Quick Consulting Service
Overview
Recently, many kinds of multi/many-core processors, such as x86, GPU, DSP, Cell/B.E., and ARM, have become available not only in the HPC industry, but also in the embedded industry. Not all processors are the same however, for example a generic GPU is quite powerful for most image processing algorithms, but has trouble with others, for example histogram calculation. This can make it difficult to estimate what sort of a speed-up you could expect with a multi/many-core processor. Fixstars has focused on parallel computing since 2005, and has extensive knowledge of both hardware and software aspects.
Fixstars' OpenCL Quick Consulting Service can quickly evaluate how effective porting your software to many/multi-core systems with opencl would be. Simply send us your source with a brief description and Fixstars' OpenCL experts will analyze your code, port it to OpenCL, and provide a report detailing the sort of performance improvements you can expect. In addition to OpenCL porting, CUDA™ and OpenMP™ porting are also available. Using this service, you can dramatically reduce the workload of evaluation.
Terms & Conditions
See below for details:
| You need to provide | - C99 source code - Makefile (Linux) or Project files (Windows) - A brief description of the source code |
|---|---|
| Fixstars will deliver |
- Optimized Source code (Binary, Source) - One-page performance report (PDF) |
| Size of source codes | below 100 LOC (commment line is not included) |
| Turnaround time | Typical 10 business days (MAX 15 business days) |
| Support OS |
- Windows 7 64bit - Yellow Dog Linux for CUDA 6.3 - Ubuntu 10.04.1 LTS 64bit |
| Target HW | Select one of the following options: - Intel i7-2600, i7-990X, Xeon 5650 - NVIDIA Tesla C2050/2070 - IBM Cell/B.E., PowerXCell |
| Target Programming model | Select one of the following options: - OpenCL(Intel, NVIDIA, IBM) - CUDA (NVIDIA) - OpenMP(Intel, IBM) |
| Price | Regular Price: USD 499.00 Trial Price (only for the initial purchase): USD 299.00 |
The one-page report includes:
- The result of profiling your source code
- Recomended platform (HW and SW)
- Expected speed-up factor using one specific platform
- Comments of optimized source codes
If your code size exceeds 100LOC and/or you are interested in other programming models, Please E-mail your requests to us and we will get back to you quickly. Other models, such as Intel TBB/ArBB, Google GO, MapReduce, and MPI are available.
How to use
Follow the following procedure:
Step 1.
Confirm that your source code can be built by gcc or MS VC or Intel C, and works normaly on Windows 7 or Ubuntu 10.04.1 or YDL 6.3.
Step 2.
Write a brief description of your source code, including build method, installation, normal inputs/outputs, and typical elapse time in your environment.
Step 3.
Choose which HW/SW you want to optimize your code to. We analyze your code from many different points of view, but port your code to only one specific platform. For example, OpenCL + NVIDIA GPU, OpenMP + Intel Xeon, and so on.
Step 4.
Submit your code and other files HERE.
Please DO NOT submit anything confidential. Fixstars reserves the right to use and/or disclose all submitted code without notice. Fixstars also has no responsibility to return the submitted data. For details please see the submission page.
Step 5.
Fixstars will confirm that your submission meets our specifications, and we will send you a URL to the payment page. All major credit cards are accepted.
Step 6.
Please wait while we port your code to OpenCL! This usually takes about 10 business days.
