Shattering the 1U Server Performance Record
Supermicro and NVIDIA® recently announced a new class of servers that combines massively
parallel GPUs with multi-core CPUs in a single server system. This unique configuration
delivers performance at least an order of magnitude better than traditional quad-core CPU-based
servers. This breakthrough technology immediately provides users with the ability to implement
tasks that were traditionally addressed only with massive supercomputers or that were simply
unsolvable. Supermicro calls this new server class, the latest in its rich history of technology
innovations as shown in Figure 1, the GPU SuperServer.
Figure 1: Supermicro Product and Market Opportunity Growth
First to Market
Supermicro demonstrated its first GPU SuperServer, the SuperServer 6016GT-GF series, at the
Computex 2009 Show in Taiwan in early June. This server series features dual Intel® Xeon®
(Nehalem) processors and two Gen2 PCI-Express x16 interfaces to support two NVIDIA
Tesla™ M1060 GPU Processors. With two double-width GPUs in PCI-E 2.0 x16 lanes, this 1U
server delivers truly non-blocking GPU performance, or up to 2 Teraflops of processing power,
which makes it the highest performing 1U server on the planet.
The SuperServer 6016GT 1U series is the first of an entire line of GPU-optimized systems
Supermicro has created to meet the requirements of the high-performance computing market. By
the end of June, Supermicro will launch a 4U/Tower system, the SW7046A-GRF, which
supports four double-width GPUs. These platforms feature Supermicro’s new Gold Level¹ (93%
efficiency) power subsystems to deliver breakthrough performance-per-watt.
1
SS6016GT-TF-TM2
SW7046A-GRF
Figure 2: Supermicro GPU SuperServers
What is GPU Computing?
GPU computing is the use of a GPU (graphics processing unit) to perform general purpose
scientific and engineering computing. The model for GPU computing is to use a CPU and GPU
together in a heterogeneous computing model (Figure 3). The sequential part of the application
runs on the CPU and the computationally-intensive part runs on the GPU. From the user’s
perspective, the application just runs faster because it is using the high performance of the GPU
to boost overall system performance.
CPU
GPU
Figure 3: Heterogeneous Computing Model
The application developer has to modify their application to take the compute-intensive kernels
and map them to the GPU. The rest of the application remains on the CPU. Mapping a function
to the GPU involves rewriting the function to expose its parallelisms and adding “C” keywords
to move data to and from the GPU.
____________________
¹ www.80Plus.org
2
GPU computing is enabled by the massively parallel architecture of NVIDIA GPUs called the
CUDA™ parallel computing architecture. The CUDA architecture consists of hundreds of
processor cores that operate together to crunch through the data set in the application.
The NVIDIA® Tesla TM 10-series GPU is built on the second generation CUDA architecture with
features optimized for scientific applications such as IEEE standard double precision floating
point hardware support, local data caches in the form of shared memory dispersed throughout the
GPU, and coalesced memory accesses.
Learn more about vertical applications that take advantage of the GPU at:
http://www.nvidia.com/object/vertical_solutions.html
NVIDIA provides a complete range of software tools to program the GPU, including a C
compiler toolchain, libraries, and other tools. These are listed at:
http://www.nvidia.com/object/tesla_software.html
History of GPU Computing
Graphics chips started as fixed function graphics pipelines. Over the years, these graphics chips
became increasingly programmable, which led NVIDIA to introduce the first GPU or Graphics
Processing Unit. In the 1999-2000 timeframe, computer scientists in particular, along with
researchers in fields such as medical imaging and electromagnetics started using GPUs for
running general purpose computational applications. They found the excellent floating point
performance in GPUs led to a huge performance boost for a range of scientific applications. This
was the advent of the movement called GPGPU or General Purpose computing on GPUs.
The problem was that GPGPU required using graphics programming languages like OpenGL and
Cg to program the GPU. Developers had to make their scientific applications look like graphics
applications and map them into problems that drew triangles and polygons. This limited the
accessibility of tremendous performance of GPUs for science.
