Section Topics:
News:
|
8 ways CodeQuest™ can improve your 454 data analysis Comparisons with HMMER and SSEARCH show that DeCypher delivers 300-1000X performance over CPU core Invitrogen and Active Motif explore use of FPGA for Next-Generation Sequencing Data Analysis TimeLogic welcomes new customers at UNC Chapel Hill, Perlegen, and MRC |
Newsletter:
|
|
|
TimeLogic has completed a set of searches to demonstrate the performance and value of our FPGA solutions in comparison to open source software running on commodity server hardware. Each search was processed on a DeCypher system (with 1, 2 and 3 SeqCruncher FPGA accelerators installed). The DeCypher system includes TimeLogic software, SeqCruncher™ FPGA cards (PCIe), and a Dell 2950 III host server equipped with 2 quad-core Xeon CPUs (3 GHz). Software comparisons were run on the same server using all 8 CPU cores, but without the help of the accelerator hardware.
Results Overview Due to the high degree of parallelization within the SeqCruncher FPGA circuitry, DeCypher is able to outpace the CPU-only option by a wide margin. The graphs below demonstrate an average of the search performance for each algorithm. Several Hidden Markov Model comparisons were used to compare HMMER software to DeCypherHMM. Using a DeCypher system with 3 SeqCruncher cards, up to 1,151 CPU cores of performance can be achieved.
To compare Smith-Waterman performance, two sample searches were processed using DeCypher and SSEARCH software. DeCypher was able to process these searches at a rate of 236-1099X faster than SSEARCH on a single CPU core.
Computational performance is only meaningful when solution cost is considered as well. In the graph below, we compare the cost of a DeCypher system (including Smith-Waterman module, 1, 2, or 3 SeqCruncher cards and a Dell 2950 III server) to the cost of a cluster with equivalent performance (Dell 1950 1U servers with 2 quad-core CPUs). The cluster requires 4-5X more budget to acquire. Cost of operation—not factored into this graph—would show the DeCypher system to be even more cost-effective.
To receive a copy of the full performance benchmarks,
|
