GPU servers—high-speed engines for simulations

In the 2U GPU server ESC4000 G4 with Intel processors from the Xeon family, four double-deck GPU cards do the actual floating-point computing work. Its RAID mass storage with eight LFF SATA hard disk bays, the redundant power supplies and its 8+3 PCIe slots complete the coherent equipment list. The fast server offers terrific computing performance, enormous scalability and high energy efficiency.High-density computing is specifically designed to fit more computing power into one rack. The two-rack ESC4000 G4 easily does this by accommodating up to four GPUs for extremely high power density. The ESC4000 G4 supports up to eight PCIe Gen3 x16 slots. This results in high expandability and great compatibility with many expansion cards. Thus, the server works with a large number of GPU cards (NVIDIA Tesla, GridTM and AMD FirePro S series). In addition, there is also a PCIe Gen3 x 24 slot, which can also fit two low-profile HBAs or proprietary network cards with FDR and 10 GbE interface by means of a riser board.

The four height units occupying GPU server from ASUS works with two Intel Xeon processors and can be equipped with up to 3 terabytes of RAM. The eight front-accessible drive bays are unceremoniously docked to the SATA controller. Optionally, there is also a controller for fast SAS disks and SSDs. In addition, the ESC8000 G4 also offers two M.2 interfaces for SSDs (SATA 6 Gbit/s & PCIe Gen3 x4 link 22110/2280/2260), as well as two NVMe slots. (If the latter are used, the number of usable SATA drives is reduced from eight to six). Of its ten PCIe expansion slots, up to eight bays can be used for GPU cards. NVIDIA Tesla V100s are suitable for this, for example. Fully equipped, the fast computing machine achieves around 112 teraflops (Floating Point Operations Per Second) according to Single Precision calculations. For comparison: A server without GPU support calculates floating point operations about 200 times slower.

The four height units occupying GPU server from ASUS works with two Intel Xeon processors and can be equipped with up to 3 terabytes of RAM. The eight front-accessible drive bays are unceremoniously docked to the SATA controller. Optionally, there is also a controller for fast SAS disks and SSDs. In addition, the ESC8000 G4 also offers two M.2 interfaces for SSDs (SATA 6 Gbit/s & PCIe Gen3 x4 link 22110/2280/2260), as well as two NVMe slots. (If the latter are used, the number of usable SATA drives is reduced from eight to six). Of its ten PCIe expansion slots, up to eight bays can be used for GPU cards. NVIDIA Tesla V100s are suitable for this, for example. Fully equipped, the fast computing machine achieves around 112 teraflops (Floating Point Operations Per Second) according to Single Precision calculations. For comparison: A server without GPU support calculates floating point operations about 200 times slower.

The ASUS GPU server, which occupies four height units, works with two AMD EPYC 9004 processors and can be equipped with up to 6 terabytes of RAM. The eight front-accessible drive bays are simply docked onto the SATA controller. There is also an optional controller for fast SAS disks and SSDs. The ESC8000A-E12 also offers an M.2 interface for SSDs (PCIe Gen3 x4 link 22110/2280/2260). Up to eight of its ten PCIe expansion slots can be used for GPU cards. NVIDIA H100, for example, is suitable for this. Fully equipped, the fast computing machine achieves around 480/240 teraflops (FP32/FP64)