Nvidia ditches first SOCAMM memory and resets the race as Samsung, SK Hynix, and Micron start afresh
- Nvidia deleted Socamm 1 after repeated failures to meet expectations
- Socamm 2 promises faster transfer speeds reaching 9,600 mt / s performance
- Adoption discussions LPDDR6 Signal scalability beyond the current SOCAMM 2 modules
NVIDIA has abandoned its previous efforts to market Socamm 1 after repeated technical problems and is now completely focused on Socamm 2, reports.
The first version was positioned as a high power and high capacity memory alternative for IA servers, but delays and design reverses prevented it from gaining ground.
Now an initiate of the industry said And news (Originally in Korean), “Nvidia originally planned to introduce Socamm 1 in the year, but the technical problems interrupted the project twice, preventing real -scale orders”.
A change in performance and objective design
This reset means that Samsung Electronics, SK Hynix and Micron all start on the same basis with the second generation design.
SOCAMM 2 Maintains the detachable module form factor with 694 E / S ports, but increases the transfer speed to 9,600 mt / s compared to the previous 8,533 mt / s.
In practice, this results in a bandwidth of the system from around 14.3 TB / S to around 16 TB / S in the ultra GB300 NVL72 blackwell, a platform already linked to the best GPU discussions in contexts of data centers.
NVIDIA’s dependence at LPDDR5X continues for the moment, although discussions on the adoption of LPDDR6 suggest that the format is designed with long -term scalability to the mind.
Despite these upgrades, the module still consumes less power than Rdimm based on standard dram, an assertion that will need validation under Real Server workloads.
The first generation of Socamm modules was only manufactured by Micron, creating a single outbuilding point that raised questions about the stability of the offer.
Socamm 2 extends the supplier’s base, with Samsung and SK Hynix preparing samples alongside Micron.
This broader participation could make production more stable and the prices more competitive, although it remains uncertain how much mass production will start.
Samsung and SK Hynix indicated that they “prepared for mass production of Socamm in the third quarter”.
However, industry estimates suggest that Socamm 2 will not be available in volume before the start of next year.
One of the most notable differences between the two generations lies in standardization.
SOCAMM 1 was developed outside of Jedec, which limited its adoption to the NVIDIA platforms.
SOCAMM 2, however, could attract the participation of JEDEC, which allows other companies to more easily adopt similar modules in their systems.
If this happens, Socamm 2 could evolve towards a new format of the industry, offering compact memory options and wide bandwidth beyond the Nvidia ecosystem.
For creation professionals, these developments could ultimately influence what counts as the best GPU for video publishing, especially when memory performance directly affects high -resolution images management.
However, analysts remain cautious, noting that if this path is promising, the moment of its arrival, as is the development of LPDDR6 accelerates, can dilute its long -term impact.
As the calculation performance of AI semiconductors improves, the demand for memory to resolve the bottlenecks of the data increases.
Whether Socamm 2 becomes the solution to this problem, or simply one more option in a crowded field, will depend on the execution, normalization and speed of the deployment of LPDDR6.
You might also love
