Unum is a Deep-Tech research company focused on designing scalable data-processing algorithms to make database, search, and AI efficient for enterprise and edge. Unum is on a mission to efficiently scale data processing, from Storage and Analytics, to Search and AI Modelling.
Key Product line
Unum is on a mission to efficiently scale data processing, from Storage and Analytics, to Search and AI Modelling.Learn more
Unum’s persistent transactional database can store media, tabular, document-like, network-like, or even vector data, combining unseen universality, with revolutionary speed.
The in-memory vector search index is capable of navigating datasets with billions of high-dimensional vectors, empowering real-time semantic search at the scale of largest marketplaces and content-providers.
The quantized Neural Networks, recognise media and texts written across numerous languages, with state of the art accuracy, but at the fraction of the cost, making Artificial Intelligence accessible for the next million applications.
What does this Partnership bring to the table?
Gcore, with its global presence and extensive network infrastructure, provides seamless 5G connectivity to cloud infrastructure and high-end equipment to the edge.
Gcore’s powerful SSR server, built on Quanta hardware—Intel Xeon Platinum 8480+, ensures that this project benefits from the latest developments in network technology, enabling efficient and lightning-fast data transfers. Additionally, Gcore’s advanced DDoS protection adds an extra layer of security to the project.
Unum’s Milti-Modal Transactional Database - UStore, in combination with its in-house transactional engine - UDisk, can help avoid expensive system calls and partially or entirely bypass the OS and Filesystem layers while exchanging data with storage devices.
UStore was designed for the age of AI with batch workloads in mind. Its native support for Vector Search brings an entire family of bundled Transformer models optimized for cheap and efficient inference on Edge, forming the first complete Database+AI product lineup for customers worldwide.
Our First Collaboration
We conducted the first UCSB 10TB Benchmark on GraphCore IPU-POD16, comparing RocksDB and UDisk. The results showed significant improvements in workload parameters like Initialization, Batch Read, Scan, Read Update, Read Upsert, and Batch Upsert.
We are significantly faster on “Batch” operations common to Intelligence-oriented applications, both for reads and writes, making UDisk the ideal engine for use cases with typical random-order reads and writes, such as graphs, AI training, and retrieval-based large-scale AI systems, like newer chat-bots.
Unum and Gcore share a common vision and expertise to make this engine even faster and use it on CID-transactional replacement for a file system. This will open doors to markets that haven’t historically used databases and preferred object stores, such as video streaming and hosting.
In addition we are also targeting to mix “Updates” with “Range Selects” in the future. We will optimize the hardware configuration even further to derive more optimal results and achieve 10M+ ops/second performance.
What does this mean for our customers?
As partners, Unum and Gcore are focused on unifying world’s best hardware vendors to bring the highest-performing managed database service to our worldwide customers.
< 30 ms latency
to access your nearest location.
> 10 Million
random operations across terabytes of data.