Oil & Gas and Financial Engineering oriented HPC Platform
The increasing speed of FPGA (Field Programmable Gate Array) performance development exceeds CPU and GPU, and it easily satisfies the ever-growing requirement of high performance computing (HPC). Different from traditional FPGA design method, ClusterTech FPGA HPC solution based on Maxeler technology is a revolutionary computing method, which allows engineers use Java to do programming in FPGA card. Engineers do not need to grasp Verilog/VHDL (Verilog Hardware Description Language) programming language, which lowers the design barrier and increases development efficiency. The solution mainly aims at Oil & Gas and Financial Engineering fields (also suitable for Scientific Computing). It uses Dataflow computers with thousands of FPGA chips to perform massive scale parallel computing tasks. Our solution can greatly optimize data transfer efficiency. Meanwhile, the cost including system operation, data processing, storage and communication consumption can achieve significant reduction.
Our Strengths and Customers’ Benefits:
Powerful Computation Ability
Based on FPGA hardware platform, ClusterTech offers comprehensive, easy-to-use software platform support for programming, which makes architecture optimization for special-purpose application as easy as software programming. Engineers can use C programming language to host program running at CPU side, and use Java programming language to kernel program at FPGA side. In other words, engineers do not need to study Verilog/VHDL programming language, which can lower the design barrier and increase their development efficiency.
Tailored for Big Data Application
We provide solutions of all sizes which cover from desktop computers (MaxWorkstation) to large scale clusters (MaxRack) , including MPC X series (full FPGA configuration), MPC C series (CPU + FPGA) and MPC N series (CPU + FPGA + SFP*). For example, individual MPC-X nodes provide large memory capacities (up to 384GB) and compute performance equivalent to dozens of conventional x86 servers, which is tailored for big data application.
* SFP denotes Small Form-factor Pluggable optical module
CPU+FPGA Computing Pattern
FPGA hardware improves computation performance by deep pipeline and high parallelism. Computation-intensive part of programs can be processed by FPGA, and other parts of programs are still processed by CPU.
Saving Network Bandwidth
FPGA computing nodes are connected by customized high speed interconnecting architecture, and these nodes are connected with CPU by PCI-Express bus. FPGA Servers are connected by infiniband or high speed Ethernet cards. These architectures can maximally reduce bottleneck of network bandwidth and then boost overall system performance.
Increase Development Efficiency
The increasing computation complexity makes the serious challenge to the current computation ability. ClusterTech FPGA solution can solve the problem very easily by adopting high performance Xilinx FPGA chips that have very large capacity, very fast computing speed, and very high parallelism.
Compared with traditional computing method, FPGA computing has very clear advantages in low energy consumption with high performance computing. Therefore, our solution can reduce operational cost and achieve green computing and environmental protection.
ClusterTech has a strong team with experienced software and hardware experts. We cooperate with customers in application level and algorithm level, help customers explore potential of newest acceleration techniques and design algorithms or products based on Maxeler software and hardware platform.
- Computation intensive
- Data intensive
- Connectivity intensive
- Oil and Gas – E.g. seismic data processing, RTM(reverse time migration), etc.
- Financial Analytics – E.g. derivatives pricing, risk analysis, etc.
- Scientific Computing – E.g. Finite Difference modeling, large scale sparse matrix solving, etc.
What our Customers are saying:
quoteApplication 200-230x Speedup
quoteOperational costs resulting from given computations are dramatically reduced.
quote30x savings in space, reducing operational costs.
quoteMaxeler MaxCompiler allows the generation of very large dataflow graphs from a Java-based interface
Supported LibrariesClusterTech Finance Library ClusterTech Monte Carlo Library
Report of Maxeler technology and FPGA: JP Morgan supercomputer offers risk analysis in near real-time