In this guest post from One Stop Systems, Tim Miller, president of SkyScale, covers how GPU cloud computing is on the fast track to crossing the chasm to widespread adoption for HPC applications.
Tim Miller, President, SkyScale
The use of cloud computing has become mainstream across many companies and industries today. The initial hesitations around security, reliability and ease of use have largely been addressed. Clearly cloud usage has moved from early adopters across the chasm to widespread adoption for traditional business applications. Today we see a similar adoption trend for HPC applications. The benefits of cloud for HPC are the same as mainstream users; savings in capital and data center management expense, off-load of maintenance and operation issues, avoidance of technology obsolescence, on-demand usage, flexibility and peak load coverage without risking idle capacity. The recent advent of GPU cloud computing is the latest wave in this megatrend toward the cloud. Interestingly enough the growth in GPU computing over traditional CPU computing for HPC applications generally becomes parallel to the overall trend toward cloud computing making this the perfect time for HPC users to consider GPU computing in the cloud.
Two good examples of very different markets adopting GPU computing and where cloud usage makes sense are artificial intelligence and high quality rendering. In the case of AI and especially machine learning, the powerful benefits of GPU acceleration open new and high impact use cases. For many companies, adding specialized GPU compute resources to their existing IT infrastructure does not make sense when they can access the amount of resources they need when they need it in the cloud. They can avoid the infrastructure costs driven by the high power requirements of GPU systems and can avoid paying for idle capacity as they move through the life cycle of their AI project. In high quality rendering, many studios and production facilities have invested heavily in large scale CPU based rendering farms. With the growing appeal of superfast GPU based rendering they now face the decision of augmenting those farms with GPU resources or simply tapping into resources available in the cloud. Again, the cloud provides the clear advantage of low investment and flexibility. Adopting GPU cloud computing now makes sense in these two application spaces.
Two good examples of very different markets adopting GPU computing and where cloud usage makes sense are artificial intelligence and high quality rendering.
Most of the hyper-scale cloud providers such as AWS, Azure and Google offer GPU Cloud Computing as an option. GPU Cloud Computing is a secondary focus of these providers and in many ways is a bolt on to their main high volume business applications. The unique requirements of many HPC users are not ideally addressed through these hyper-scale cloud vendors. On the other hand, SkyScale offers a complementary service to hyper-scale providers specifically tuned to the needs of the HPC user using GPU computing. Simply put, SkyScale’s difference is the three ‘P’s: Performance, Price and Personalized service.
Performance: HPC applications are by nature performance hungry where true value is derived by both performing quicker analysis and problem solving or by addressing larger computational problems in a fixed amount of time. Often the computation work integrally ties to the creativity of the scientist, engineer or artist where the speed of iteration drives innovative breakthroughs. SkyScale generally provides 30% higher raw performance than other cloud vendors using the same compute technology. This performance difference is provided through the bare metal dedicated server access SkyScale provides versus the shared, virtualized server model of other providers. With a leased server from SkyScale, the user has sole access to the entire system avoiding the performance and security issues associated with the multi-tenant paradigm of other providers. Additionally, SkyScale provides higher end systems than any other cloud provider including systems with up to 16x NVIDIA® V100 PCIe GPUs. All other vendors top out at 8x GPUs. The chart below demonstrates the performance advantage on a widely used machine learning framework.
Price: Value is a critical driver in any industry, but in HPC the expense associated with massive compute resources can quickly overwhelm the user. SkyScale provides the absolute highest performance GPU compute nodes available in the cloud while maintaining a 30-50% price advantage as well. The advantage of price for equivalent performance approaches 70-80%. When factored across projects that can be measured in weeks or months the savings are large. Also, in keeping SkyScale’s focus on simplicity of business, there are no additional fees for data transfers or storage or high speed networking.
Personalized Service: With SkyScale ‘bare metal’ does not mean customers are on their own. SkyScale includes, at no additional cost, direct support from engineers, both pre- and post-sale, including help with configuration and tuning to maximize performance. Often deploying optimized HPC resources involves a collaborative approach not available through hyper-scale providers. SkyScale’s goal is to reduce the complexity of HPC so customers can succeed.
GPU Cloud Computing is on the fast track to crossing the chasm to widespread adoption for HPC applications. Many options are available, but SkyScale is uniquely positioned to deliver in the dimensions specifically important to HPC users: performance, price and personalized service.
Tim Miller is president of SkyScale, provider of cloud-based dedicated multi-GPU servers and clusters.
