Desktop Deployed Supercomputing
FPGA-computers in small to medium configurations
RIVYERA S3-5000 DDS has been created to deploy supercomputing to your desktop. It can be equipped with up to 4 RIVYERA FPGA-cards with e.g. 32 Spartan-3 5000 FPGAs. One or multiple I/O cables connect this FPGA-computer to an in-system off–the–shelf server based on an eATX mainboard. The low energy-consumption in combination with the improved bus-system makes the RIVYERA DDS a perfect choice for individuals and workgroups. Featuring RIVYERA scalability and APIs, a cluster setup can easily grow beyond a single machine when additional resources are needed. Code is 1:1 portable from RIVYERA DDS to any machine of the RIVYERA architectural family.
Main application areas include - but are not limited to - bioinformatics, financial computing, scientific or engineering simulation, cryptography, electrical engineering and many others.
|typical RIVYERA DDS setup
||fully equipped DDS machine
- 2 RIVYERA S3-5000 FPGA Cards
- 16 Spartan-3 5000 FPGAs
- 512 MByte DRAM, up to 512 GByte based SDHC Flash
- 80 mio. System Gates, more than 0.5 mio Slices
- 29.9 Mbit BlockRAM, 1664 18x18 multipliers
- Intel® Corei7 950 or better
- 12 Gbytes of ECC DDR3 RAM
- 1 TByte Hard Disc Drive
- 4 RIVYERA S3-5000 FPGA Cards
- 32 Spartan-3 5000 FPGAs
- 1 GByte DRAM, up to 1 TByte based SDHC Flash
- 160 mio. System Gates, 1.06 mio Slices
- 59.9 Mbit BlockRAM, 3328 18x18 multipliers
- Intel® Corei7 950 or better
- 24 Gbytes of ECC DDR3 RAM
- 10 TByte Hard Disc Space (bto)
|Already with such a reduced setup, a RIVYERA provides several hundred CPU cores performance.
Introducing Xilinx "Spartan" FPGAs
Todays general purpose processing units are designed to provide a processor which is optimized for general purpose applications. Which means that you have to optimize your code for the processor as it is physical predefined silicon which can not me modified to fit to your application. The Field Programmable Gate Array, in short "FPGA", is different. They allow you to modify the silicon via software configuration to be the perfect Application Specific Circuit or short "ASIC" for your personal application. Therefore they are like your private 'ASIC' fabrication via reconfigurable logic. And further more you can speedup your application inside each Xilinx FPGA using typically hundreds of parallel cores.
Faster per clock.
Usually multiple parallel processing elements are synthesized into an FPGA. Each Field Programmable Gate Array typically provides more computational performance than a higher clocked general purpose CPU or GPUs. Furthermore RIVYERAs unique architecture allows to scale this computational performance to a massively parallel level. Allowing automatic algorithm distribution and concurrent applications of multiple users on different FPGAs. Programmable Clock generators allow high level clock optimization to maximize computational performance.
Benefits of two worlds
The Hex-Core Intel "Gulftown" or Quad-Core Intel “Nehalam” processor has been combined with a high performance FPGA Super Computer. Joining the benefits of Intels single-die, 64-bit architecture makes 8MB of fully shared L3 cache readily available to each of the processor cores. This fully scalable FPGA computing setup allows to accelerate computationally expesive applications dramatically. The result is fast access to cache data via a SciEngines unique Interface or open standard interface structures like PCI Express. Combine that with the other technological advances and you get a supercomputer which is up to five orders of magnitude faster than current general purpose architecture.
More I/O bandwidth.
The new RIVYERA Family delivers up to twice the overall I/O bandwidth of the previous generation. Users can choose from one up to seven 1-lane (x1) or 4-lane (x4) wide PCI Express interface cards as well as from SciEngines unique ultra high bandwidth memory coupling. Because we provide mainboards with fully independant PCIe-slots, the bandwidth isn’t shared. So you get all the bandwidth you need by use of the latest I/O cards, e.g. 10Gb Ethernet cards, multiport 4Gb Fibre Channel cards, or InfiniBand interface cards for connecting your FPGA super computer.
