Practical performance measurement is a key aspect when it comes to identifying potential for optimization. Benchmark analyses help you select the right HPC components for you. We therefore work closely with the sector’s technology leaders and offer the option of testing your applications using up-to-date hardware not yet available on the market.
Thanks to its many years’ experience, the MEGWARE Benchmark Center boosts the necessary methodical expertise for the meaningful comparison of figures. Our HPC and AI engineers will be more than happy to advise you on the best configuration for you and provide you with access to the following hardware and software remotely or on site at our company:
Available Hardware
- HPC servers from MEGWARE, Intel, Supermicro, Quanta, Gigabyte and ASUS
- Processors: Intel Xeon Scalable (Granite Rapids, Sapphire Rapids, Ice Lake), AMD EPYC (Bergamo, Genoa, Milan) and Huawei ARM v8.2 Kunpen 920
- Accelerators: NVIDIA Tesla H100, AMD Radeon Instinct (MI210) and Xilinx Alveo U280
- Parallel file systems from Panasas as well as IBM Spectrum Scale (GPFS), Lustre and BeeGFS
- InfiniBand interconnect EDR or HDR from NVIDIA Networks and Omni-Path from Intel
- DDR4 memory from Samsung and Micron
| Manufacturer | Type | Cores | Clock in GHz | Quantity |
| AMD | EPYC "Genoa" 9334 | 32 | 2,7 | 2 |
| AMD | EPYC "Genoa" 9534 | 64 | 2,45 | 4 |
| AMD | EPYC "Genoa" 9654 | 96 | 2,4 | 4 |
| AMD | EPYC "Milan" 7713 | 64 | 2,0 | 8 |
| AMD | EPYC "Milan-X" 7773x | 64 | 2,2 | 8 |
| Intel | Xeon "Granite Rapids AP" 6980P | 128 | 2,0 | 4 |
| Intel | Xeon "Sapphire Rapids" 8470 | 52 | 2,0 | 8 |
| Intel | Xeon "Sapphire Rapids" 8480 | 56 | 2,0 | 4 |
| Intel | Xeon Max "Sapphire Rapids HBM" 9468 | 48 | 2,1 | 4 |
| Intel | Xeon "IceLake" 8362 | 32 | 2,8 | 2 |
| Intel | Xeon "IceLake" 8368 | 38 | 2,4 | 2 |
| Intel | Xeon "IceLake" 6346 | 16 | 3,1 | 2 |
| Intel | Xeon "IceLake" 8360Y | 36 | 2,4 | 8 |
| Huawei | Kunpeng 920-4826 | 48 | 2,6 | 8 |
| Manufacturer | Type | Quantity |
| AMD | Instinct MI300X | 1 |
| AMD | Radeon Instinct MI210 | 2 |
| Nvidia | H100 | 2 |
| Nvidia | A100 | 4 |
| Xilinx Alveo | U280 (PCIe Gen4) | 1 |
| Manufacturer | Type | Capacity | Interface | Seq. Read | Seq. Write | Quantity |
| Intel | Optane DC P4800X | 375GB | U.2 | 2400MB/s | 2000MB/s | 2 |
| Micron | 7500 Pro | 15.3TB | U.3 / PCI Gen4 | 7000MB/s | 5900MB/s | 2 |
| Micron | 7500 MAX | 12.8TB | U.3 / PCI Gen4 | 7000MB/s | 5900MB/s | 2 |
| Micron | 9400 MAX | 6.4TB | U.3 / PCI Gen4 | 7000MB/s | 7000MB/s | 1 |
| SOLIDIGM | D5-P5430 | 15.36TB | U.2 / PCI Gen4 | 7000MB/s | 3000MB/s | 1 |
| SOLIDIGM | D7-1010 | 3.84TB | U.2 / PCI Gen5 | 14500MB/s | 8200MB/s | 1 |
| SOLIDIGM | D7-1030 | 3.2TB | U.2 / PCI Gen5 | 5200MB/s | 900MB/s | 1 |
| Western Digital | SN655 | 7.68TB | U.3 / PCI Gen4 | 6800MB/s | 2000MB/s | 1 |
| Manufacturer | Type | Bandwidth | Interface | Quantity |
| Cornelis Networks | OmniPath | 100GBit | PCIe3 x16 | 12 |
| NVIDIA Networks | Connect-X6 HDR InfiniBand | 200GBit | PCIe4 x16 | 4 |
| NVIDIA Networks | Connect-X4 EDR InfiniBand | 100GBit | PCIe3 x16 | 4 |
Available Software
- Operating systems: Linux (Red Hat, SUSE) and Windows HPC Server
- Compilers: Intel C and Fortran compilers, GCC
- MPI implementations : Intel MPI, Open MPI, MVAPICH
- Optimized software libraries, such as ACML and Intel MKL
- Cluster management software ClustWare® from MEGWARE
- Intel Parallel Studio Cluster Edition including Intel® VTune™ Amplifier XE, Intel® Inspector and Intel® Advisor
- Artificial Intelligence / Machine Learning Frameworks: caffe2. cntk, digits, mxnet, pytorch, tensorflow tensorrt, theano, torch
