It wasn’t too long ago that GPUs only focused on graphics and single-precision performance and either didn’t support double-precision operations, or only supported them at dramatically reduced performance levels. With the AMD FirePro™ S9150 server GPU, AMD now delivers exceptional compute performance with its newest and most powerful server GPU ever built. Based on our latest Graphics Core Next architecture, the AMD FirePro S9150 delivers maximum double-precision performance, driving circles around the competition, offering up to 77% more double-precision performance than the Tesla K40.1
Our performance superiority has recently been demonstrated using DGEMM, with performance measurements taken from a single AMD FirePro S9150 GPU. DGEMM, or Double-precision GEneral Matrix-Matrix, measures the floating point execution rate for double precision real matrix-matrix multiplication. DGEMM computations are part of the BLAS (Basic Linear Algebra Subprograms), a specified set of low-level subroutines that perform common linear algebra operations such as copying, vector scaling, vector dot products, linear combinations, and matrix multiplication.
There are many real-world applications that take advantage of double-precision matrix operations. These include computational fluid dynamics, finite element analysis and structural modelling, and molecular dynamics.
With our AMD OpenCL™ BLAS implementation, we are able to achieve 2 TFLOPS of sustained DGEMM performance with the AMD FirePro S9150. This is a first for a single GPU solution, and with the Tesla K40 only achieving a theoretical 1.43 TFLOPS of peak double-precision, we are able to show the world that AMD can win by a wide margin when comparing actual measured results versus Nvidia’s theoretical performance.1 With industry-leading performance/watt2, industry-leading memory configuration3, and support for the latest OpenCL™ standards4, the AMD FirePro S9150 Server GPU is clearly unmatched when it comes to compute performance.
AMD is serious about HPC and we want to show you that we are able to lead in this space. We are focused, we are committed, and most importantly, we are here to stay.
The AMD FirePro S9150 server GPU is available for purchase today. Please visit http://ift.tt/UWK5nw for more details.
JC Baratault is a senior business development manager, Global GPU Computing for professional graphics at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
OpenCL and the OpenCL logo are trademarks of Apple Inc. used by permission by Khronos. PCI Express is a registered trademark of PCI-SIG.
1.AMD FirePro™ S9150 delivers up to 2.53 TFLOPS peak double precision floating point performance, and Nvidia’s highest server GPU in the market as of June 2014 is the Tesla K40 with up to 1.43 TFLOPS peak double precision. Visit http://ift.tt/L3fL2T for Nvidia product specs. FP-95
2. AMD FirePro™ S9150 max power is 235W and delivers up to 2.53 TFLOPS peak double and up to 5.07 peak single precision floating point performance. Nvidia’s highest performing server cards in the market as of June 2014 are the Tesla K40, max power of 235W, with up to 1.43 TFLOPS peak double and up to 4.29 peak single, and the K10, max power 225W, with up to 4.58 TFLOPS peak single and 190 GFLOPS peak double precision. Visit http://ift.tt/L3fL2T for Nvidia product specs. FP-97
3. AMD FirePro™ S9150 features 16GB GDDR5 memory, and Nvidia’s highest performance server GPU in the market as of June 2014 is the Tesla K40 with 12GB GDDR5 memory. Visit http://ift.tt/L3fL2T for Nvidia product specs. FP-98
4. OpenCL™ 1.2 conformance expected for S9150 and S9050. AMD plans to release OpenCL™ 2.0 drivers for enabled AMD FirePro S9150 server GPUs in Q4 2014; conformance testing is planned at that time. Previous generation AMD FirePro products may not support OpenCL™ 2.0.
via AMD Blogs http://ift.tt/1xfoaJf