sub_a3f1→us-e.07·ok·47ms//POE · SHA-256 COMMITMENTsub_a3f2→eu-w.03·ok·51ms//DISPATCH · CAPABILITY-MATCHEDsub_a3f3→ap-n.04·queued//MESH · GLOBALsub_a3f4→us-w.11·ok·92ms//COORDINATION · ETHEREUM L1sub_a3f5→eu-w.08·active//PARALLEL-NATIVE · SUB-TASK ROUTINGsub_a3f6→lat.02·queued//$PRLX · 100M FIXED · 0% INFLATIONsub_a3f7→ap-n.10·ok·38ms//VALIDATION · SAMPLED RE-DISPATCHsub_a3f8→us-e.05·queued//OPEN MARKETPLACE · PHASE 2sub_a3f9→mea.01·ok·124ms//ZK PROOFS · RESEARCHsub_a3f1→us-e.07·ok·47ms//POE · SHA-256 COMMITMENTsub_a3f2→eu-w.03·ok·51ms//DISPATCH · CAPABILITY-MATCHEDsub_a3f3→ap-n.04·queued//MESH · GLOBALsub_a3f4→us-w.11·ok·92ms//COORDINATION · ETHEREUM L1sub_a3f5→eu-w.08·active//PARALLEL-NATIVE · SUB-TASK ROUTINGsub_a3f6→lat.02·queued//$PRLX · 100M FIXED · 0% INFLATIONsub_a3f7→ap-n.10·ok·38ms//VALIDATION · SAMPLED RE-DISPATCHsub_a3f8→us-e.05·queued//OPEN MARKETPLACE · PHASE 2sub_a3f9→mea.01·ok·124ms//ZK PROOFS · RESEARCH
trace
ParalleliX.
Decentralized Parallel Computing Network

Run EverythingIn Parallel

One task in. Sub-tasks out. Dispatched in parallel across the node pool. Throughput scales with the network.

Run TaskRead the docs
parallelix.console9c27a48f
task.run
dispatch.live
settle.txn
MESHokCHAINokPOEok
$parallelix submit ./model.task --gpu 4 --regions 3
✓task IDprlx_7f3a91c8e2…
✓decomposed142 sub-tasks
›dispatched to 38 nodes · US · EU · ASIA
›executing in parallel
✓proof-of-executionsha256 commitment
✓task complete12.4s · 0.18 PRLX
MESHGLOBALSHARDS142POESHA-256L1ETHEREUMFRAME 04v0.1.0// SAMPLE
  1. 01
  2. 02
  3. 03
  4. 04
  5. 05
Plug In

Plug Into The Parallel Stack.

Three roles. One network. Submitters buy compute. Operators sell it. Holders carry the work currency.

  • // submitters

    Run Parallel Workloads.

    Open the Console. Upload a workload. Choose how it splits. Sub-tasks land across the pool in parallel. Pay per validated result in $PRLX.

    Open Console
  • // operators

    Sell Your GPU Cycles.

    Install the Node CLI on your own GPU. Advertise what your hardware can run. Earn $PRLX for every sub-task that passes Proof-of-Execution.

    Operator brief
  • // holders

    Hold The Utility Token.

    $PRLX is the work currency of the network. ERC-20 on Ethereum. 100M fixed. No inflation. Holding does not stake and does not yield.

    Token details
How It Works

One Workload. Many Machines. One Result.

Submit once. Decompose. Dispatch. Aggregate. The pool runs the work in parallel and returns one result.

// 01

Decompose

decompose(task) → [ST-001..ST-008]
PARSEPARTITIONTAG

The orchestrator splits the workload into independent sub-tasks. Dependency graph in, sub-task list out.

// 02

Dispatch

dispatch(8) → workers[5]
MATCHENCRYPTROUTE

Sub-tasks route to nodes that match the work. Hardware. Region. Live load. All of them run at the same time.

// 03

Aggregate

aggregate(8) → 0xa3f8…
VERIFYMERGESETTLE

Results return signed with a Proof-of-Execution hash. Validated. Merged. One output back to the submitter.

Step 1 of 3: Decompose
Inside A Node Session

A Node In The Mesh. Skin In The Game.

Operators run real hardware. Declare what it can do. Stake $PRLX as collateral. Get paid per verified sub-task.

Network Foundations

What The Network Guarantees.

