India
Breadcrumb navigation
SX Aurora TSUBASA
- Vector processor +x86/ Linux architecture
- 1.35 Tb/s memory bandwidth
The SX Aurora TSUBASA is a cutting edge energy-saving server technology that combines the power of multiple PCIe card-based Vector Engines(VE) and Vector Hosts (VH). The VE consists of a powerful vector processor with eight vector cores, using high bandwidth memory modules (HBM2) for attaining the utmost memory bandwidth of 1.35 TB/s. This new SX architecture executes an entire application on the VE and the OS on the VH, ensuring the highest sustained performance in a well-known x86/Linux environment.
Features
Powerful computing environment
High capability core and processor with maximum memory bandwidth
The vector core on the VE processor is the most powerful single core in the High Performance Computing environment (HPC) today, with performance per core at 307 GFLOPS and memory bandwidth per core at 150 GB/s. With eight cores, the processing power multiplies leading to the highest sustained performance (2.45 TF peak ) and World's highest memory bandwidth per processor ( 1.35 TB/s).
State of the art technology
The vector processor employs 16nm FinFET process technology for extremely high performance and low power consumption
Increased ease-of-use
Compiler with automatic vectorization and parallelization
Supporting a GNU environment, the SX-Aurora TSUBASA offers Fortran/C/C++ compilers with advanced automatic vectorization and parallelization for sustained performance.
Rich scientific computing library
Scientific computing libraries optimized for SX-Aurora TSUBASA are available and include the industry-standard BLAS, FFT, LAPACK, and ScaLAPACK
NEC SX Aurora Tsubasa Architecture
NEC TSUBASA boasts of a powerful architecture to meet the demanding data-intensive workloads.
The vector processor is loaded with eight independent vector cores with 64 fully functional vector registers per core. These registers can, in turn, feed the functional units or receive results from them. This increases the efficiency of the compiler significantly as they have a lot more options to optimize the register usage. Each vector register has 256 entries, each having a width of 8 bytes. This makes the capacity of a full set of registers to 128k bytes which enables the system to run double precision data at full speed.
SX Aurora TSUBASA comes with three FMA ( Floating point Multiply Add) units with one for divide and square root per core. Div/Sqrt, wherever needed, offers faster calculation when compared to other processors in the market that employ time-consuming software to deal with such calculations. The peak performance of this super processor is 3*2*32 = 192 double-precision floating-point operations per cycle.
Memory
- Maximum memory efficiency to tackle multiple mathematical simulations quickly
- Latest ‘Last-Level-Cache’ (LLC), vector cache memory with a "write-back" feature
- Capable of shared memory parallelization, by auto parallelization/ OpenMP. The last level cache has a line-size of 128 bytes.
- Utilizes 2nd generation "High Bandwidth Memory" standard, HBM2 which can stack four or eight dies together, to achieve 200 GB/s bandwidth while providing either 4 or 8 GB of capacity. Six of these memory blocks and the Vector Engine Processor are connected by a "Silicon interposer", a special die to mount on and connect memory and processor.
- Constitutes of "Silicon interposer", a special die to mount on that connects memory and processor, which provides a total of 24 GB to 48 GB per Vector Engine with 1.35 TB/s bandwidth.
Platforms
- A100-1: Tower server with one VE card of type 10C.
- A300-2: Single socket 1U rack-mountable server with up to two VE cards
- A300-4: A dual-socket 1U rack-mountable server equipped with up to four VE cards of type 10B or 10C.
- A300-8: A dual-socket 4U rack-mountable server with up to eight VE cards of type 10B or 10C.
- A500-64: Proprietary rack equipped with 64 VEs of type 10A or 10B
Applications
University, research laboratories
Power research facilities at R&D units and large scale supercomputer centres by enabling fast data crunching and analysis mechanisms.
Nature
Create a holistic technological environment to study and analyse climate change, weather forecasting, disaster prevention, and mitigation, resource exploration
Manufacturing
Assist in all civil projects and help engineers and research developers in analysing the strength, composition, and durability of structures, fluids, and supporting new material development
Social Infrastructure
Empower safe, connected and smart cities through AI, IoT, Image analysis and new energy research
Healthcare and life science
Support the health care industry in advancing research facilities to develop bio healthcare, enable drug discoveries, and empower gene analysis.
Marketing
Improve customer value, product targeting and market segmentation by streamlining big data analytics, financial analytics and building next-generation distribution technology
Sources
Contact US