Going to conferences and meeting new people, folks you are hoping to collaborate with, and generally getting a vibe from the people around you is always fun. Memcon 2024 was no exception. We had a lot of great and productive conversations and the best part of this conference is it is people with significant influence who are in attendence and accessible for freeform conversation. This is the first blog in our whole Memcon series to give you the jist of what the happenings are in the memory industry as of the spring of 2024.
I’m going to start with the biggest thing that I think needs to change to make future Memcon’s better:
More end-user attendees and stories.
I get that technically speaking, hyperscalers like Microsoft (who was the featured keynote speaker) are “end-users.” However in reality, Microsoft and their contemporaries are more akin to intermediate users. Are they in charge of driving trends in the emerging memory space? I mean, not really. They deploy memory at a scale that few others on the planet do, but when we are specifically talking about the Azure side of Microsoft, they are building things to sell to end users, it’s a somewhat singular use case that cannot be generally applied to really anyone else including other hyperscalers. I find it odd to consider them the ones to set the trend, when it’s clear they are looking to the market to set a trend and optimize it to serve the eventual end users. \rant
Now that we’ve got that out of the way, lets talk about the high level themes and break those down further than quips and nuggets. In the following week and change expect a short blog on each of the topics below (which will expand in the series dropdown above to make navigation easy). Enjoy!
- HBM is poppin’, but it’s not going to dominate share of the memory market and will likely stabilize in the next 18 months from a raw outpacing of demand.
- NAND is still technically memory, but no one is paying that much attention to it, and they are really hoping for an optane replacement sooner rather than later.
- There is a lot going on in the low power, industrial, edge and consumer segments that was just glazed over with regard to memory (LPDDR5X, CAMM, etc) because it’s just not really all that sexy.
- CXL is very much still in it’s infancy. It needs some serious care and feeding to get out of the “baby” phase of existence and move onto self-reliance from the customer base.
- Memory Pooling is cool, but it’s not the universal value unlocker that many previously thought it would be.
- Memory Tiering is hard. Like, really hard. It’s one of the biggest gaps in the market in terms of solutions, but Memverge and Kove are both taking different approaches to solving it.
- Cache Coherent Memory Sharing AKA CXL 3.1 is what everyone truly considers “CXL” except for the hardware companies that have to iterate through 1.1 and 2.0 to get there.
- GenAI dominated the conversation and said absolutely nothing, much like your favorite LLM (but we do love the memes).
- In or near memory processing is an obvious problem that no one seems to want to tackle just yet.
- The new paradigm in hardware for chiplets is just getting started, and there’s a lot to be excited about from an efficiency, performance and capability perspective in the next few generations.
Shameless plug: If you are also just looking to bring in experts who know CXL, it’s capabilities, and want to marry that with your current infrastructure or application deployment the experts at Jackrabbit Labs can do deep dive sessions and use-case analysis. Reach out to us if you are interested.