Five primitives. The whole network rests on them. Parallel execution. SHA-256 commitments. Four-signal scheduling. Encrypted dispatch. Capability-aware routing.

// 01Execution

Parallel-native execution.

Tasks split into independent sub-tasks at the coordination layer. Each one ships in parallel. The unit of work is the sub-task, not the box.

task
A3F8 · matmul-4096
submitted
14:22:07 UTC
shards
6 / 6
idnodetimeline · wall-clockelapsed
S1node-04
0.82s
S2node-09
0.91s
S3node-15
78%
S4node-22
62%
S5node-31
55%
S6node-07
41%
0s250ms500ms750ms1.0s
wall-clock
0.91s
seq est
5.4s
speedup
5.9×
6 nodes · concurrentwall 0.91s
// 02Verification

Proof-of-Execution.

Each result returns with a SHA-256 hash commitment over (task_id, result, node_id). The network reconstructs the hash to validate.

inputs3 fields · hashed in canonical order
task_id
0x9c4d…f2a1
result
0xb3e7…a8f0
node_id
nx-0049
↓ sha-256 · 256-bit · deterministic
commitment verified
0x
4a91c91d
b8e6 · 5a12 · f3d7 · 8ea2
a3c1 · 047b · 9d2e · 6f70
algorithm
SHA-256
output
256 bit
ledger
anchored
network ledger✓ committed
// 03Routing

Decentralized scheduling.

Sub-task to node matching uses four signals in priority order: capability match, capacity-aware queuing, best-fit allocation, reputation weighting.

candidate pool
1,420nodes
evaluated in
14ms
01capability1,420 → 312−1,108
02capacity312 → 88−224
03best-fit88 → 17−71
04reputation17 → 1✓ 1
routed
node-04
RTX 4090 · US-EAST
reputation
97.4
of 100
4 signals✓ dispatched
// 04Privacy

End-to-end encryption.

Sub-task payloads are encrypted to the assigned node's public key before dispatch. Only the assigned node can decrypt. Nodes sign every message.

handshake4 frames · 1.4ms
T+0.0→KEY_REQUEST384 B
T+0.3←PUB_KEY_EXCHANGE288 B
T+0.7→SESSION_AGREE128 B
T+1.4⇄CHANNEL_OPENAEAD
●x25519 outer
●ChaCha20 mid
●SEALED core
derived keysHKDF · 3 outputs
session_keyactive
ed25:7f3a · 91c8 · 4b22
mac_keyvalid
bc8d · 12e4 · 99af · 62
nonce_counterfresh
0000…c8d1f2a4
ChaCha20 active✓ sealed · authenticated
// 05Capability

Industry-aware node profiles.

Nodes declare workload classes: AI, scientific compute, big-data analytics, blockchain, rendering. The scheduler routes sub-tasks to compatible hardware.

class filter
AI · GPUSCIENTIFICRENDERINGBLOCKCHAINBIG DATA
workload
5 signals
CUDA · FP32 · TENS · VRAM · LAT
nodehardware · classcapsregscore
node-04
RTX 4090 · 24 GB
AI · GPU
US-E5/5
node-09
HPC · 256 GB
SCIENTIFIC
EU-W2/5
node-15
RTX 4080 · 16 GB
RENDERING
AP-N3/5
node-22
EPYC · 64-core
BLOCKCHAIN
US-E1/5
filter: AI · GPU→ node-04
Workload Classes

Where Parallelism Wins

Workloads where adding nodes actually makes things faster. Most code stalls on serial sections. These don't.

ai // training runepoch 14/40
0.184train loss
model · llama-7b · ckpt 14000
running · 2h 14m elapsed · eta 1h 32m
GPU
32
BATCH
256
LR
3e-4
MOM
0.9
epochtimelosslr
ep 1418:420.1843e-4
ep 1318:240.1923e-4
ep 1218:060.2014e-4
ep 1117:480.2184e-4
// ai

Artificial Intelligence

Training fans out across many GPUs in one shot. Inference fans out per request, not per region.

TrainingInferencePreprocessing
sim // parameter sweep◇ genomics
seq.aligned2.4M reads
step
14,820 / 19k
mesh
512 × 512
error · 1.2e-6 · tolerance · 1e-5converged
workers
64 nodes
recent timesteps · sim 14820 of 19000
w1
88%
w2
94%
w3
71%
w4
82%
// science

Scientific Research

Parameter sweeps and ensemble runs are embarrassingly parallel. The bottleneck is seat count, not the algorithm.

