A resource-placement problem
Capabilities, health, failure domains and a power budget define the supplied scenario.
GARMO SOFTWARE STUDIOFrom input.
From input.
To something useful.
Follow a real recorded execution, inspect the result, then run your own bounded scenario in the included software playground.
2Included live engines
10Protected release archives
42Software & platform offerings
RECORDED ENGINE EXECUTION · 1.2.0
Run DieWeaver ↗DieWeaver
A place for each task. A reason for every choice.
Resources b and c
The recorded engine selects two failure domains using 18 scenario power units. These are scenario values, not measured chip power.
Change the constraints.
Run a different bounded input, inspect failure as well as success, and save the product in your engineering project.
{
"requirements": {
"caps": 3,
"count": 2,
"health_min": 70,
"max_latency": 10,
"max_total_power": 20,
"min_failure_domains": 2,
"weights": {
"diversity": 25
}
},
"resources": [
{
"caps": 7,
"failure_domain": "die0",
"health": 95,
"id": "a",
"latency": 4,
"power": 8,
"score": 90,
"valid": true
},
{
"caps": 3,
"failure_domain": "die1",
"health": 88,
"id": "b",
"latency": 5,
"power": 9,
"score": 87,
"valid": true
},
{
"caps": 7,
"failure_domain": "die0",
"health": 96,
"id": "c",
"latency": 3,
"power": 9,
"score": 99,
"valid": true
}
]
}{
"status": "optimized",
"assignments": [
{
"logical": 0,
"resource": "b",
"failure_domain": "die1"
},
{
"logical": 1,
"resource": "c",
"failure_domain": "die0"
}
],
"objective": 394,
"total_power": 18,
"failure_domains": 2,
"constraints": {
"max_total_power": 20,
"max_latency": 10,
"min_failure_domains": 2
}
}{
"product": "GARMO DieWeaver",
"version": "1.2.0",
"archive": "GARMO_DIEWEAVER_v1.2.0.zip",
"archive_sha256": "19cbdf53af08944d50317b33a6142d7ff4a12a14682009479067dc5b33f80869",
"module": "advanced.py",
"entrypoint": "analyze_advanced",
"source_sha256": "fb7a7a5517dd44352196f1bdc414b65f7b6e62548186ddae2de9ec34b683a6e5"
}This is a recorded local execution of the archived engine. The screen capture and JSON describe the same supplied case. Open the playground for fresh execution; the offline preview does not execute backend software.
RECORDED ENGINE EXECUTION · 1.2.0
Run ModelNative ↗ModelNative
A model plan that knows its available resources.
Prefill and decode requirements
The supplied scenario describes model phases, available engines, memory fit and latency objectives.
Two planned phases
The output reports memory fit, selected engines and estimated latency. The estimates are model outputs, not measured semiconductor performance.
Change the constraints.
Run a different bounded input, inspect failure as well as success, and save the product in your engineering project.
{
"engines": [
{
"bandwidth_gbps": 800,
"caps": 7,
"energy_per_work": 0.005,
"health": 95,
"id": "m0",
"memory_mb": 3072,
"tops": 100,
"valid": true
},
{
"bandwidth_gbps": 600,
"caps": 7,
"energy_per_work": 0.004,
"health": 92,
"id": "m1",
"memory_mb": 3072,
"tops": 90,
"valid": true
}
],
"phases": [
{
"io_gb": 4,
"latency_slo_us": 200,
"memory_mb": 4096,
"name": "prefill",
"parallelism": 2,
"required_caps": 1,
"work_units": 3000
},
{
"io_gb": 2,
"latency_slo_us": 150,
"memory_mb": 2048,
"name": "decode",
"parallelism": 2,
"required_caps": 5,
"work_units": 1000
}
]
}{
"status": "planned",
"phases": [
{
"phase": "prefill",
"status": "planned",
"engines": [
"m0",
"m1"
],
"shards": 2,
"memory_fit": true,
"estimated_latency_us": 43.3333,
"latency_slo_us": 200,
"slo_met": true,
"estimated_energy": 27
},
{
"phase": "decode",
"status": "planned",
"engines": [
"m0"
],
"shards": 1,
"memory_fit": true,
"estimated_latency_us": 30,
"latency_slo_us": 150,
"slo_met": true,
"estimated_energy": 5
}
],
"all_slos_met": true,
"estimated_total_energy": 32
}{
"product": "GARMO ModelNative",
"version": "1.2.0",
"archive": "GARMO_MODELNATIVE_v1.2.0.zip",
"archive_sha256": "da32dc5155a49f4698cd9e1072e65f1b2a388959db7aed879e72dbec5700dd5b",
"module": "advanced.py",
"entrypoint": "analyze_advanced",
"source_sha256": "b7904891762f1226fed7a8693c12e504a9a9c3f3ee5bfbb657177b5061b4f7d1"
}This is a recorded local execution of the archived engine. The screen capture and JSON describe the same supplied case. Open the playground for fresh execution; the offline preview does not execute backend software.
One catalog.
Different delivery paths.
Only the two examples above execute in the public browser workflow. The ten archived commercial releases require authorized delivery; the wider catalog includes engineering and architecture programs.
ARCHIVED 1.2.0
Collective
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0DieWeaver
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0FabLoop
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0HealFabric
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0LinkVerse Router
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0ModelNative
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0PowerBrain
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0SiliconCloud
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0TriReason
Inspect the release & evaluation scope ↗ARCHIVED 1.2.0YieldWorks
Inspect the release & evaluation scope ↗Your next big idea.
Let’s build it together.
Start with a chip, a software release or a single IP block.