NVIDIA realized the potential to bring this performance to the larger scientific community and
decided to invest in modifying the GPU to make it fully programmable for scientific applications
and added support for high-level languages like C and C++. This led to the CUDA general
purpose parallel computing architecture for the GPU.
CUDA Parallel Computing Architecture and Programming Model
The CUDA parallel computing architecture is accompanied by the CUDA parallel programming
model that provides a set of abstractions that enable expressing fine-grained and coarse-grained
data and task parallelism. The programmer can choose to express the parallelism in high-level
languages such as C, C++, Fortran or driver APIs such as OpenCL™ and DirectX™ Compute.
3
Figure 4: CUDA Programming Model
The first language support NVIDIA provided was for the C language. A set of C for CUDA
software development tools enable the GPU to be programmed using C with a minimal set of
keywords or extensions. Support for Fortran, OpenCL, and others will follow soon.
The CUDA parallel programming model guides programmers to partition the problem into
coarse sub-problems that can be solved independently in parallel. Fine grain parallelism in the
sub-problems is then expressed such that each sub-problem can be solved cooperatively in
parallel.
The CUDA parallel computing architecture and the corresponding CUDA parallel computing
model are now widely deployed with hundreds of applications and over one thousand published
research papers. CUDA Zone lists many of these applications and papers.
Widest Variety of GPU-Optimized Servers
Supermicro has developed the widest variety of GPU SuperServers in the industry. These include
the SS6016GT series of 1U SuperServers announced at Computex and the SW7046A series of
4U/Tower SuperWorkstations. These systems are identified in bold in Table 1 below together
with the maximum number of double-width GPUs, and expansion slots that they can support
simultaneously.
Supermicro’s GPU SuperServer family also includes a growing number of systems that can be
customized using Supermicro’s unique Server Building Block Solutions®. These are unique
combinations of Supermicro’s servers, motherboards, and other components that are optimized
for the customer’s specific GPU application. These Server Building Block Solutions® are
identified in Table 1 below by the number of dual-height GPUs and expansion slots that they can
support for their customer application.
4
* Some Gen2 PCI-E x8 in x16 slot
** requires video output card
Table 1: Supermicro GPU SuperServers (Bold) and Server Building Block Solutions®
(All Expansion Slots are Full-Height, Full-Length unless otherwise indicated)
Initially one 1U and three 4U/Tower chassis support GPU computing. A wide variety of
Supermicro’s next-generation motherboards supporting the Intel® Xeon® Processor 5500 Series
(Nehalem) are also available as Building Blocks.
_________________________________
More information on these Supermicro chassis is available at: http://www.supermicro.com/products/chassis/
Additional motherboard information can be found at: http://www.supermicro.com/products/motherboard/Xeon1333/
Supermicro’s GPU SuperServer webpage is: http://www.supermicro.com/GPU/
5
Advantages of Supermicro GPU SuperServers
Supermicro’s GPU SuperServer family provides many feature advantages of interest to the user
community.
Widest Selection
Supermicro’s wide line of GPU systems, outlined in Table 1, can satisfy any GPU
Supercomputing server application. The family comprises a wide variety of 1U and 4U/Tower
systems, chassis and motherboards.
Highly Reliable Thermal Optimization
Supermicro’s advanced thermal subsystem designs optimize the cooling elements of the server
with the thermal characteristics of the GPU, CPU, motherboard, and other components. Each
thermally controlled fan module is 1+1 paired in a counter rotating matched set, which
maximizes the cooling effect. This design enhances system reliability and simplifies
maintenance. In addition, by monitoring both CPU and GPU thermal readings though the I²C
bus, Supermicro’s unique intelligent cooling control and auto fan-speed adjustment keeps the
entire system in stable operation.
Direct GPU Connect Architecture
The Supermicro design architecture provides direct Gen2 PCI-E x16 non-blocking connectivity
to each GPU to take advantage of the full bandwidth of the GPU. No extra cabling is required,
since the GPUs are connected directly to the server motherboard via riser cards, thus improving
reliability, airflow, and maintenance, as well as reducing costs.
