Graph Analytics is going to dominate BI/ML/AI processing in coming years.
Zettabolt has recognized the need for accelerating Graph workloads and created products which can provide seamless integration of Graph Analytics tools with hardware accelerators.
WHAT IS ZETTAACCEL?
Zettabolt has leveraged its experience in CPU, GPU and FPGA based optimizations to create a middleware which can plug-in to any Graph processing tool (like Tiger Graph) and offload runtime heavy computations to appropriate accelerator.
User does not need to make any changes in the setup!
Graph
Call to host
zettaAccel
Transfer graph data to kernel
Call kernel computation
Pass result back to tool
Call to kernel
CPU
GPU
FPGA
WHY US
Zettabolt’s expertise in graph analytics and acceleration technology has helped create zettaAccel product for seamless integration between Graph Tool and accelerators. Technology works out-of-the-box for commonly used graph algorithms like PageRank, Louvain, Jaccard, Cosine Similarity etc. It is very easy to adapt zettaAccel to any variation of these algorithms. Zettabolt can make any bespoke changes required for specific requirements and provide continued maintenance and support.
ZETTAACCEL FOR CPU
ACCELERATED GRAPH ANALYTICS
We have created a library of SQL and Graph operators which uses vector-processing based optimizations to provide much superior runtime for popular Graph algorithms than what users get with Graph DB tools. Our implementations of PageRank, Louvain, Jaccard and Cosine Similarity are more than 10x faster than the implementations provided by modern Graph DB tools. Through zettaAccel our algorithms can plug-in to the Graph DB tools as UDFs (User Defined Functions). For a major Credit Card company (case study link), their fraud detection algorithm saw a speed-up of over 100x for both training and inferencing scenarios.
ZETTAACCEL FOR GPU
ACCELERATED GRAPH ANALYTICS
GPUs can provide superlative speed-ups compared to even best CPU optimized graph algorithms. Integration of GPUs with Graph DBs remains a major hurdle towards their wider adoptions. Zettabolt’s zettaAccel product makes it super easy to offload graph algorithms to multi-node-multi-GPU compute clusters. For a variety of commonly used Graph Algorithms (PageRank, Jaccard, Louvain etc.) and their customized implementations, zettaAccel framework has helped provide 100-200x performance improvements (link to blog).
ZETTAACCEL FOR
GNN ACCELERATION
zettaAccel simplifies porting of GNN algorithms to GPU based acceleration. User can configure modelling to run on single or multiple GPUs across servers. Task of configuring the GPUs, offloading computation and bringing the finished models back is accomplished by zettaAccel. Up to 60x performance improvements can be achieved with minimal or no changes to the user setup.
OUR BLOGS & WHITEPAPERS
