I’m pleased to announce the release of Jacket v1.1! This release represents a major milestone in Jacket’s development and a great boost in functionality for Jacket customers. The major feature of this release is the inclusion of new GPU datatypes, most notably double-precision. We are very pleased with the performance we’ve seen for double-precision computations.
At the time of this writing, the NVIDIA Tesla T10 series is the newest GPU on the market and NVIDIA’s first in what will become a great line of double-precision enabled GPUs. Even on this first double-precision generation card, we are seeing ~20x speedups for a lot of our examples and test cases.
Of course, GPUs still give higher speedups when comparing single-precision GPU to single precision CPU than when comparing double-precision GPU to double-precision CPU. Nonetheless, both cases provide great overall speedups and in the double-precision case, you get better accuracy for some simulations than you would in the single-precision case.
Jacket 1.1 also includes a big advancement in the ability for Jacket customers to integrate custom CUDA code. The new Developer SDK Upgrade is now available for this. This feature is so cool, it deserves its own post (more to come on that later).
In terms of Jacket’s architecture, this 1.1 release is a major milestone. Since releasing Jacket 1.0, the AccelerEyes team has more than doubled in size. The new team members have had a great impact on Jacket and the product continues to mature, stabilize, and expand. We’re especially thankful for our customers and partners who provide valuable feedback that helps us to best allocate our resources to build a great product.
To give everyone access to this new release, we are resetting all trials so that you can retrial Jacket even if you’ve trialed in the past. To access it, just login, download, and have fun!