Silicon Proven. First-of-Its-Kind Computational Memory Chip.
The growing gap between data volume and our ability to process it. Or more precisely, the flow of data between storage, memory and processing.
Processing inside memory. Leap over memory bottlenecks and eliminate unneeded data movement.
Just process. Faster.
Thousands of processors at unmatched bandwidth
Processors and memory on a monolithic die
Cut down on data movement and wasted power
Tailor-Made for Low-Latency & High Bandwidth Applications.
Data Analytics teams are tasked with a whole new scale of workloads as data volume continues to grow at an unprecedented pace. An analysis of existing system architectures shows that data movement is the root cause of processor bottlenecks and slow response time.
What is needed is an new, innovative computational architecture that eliminates data movement requirements, therefore improving analytics performance.
compared to DDR5
Memory Module
Random access
at Sequential
access speed
What does it mean? There is a growing gap between compute performance and memory performance. External memory bandwidth does not keep up with data processing needs.
A new solution is needed. One that bridges the memory-to-compute gap, allowing data-intensive workloads to sprint forward.
Computational
Memory
Closes the
compute-to-memory gap
Increasing compute speed does not accelerate the processing of data-intensive applications.
More memory bandwidth does.
The XRAM™ does.
NeuroBlade XRAM™ is a DRAM memory with embedded processing logic. The processing elements are integrated near the memory banks, providing massively parallel execution at high bandwidth.
As a result, data-intensive workloads are now processed seamlessly, with far less data movement and much higher performance. It enables operations such as data comparison and manipulation, aggregations, filtering, and lookups to be dramatically accelerated.
High latency for relevant data
High power consumption
Bus is fully utilized with relevant data
x10 energy efficiency
