We have a thesis-defined approach to capability design and prefer to play slightly at variance with much of the noise of the current compute milieu.
SLM is focused on securing leap advances at multi-term intervals, across the near- and longer-term. Our vagueness notwithstanding, for us, this generally means we hold a far greater bar for what we think complex, efficient computation can and should do, for higher, sustained leverage and imperatives for the world.
The architectural and efficiency advantages we hold over the kind of high-throughput compute most people expect allows us to capture these gains even more durably. But the advantages we’re seeing in the multiplicative classes of scaling matrices we’ll be introducing together places us on a stronger path for the tenor of practical leaps we target, instead of being constrained to aim only for incremental speed-ups every two years.
At our current timeline, we may sustain a tapeout cadence of 5-8 months per generation per accelerator lineage. We’re focused on doing this while still continuing advanced programs of nontrivial architectural and research and engineering import.
:: How our team profile is shaped
02 Our character and animating principles
03 Our general modes
xy [at] sl-machines.com