The ACPU-20-10 is a dual socket soldered memory (up to 128 GB) board, with 2 GPGPU PCIe slots and optional FPGA. The board can be configured in different ways:
High end high power configuration, with a peak performance of 3.3 TFlop/s peak.
Also, for applications under restricted power budget, the board can be configured on low power configuration, with an average power absorption of less than 100 W.
The board is optionally liquid cooled and can be used in a chassis that fits a standard 19’’rack and can also come ruggedized.
The board can be mounted on third party devices.
SUPERCOMPUTING CAPABILITIES – ACPU-20-10 boards are used in the Eurotech Aurora Tigon supercomputers, the greenest and most dense machines in the high performance computing market. GPU
ACCELERATION – ACPU-20-10 is able to perform at 3.3 Tflop/s. It has very fast I/O with Infiniband or Gigabit Ethernet.
ENERGY EFFICIENT – ACPU-20-10 marks a record in energy efficiency with > 3.15 GFlop/s per Watt.
MIL-STD Versions – The CPU 71-10 is offered in versions that support high shock and vibration immunity, and conduction-cooling. Conformal coating is available for all versions.
RELIABLE – No moving parts eliminate vibrations. Direct water cooling avoids hot spots while the soldered memory provides speed and robustness. FPGA – ACPU-20-10 is provided with an Altera Stratix FPGA connected with 2 links PCIe3 x8 to the CPUs.
WATER COOLED – Cooled with direct hot liquid cooling that doesn’t require air conditioning and ventilation. Pervasive cooling is applied to every component of the board to maximize the cooling effectiveness.[/tab]
• 2.8 TFlop/s per board (peak)
• 2x Intel Xeon E5 v2 Processors (Sandy Bridge) • 6,8 and 12 core options
• 20MB Intel Smart Cache • 64-Bit Instruction Set
• Up to 2.90 GHz Clockspeed (3.8 GHz w/ Intel Turbo Boost 2.0)
• Support for Intel vPro, Hyper-Threading, Virtualization, Enhanced SpeedStep, Trusted Execution, Intel 64 Technologies
ACCELERATORS (GPGPU or MIC)
• GPGPU 2 x Nvidia Kepler K20 or K40 small form factor Each 2,496 CUDA cores clock at 705 MHz (single-precision floating point performance at 3.52 TFLOP/s and double-precision 1.17 TFLOP/s) 5 GB GDDR5 memory – bandwidth of 200 GB/s.
• INTEL XEON PHI COPROCESSOR 2 x Intel Xeon Phi 5115D per node Each 60 cores/1.053 GHz/240 threads Up to 1 teraflops double-precision performance 8 GB memory and 320 GB/s bandwidth Standard PCIe x16 form factor Linux operating system, IP addressable Supported by the latest Intel® software development products 512-bit wide vector engine 32 KB L1 I/D cache, 512 KB L2 cache (per core) 8 GB GDDR5 memory (up to 320 GB/s)
• 2 CPUs connected via 8.0GT/s QPI link
• 2 accelerators connected via PCIe2 to the board
• 2 x PCIe sockets
• Up to 128 GB soldered RAM per node. ECC DDR3 SDRAM 1866 MT/s
• Stratix V SGXB6 CHIPSET
• Intel I/O hub (Patsburg)
• 1 x 40 Gbps QDR Infiniband
• 1 x 1Gbs Ethernet (Intel PXE)
• 2 x USB host (internal)
• 1 x standard VGA – Matrox MGA G200 with 8Mb RAM integrated in the BMC • Optional: 1+1 3D Torus or 3D mesh BW:up to 240+240Gbps, Latency: ~1us
• 2 x PCI Express x 16 SERIAL ATA
• 1 x 3 Gbps standard connector BIOS
• Insyde H2O
• up to 2 TB GB 2,5” Sata Disk or up to 512 TB 1,8” microSATA SSD HW MONITORING AND CONTROL
• IPMI – temperature, current (hall effect current sensor) and voltage
• CAN BUS OPERATING SYSTEMS
• Linux, Windows 7, Windows Embedded