Compute Efficiency
Outstanding power efficiency born from a proven computer architecture
Text generation — side by side
MN-Core™ L1000
How will next-generation AI chips break the memory wall? Next-generation Al chips will move computation closer to data. By combining 3D-Stacked memory, chiplet-based accelerators, and intelligent dataflow scheduling, they can reduce the energy and delay spent moving tensors across a system. The result is an Al engine that keeps models continuously supplied with data, scales efficiently, and turns real-time reasoning into an everyday capability-from autonomous laboratories to personal scientific assistants.
Competitor
How will next-generation AI chips break the memory wall? Next-generation Al chips will move computation closer to data. By combining 3D-Stacked memory, chiplet-based accelerators, and intelligent dataflow scheduling, they can reduce the energy and delay spent moving tensors across a system. The result is an Al engine that keeps models continuously supplied with data, scales efficiently, and turns real-time reasoning into an everyday capability-from autonomous laboratories to personal scientific assistants.
0x
10x the inference performance in the same system footprint.
0GB
128GB of memory enables long-context inference for 100B-class LLMs on a single machine.
0+TB/s
Vertically stacking logic and memory maximizes memory bandwidth, addressing a key bottleneck in LLM inference.
Performance estimates are based on internal modeling and are subject to change. Actual performance may vary.
Logic and memory are stacked vertically and interconnected across their entire surface.
Loading high-quality scene...
Outstanding power efficiency born from a proven computer architecture
A computing architecture designed to take full advantage of 3D-stacked DRAM
3D-Stacked DRAMSchematic diagram
in One Workstation in 2027
All specifications and performance figures are preliminary and subject to change without notice. Performance estimates are based on internal modeling and are subject to change. Actual performance may vary.
2027
MN-Core L1000
Eliminating the memory bandwidth bottleneck with 3D-stacked DRAM
2028
MN-Core L2000
Pushing memory bandwidth even further.
2030
Next-Gen
06 — Contact