Industry-Leading Power Efficiency
Supermicro’s GPU SuperServer platforms feature Gold Level (93% efficiency) power
subsystems, along with high-efficiency motherboard and thermal designs to deliver breakthrough
performance-per-watt and increased system reliability.
Flexible Networking Connectivity
The SS6016GT GPU SuperServer series includes a single half-height PCI-E Gen2 slot for
additional connectivity. The SW7046A-GRF 4U/Tower server with 4 double-width GPUs has 3
additional PCI-E expansion slots for add-on cards.
Advanced System Management
Supermicro’s advanced IPMI capability allows the GPU system to be managed directly via the
remote monitoring / controlling features. All the key elements of the server system- CPU, GPU
and power supply- can be remotely monitored, along with Supermicro standard onboard IPMI
functionalities.
6
Applications and Industries
This new line of highly parallel, multi-core, multi-GPU systems is an excellent choice for an
extensive range of graphics and computationally intensive applications in a wide variety of
industries. In general, these systems are expected to make the fastest teraflop clusters much more
affordable and accessible for users throughout the world.
Applications requiring high arithmetic intensity, such as dense linear algebra, partial differential
equations, n-body problems, and finite difference formulas, can be easily accommodated using
these Supermicro SuperServers. High bandwidth problems such as sequencing (virus scanning,
genomics), sorting, and database problems are also potential applications. Finally, visual
computing problems such as graphics, image processing, tomography, and machine vision, being
original applications for GPUs, are especially accessible.
Fields where these types of problems are common and can benefit from Supermicro’s GPU
SuperServers include medical, energy, telecommunications, finance, science, and engineering in
a wide array of application areas such as:
- Computational Chemistry
- Genomics
- Fluid Dynamics
- Medical Imaging
- Digital Content Creation
- Oil & Gas Exploration
- Electronic Device Automation
- Research & Scientific Computing
- Financial Markets Modeling
- Signal Processing
- Game Physics
- Weather Simulation
Below in Figure 5 are some application examples from Supermicro’s GPU partner NVIDIA®:
Figure 5: Example GPU Applications
http://www.nvidia.com/object/cuda_home.html#
7
Conclusion
Supermicro’s GPU SuperServer family includes powerful 1U and 4U systems integrating up to
four GPUs. These server systems offer many feature advantages applicable to a wide range of
problems in many industries. Supermicro and GPU technology partner NVIDIA have
aggressively positioned themselves to lead this exciting new business, beginning with the
SS6016GT GPU SuperServer series, which shatters the 1U server performance record.
Additional Resources
Those wishing to learn more about this topic may find the following resources helpful places to
start:
•
Supermicro GPU: http://www.supermicro.com/GPU/
•
YouTube video (Introducing the World's Fastest 1U Server):
http://www.youtube.com/watch?v=KGoT7C8y5rQ
•
GPGPU.org is a central resource for GPGPU news and information: http://gpgpu.org/
•
Dr. Dobb’s Journal discusses applications software for GPUs:
http://www.ddj.com/hpc-high-performance-computing/206900471
•
NVIDIA CUDA Zone: http://www.nvidia.com/object/cuda_home.html
•
NVIDIA Tesla Vertical Applications:
http://www.nvidia.com/object/vertical_solutions.html
•
NVIDIA Tesla CUDA Software Development Tools:
http://www.nvidia.com/object/Tesla_software.html
•
Who’s working on GPU apps: “GPGPU People”
http://www.gpgpu.org/w/index.php/GPGPU_People
For further information regarding Supermicro GPU systems and solutions please contact:
Supermicro UK
The Kinetic Centre
Theobald Street
Borehamwood
Hertfordshire
United Kingdom, WD6 4PJ
Tel: 0208 387 1398
Email: jasg@supermicro.co.uk or manib@supermicro.co.uk
8