[01] // identity node-00.simulai

simulai.tech

real-time simulation compute · gpu kernel orchestration · deterministic state propagation

build:0xA47F.c2 env:prod region:us-west-2c uptime:14d 03:42:11
[02] // process utilization
k0k1k2k3
compute 87.3%
[03] // kernel descriptor

$ cat ./manifest.txt

simulai.tech operates a fleet of distributed simulation kernels engineered for high-throughput physical, agent-based, and probabilistic models. The platform is a working tool, not a marketing surface: every panel on this page reflects the same brutalist data-density philosophy that drives the runtime itself.

The runtime supports deterministic replay, sub-millisecond tick scheduling, and zero-copy memory passing between kernel boundaries. Workloads are dispatched onto heterogeneous compute -- CUDA, ROCm, CPU SIMD lanes -- via a unified IR that compiles ahead-of-time. State is propagated through immutable snapshots; nothing is mutated in place.

Operators get raw access. No abstraction layers, no hand-holding, no "no-code" pretensions. You drop into a terminal, you write the kernel, the system executes it across the cluster. Logs stream to stdout, metrics emit on a dedicated UDP channel, snapshots persist to columnar storage.

> 4 active modules · 2,184 ops/cycle · deterministic_v3

  • module/scheduler: adaptive_dag rev.41
  • module/transport: zerocopy.shm.v2
  • module/persist: columnar.snap.gz
  • module/observe: udp.metrics.bin
[04] // memory.map heap 64/64
idle active peak
[05] // telemetry live
  • latency 0.30 ms
  • throughput 847 ops/s
  • vram 21.4 gb
  • tensor.core 94.1 %
  • err.rate 0.002 %
  • queue.depth 128 jobs
[06] // stream stdout · tail -f
[03:14:07] kernel.exec: batch_size=64 latency=0.3ms status=OK • [03:14:08] sched.dispatch: dag=adaptive workers=8 backpressure=0 • [03:14:09] mem.alloc: blk=0x7fa44 sz=2048kb pool=ring-2 • [03:14:10] gpu.0 util=87% temp=64C power=312W clk=1.92ghz • [03:14:11] snap.persist: cols=42 rows=1.8M zstd=4.1x ok • [03:14:12] kernel.exec: batch_size=64 latency=0.28ms status=OK • [03:14:13] xport.shm: peer=node-04 throughput=2.4gb/s • [03:14:14] observe.emit: udp.metrics packets=2048 drops=0 • [03:14:15] tensor.core util=94.1% saturation=high • [03:14:16] kernel.exec: batch_size=128 latency=0.41ms status=OK • [03:14:17] sched.preempt: kernel=k2 reason=priority_boost • [03:14:18] persist.flush: bytes=128mb fsync=ok dur=12ms •
[03:14:07] kernel.exec: batch_size=64 latency=0.3ms status=OK • [03:14:08] sched.dispatch: dag=adaptive workers=8 backpressure=0 • [03:14:09] mem.alloc: blk=0x7fa44 sz=2048kb pool=ring-2 • [03:14:10] gpu.0 util=87% temp=64C power=312W clk=1.92ghz • [03:14:11] snap.persist: cols=42 rows=1.8M zstd=4.1x ok • [03:14:12] kernel.exec: batch_size=64 latency=0.28ms status=OK • [03:14:13] xport.shm: peer=node-04 throughput=2.4gb/s • [03:14:14] observe.emit: udp.metrics packets=2048 drops=0 • [03:14:15] tensor.core util=94.1% saturation=high • [03:14:16] kernel.exec: batch_size=128 latency=0.41ms status=OK • [03:14:17] sched.preempt: kernel=k2 reason=priority_boost • [03:14:18] persist.flush: bytes=128mb fsync=ok dur=12ms •