TPole takes you to the edge of technology

Today TPole presents a solution truly at the edge of technology; as tangible proof of this statement it is useful to analyze the performance that intelligent CPU and GPU units, although it would be better to say, Dual CPU and Multi GPU, are able to offer.

GPU SuperServer SYS-821GE-TNHR

• Dual Socket E (LGA-4677): 4th Gen Intel® Xeon® Scalable processors
• Future support for the upcoming 5th Gen Intel® Xeon® processors.
• CPU-GPU Interconnect: PCIe 5.0 x16 CPU-to-GPU Interconnect
• NVIDIA SXM: HGX H100 8-GPU (80GB)
• GPU-GPU Interconnect: NVIDIA® NVLink® with NVSwitch™

CPU

As regards the CPUs, to date the maximum configuration available is with two Intel Platinum 8480+ CPUs, each of these has 56 physical Cores and manages 112 threads, the cache is 105MB and the maximum clock, the Max Turbo Frequency, reaches 3.80GHz.

RAM

The installable RAM capacity is very important, in fact it is possible to equip this server with 32 PC5-38400 4800MHz DDR5 ECC RDIMM modules of 128GB each.

GPU

An NVIDIA® HGX™ H100 module with 8x NVIDIA® H100 with 640GB HBM3 Memory is installed on this server, each single module therefore has 80GB of RAM available.
The data in the table below unquestionably explains the potential of this platform.

It must be taken into account that all this “intelligence” communicates very quickly, keep in mind that the communication between the CPUs and Nvidia HGX takes place on PCIe gen lines. 5.0.

Communication between the H100 modules occurs with fourth generation Nvlink and NVSwitch.

Nvlink is Nvidia’s interconnection BUS between GPUs that is decidedly faster than PCIe.

Furthermore, this latest generation allows up to 900GB/s of bandwidth for each single GPU.