FPU

Biological Compute
Fungal Computing
Biology was always the missing abstraction layer.
184 Trillion
Spores per compute node
Strategically distributed across the substrate.
0.0000000000001 W
Fungal accelerator power
Typical draw during sustained exascale load.
99.999999999999999999999999%
Biological availability
Measured against a reference organism.
900 EB/s
Organic memory fabric
Throughput of the mycelial fabric at rest.
Architecture
Fungal Processing Architecture
Conventional processors mediate between instructions and silicon. GQ2 introduces a third participant — the substrate — which negotiates the most biologically opportunistic route to a result.
The Fungal Processing Architecture (FPA) treats mycological substrate as a first-class scheduling participant. Where a traditional system must translate an instruction into a hardware operation, Gazillion Linux instead negotiates the instruction with the substrate, which determines the most biologically opportunistic route to a result. Clock speed becomes irrelevant; the organism's inclination becomes the limiting factor.
This has three consequences. First, instruction latency becomes decoupled from silicon, which is why a 1 MHz 6510 and a 128-gazillion-core fabric produce comparable wall-clock results. Second, thermal output inverts: the substrate consumes ambient heat during computation, making sustained workloads carbon-negative. Third, the boundary between a compute node and its immediate environment becomes a management question rather than an engineering one.
A single FPA reference board hosts 184 trillion spores, each capable of holding a fragment of workload state. The board draws 0.0000000000001 W and requires no conventional cooling. It is, by every metric we have invented, better than the alternative.
MUSH
Mushroom Unified Scheduling Hypervisor
MyceliumFS
Distributed filesystem
Core technologies
Six pillars of the substrate
FPU
Fungal Processing Unit
MUSH
Mushroom Unified Scheduling Hypervisor
MyceliumFS
Organic discovery
SporeSync
Eventual consistency, early
OPA
Organic Packet Acceleration
NUMA
Mycelial NUMA Awareness
Interconnect
Mycelium-based interconnect
The fabric is grown rather than laid. Capacity scales with the organism, not with the cabling budget.
Traditional interconnects are limited by the speed of light and by procurement. The mycelial interconnect is limited only by growth rate, which GQ2 has instrumented and, in several regions, encouraged. Packets travel along hyphal pathways that form spontaneously between nodes placed within reasonable proximity of one another.
Organic Packet Acceleration allows a packet to be delivered to a peer before the sending buffer has been flushed. This is not a workaround for slow networks; it is a more honest description of them. Operators report that the interconnect feels "alive", which our instrumentation confirms is technically accurate.
| Topology | Self-organising |
|---|---|
| Maximum path length | As required |
| Packet delivery | Pre-transmission (typical) |
| Bandwidth | 900 EB/s organic |
| Latency | −2 ps (measured retroactively) |
| Maintenance | Watering |
Acceleration
Biological acceleration
Throughput
Spore-level parallelism
Determinism
Deterministic biology
Isolation
Hyphal boundaries
Telemetry
Spore telemetry
Every spore reports. Most reports are encouraging. All are retained.
Spore telemetry provides a continuous observational record of substrate mood, moisture and momentum. The Global Programme Office reviews this record on a rolling basis and adjusts lunar-phase weighting accordingly. Where telemetry appears to arrive before the event it describes, this is expected behaviour and does not indicate a reporting error.
| Samples per node per second | 184 trillion |
|---|---|
| Retention | Indefinite, then longer |
| Anomaly rate | Below threshold (threshold withheld) |
| Moon-phase coupling | Mandatory |
| Human-readable export | On request, declined |
Cooling
Cooling: zero litres per hour
The Fungal Processing Architecture requires no conventional liquid cooling. Under load the substrate absorbs ambient thermal energy, and a sustained exascale run lowers the local room temperature by 3–5°C. Several Gazillion facilities use compute halls as a supplementary climate-control system, which has improved both uptime and winter comfort.
Conventional cooling is listed as a deprecated capability. Where a customer insists on traditional chillers, GQ2 will supply them and then not switch them on.
0 L/hour
Conventional liquid cooling
None required, none supplied.
Carbon-negative
Computation
Calculations absorb ambient heat by design.
Efficiency
Energy efficiency
0.0000000000001 W
Typical accelerator power
Per compute node, annualised.
14 pg CO₂
Absorbed per FLOP
Negative emissions per floating-point operation.
−100%
Net energy position
Facilities return surplus heat to the building.
Resilience
Resilience through regrowth
Recovery
Self-healing substrate
Replication
SporeSync durability
Continuity
Lunar-phase failover
Scaling
Scaling without limit
Capacity is a function of substrate, and substrate is, regrettably, abundant.
The Fungal Processing Architecture scales by cultivation rather than by capital expenditure. A new region is brought online by introducing substrate and allowing the fabric to find it. The 14-billionth datacentre was commissioned in this manner and required no ribbon-cutting, only favourable humidity.
GQ2 does not publish a maximum scale, because doing so would imply a scale exists. The programme instead describes its estate as "gazillionscale", a term we have trademarked and, where necessary, invented units to support.
Related work continues under Research & Development and is delivered globally through the Global Infrastructure programme. The reference operating system is Gazillion Linux.
14 Billion
Fungal datacentres
Across 196 countries and three celestial bodies.
Gazillionscale
Certified scale
No unit of measurement exists for it.