ClimateGenomicsPhysics
STREAM · ETL◇ 12k tps · live
SOURCE
8 streams
142 GB/s
TRANSFORM
4 ops
filter
map
agg
join
89 GB/s
SINK
4 partitions
p0
p1
p2
p3
134 GB/s
THROUGHPUT · 60s windowlag · 12ms · ok
queue depth · 34%errors · 0partitions · 4 / 4 healthy
p0
4ms
offset · 14.2M
p1
7ms
offset · 14.1M
p2
12ms
offset · 14.3M
p3
5ms
offset · 14.0M
// data

Big Data Analytics

Map-reduce shapes. Partition the input, run one operator across the pool, fold the results back.

RiskDiagnosticsLogistics
zk // merkle commitroot anchored
root0x9b5e…f211
proof · 248B
batch 14,221 · 256 proofs/batchprover · 1.2s avg
groth16
plonk
stark
pf_a3f1groth161.2s✓
pf_a3f2plonk1.4s✓
pf_a3f3stark0.9s✓
pf_a3f4groth161.5s...
// blockchain

Blockchain Applications

Proof generation is heavy and offloadable. The chain only needs the verification; the witness can be computed anywhere.

zk-SNARKsOraclesContracts
render // queue◇ 1680 frames
72%scene 04 · 4k
3840 × 2160 · 48 fps · h.265 · 85 Mbps
frame 1209eta 04:18
scene 01
420f
scene 02
380f
scene 03
480f
scene 04
400f
batch 14224f4.2s92%
batch 14124f4.1s✓
batch 14024f4.3s✓
batch 13924f4.4s✓
// media

Media & Rendering

Frames are independent. So is encoding. A render queue is a sub-task pool with a different name.

RenderingEncodingAR/VR
Parallel Token

Pay. Run. Earn.

The work currency of the network. Submitters pay, operators stake, sub-tasks settle in $PRLX. 100M fixed. No inflation. No holder yield.

// SUPPLY

100,000,000

$PRLX · ERC-20 · Ethereum

NameParalleliX Network
Decimals18
StandardERC-20
/View full tokenomics

// ON-CHAIN

CONTRACT

Live · Etherscan verified
0x93ff39f65cc1d21067939961993adf3f36bbf893
ETHERSCAN DEXSCREENER BUY

ERC-20·Ethereum Mainnet·No mint fn·5% buy/sell tax

// pay

Task Payments

Every sub-task is paid in $PRLX. Demand scales with network throughput.

settle in
$PRLX
unit
per sub-task
driver
volume, not speculation
// stake

$PRLX Staking

Operator collateral. Locked while a node runs. Slashable on misbehavior.

stake
50,000 $PRLX / node
cooldown
7 days
scope
operators only
// earn

Operator Rewards

Paid per validated sub-task. Weighted by complexity and uptime.

phase 1
25% bootstrap pool
phase 2
task-fee share
weight
complexity × uptime

// DISTRIBUTION

LiquidityUNCX · 3 + 6 mo60%·60,000,000

Deep liquidity for trading and price stability

Staking RewardsSablier · 24 mo linear25%·25,000,000

Bootstrap subsidy paid to staked operators for validated work

TeamSablier · cliff 12 + vest 2410%·10,000,000

12-month cliff + 24 months linear vest

TreasurySablier · streamed5%·5,000,000

Operational reserves and partnerships

Locker proofUNCX · view lockers

// TAX DISTRIBUTION

5%

buy/sell tax

split 3 ways

tax3-way
DEV2%

platform upgrades

OPS2%

infrastructure

MKT1%

awareness, partners

where it flows9 line items · on-chain transparent
  • DEVprotocol security · scheduler upgrades · SDKs
  • OPScoordinator nodes · monitoring · operator support
  • MKTbuilder outreach · partnerships · awareness

Tax does not fund operators · bootstrap covers that.

The Product Surface

Deploy. Read. Operate.

Three places to interact with the network: Console for submitters, Docs for builders, Node Client for compute providers.

phase 01

Built For The First Sub-Task.

Decompose. Dispatch. Aggregate.

Read tokenomics for the $PRLX spec. Read the docs for the protocol, the scheduler, the operator path.

Read tokenomics$PRLX referenceRead the docsdocs.parallelix.io