NVSwitch™ for GPU to GPU interconnect. This server has 8-GPU compatibility to handle all the GPU workloads throughout the cloud.
NVMe – 2U 24 x NVMe Server
For the NVMe element, this server model supports 24x 2.5″ hot-swap NVMe/SATA/SAS drive bays (22x 2.5″ NVMe hybrid) and we’ve found it to be optimally a strong server for these types of configurations.
With all of that considered, you’ll be on your way to building a GPU cloud. Talk to us about building your GPU cloud.
Enquire today by using the form below, and one of our expert team will be in touch.
General Enquiry Form
More from our blog
Intel has released its latest Raptor Lake CPU’s and AMD has released a mass of its planned AMD Zen 4 processors. So what comes next?
GIGABYTE has released a large range of servers for what it is calling the exascale era. With a large focus on AI, HPC, Cloud & Big Data, the servers are gpu-centric and focus on maximizing performance.
SuperMicro has released the H13 generation servers which have been built for the AMD EPYC 9004 processors. The range includes several GPU-optimized servers, multi-node servers, Hyper systems, and CloudDC systems.