What you’ll learn:
- How lossless data compression can reduce memory and power requirements.
- How ZeroPoint’s compression technology differs from the competition.
One can never have enough memory, and one way to get more memory is to use compression. However, prior to ZeroPoint's entry at the cache level, compression was something that was relegated to block-level solutions. The technology is optimized for 64-byte cache entries.
I talked with Nilesh Shah, Vice President of Business Development at ZeroPoint, about the company’s technology and family of compression products (see video above).
ZeroPoint’s CacheMX, which works at the cache level, is IP that’s included with a processor’s IP. The lossless compression system also manages the compressed data (Fig. 1).
Speed is the reason CacheMX can work at the cache level (Fig. 2). Here, compression times must be on the order of 3 to 5 ns. That’s 1000X faster than software compression used at the application level.
The same hardware approach can be used with CXL and CXL-attached memory (Fig. 3). CXL-level compression only needs to run at about 100 to 200 ns.
CXL use is growing in the cloud, where massive amounts of memory are needed to address technologies like generative artificial intelligence (AI) as well as general AI training for neural-network models. At this point, CXL is lowering the total cost of ownership (TCO) of large server farms and disaggregated cloud servers, but additional compression helps reduce TCO by requiring less real memory (Fig. 4). Of course, these days, this will probably translate into the ability to provide more memory at a lower cost and lower power requirements.