Increased performance per watt.
The new RIVYERA delivers up to 50 percent reduction in idle power consumption and provides percent better performance per watt over the previous generation. A peak power consumption of 650 Watts allows you to save enough energy to run applications on those ordinary PCs you may what to keep. Compared to standard PC clusters you can save up to 500% energy depending onto the application. How do you get more performance with less energy? For starters, the FPGA is built on an industry-leading technology process allowing your algorithm to utilize typically 97.3 percent of the silicon area for your application. Therefore no idle units have to be powered and in a fully pipelined and optimized design it delivers a result every clock cycle. Therefore RIVYERA delivers an outstanding energy efficiency and performance. A smart system design combined with modern power management in RIVYERA, means a supercomputer that’s less expensive to power and cool.
RIVYERA Upgrade Cards
The RIVYERA architecture is highly scalable and customers may have initially chosen a computer e.g. with only 1 card but later require additional processing power. Rather than having to buy an additional computer, simply obtain one or more RIVYERA FPGA cards and slide them into one of the available slots in the backplane. It's as easy as adding a new PCIe card to your personal computer at home.
Quality and Reliability
Our goal is to provide our customers the maximum of quality and reliabilty they could expect for their money. Using industry and military proven components and technics SciEngines products and especially RIVYERA offers one of markets most reliable high performance reconfigurable platforms world wide. All our products are testet in an 48h stress-test to ensure minimum failures. Continuously thermal monitoring and redundant supplies allow our architecture to ensure operation even in tough environmental conditions. Some of our systems are running 24/7 at a German governmental association without any error nor fail or need of any maintenance in 5 years of operation.
|In the box
- RIVYERA DDS
- Power cords
- I/O cable (depends on option)
- Printed and electronic documentation
- Drivers and Utilities CD–ROM
- 16 to 32 Xilinx Spartan–3 5000 (XC3S5000)
- optional: 32–GB SDHC Card per FPGA
- Supermicro off-the-shelf server grade Mainboard
- Intel Corei7, 930
- 12 GB RAM, ECC DDR3
- SATAII RAID-Controller
- max. 1 PCIe4x slot for i/o card
or 4 PCIe1x slots for i/o cards
- (standard configuration, b2o available)
- optional: up to five 2–TB SATA–II hard disc drives
- 4 identical Slots
- typically 1 occupied by FPGA cards
- dual 10⁄100⁄1000BASE–T (Gigabit) RJ–45 Ethernet interface (supporting jumbo frames)
- optional: 10G-Ethernet
- optional: fibre channel
- optional: firewire 400
- optional: firewire 800
- offered interface depends on configuration
- Output power: 865W (cooling redundant) w/ PFC
- Line voltage: universal input (100V to 240V AC), power factor corrected
- Maximum input current: 32.0A (100V to 120V) or 16.0A (200V to 240V)
- Frequency: 50Hz to 60Hz, single phase
- Certification: 80 PLUS Certified
- Regulatory (Power Supply): USA - UL listed, FCC, Canada - CUL listed, Germany - TUV Certified, Europe/CE Mark, EN 60950/IEC 60950-Compliant
- 4x 5000 RPM Hot-swap PWM Cooling Fans
- Operating temperature:
10°C to 35°C (50°F to 95°F)
- Storage temperature:
-40°C to 70°C (-40°F to 158°F)
- Operating Humidity (non condensing):
8% to 70% RH
- Storage Humidity (non condensing):
5% to 95% RH
- Optional: Military Quality Hardware
Size and weight
- Height: 45,2 cm (17.8 inches)
- Width: 17.8 cm (7.0 inches)
- Depth: 64.8 cm (25.5 inches)
- Weight: depends on configuration, e.g. 56 lbs (25.4 kg) base configuration
- Communication Test
- Communication Framework
- Target Application Specific Software Packages (optional)