Split the workload.
Each task gets a deterministic input. Independent jobs can be assigned without waiting for the previous result.
THREAD ENGINE / CONCEPT ARCHITECTUREWORK QUEUES → COMPUTE LANES → RESULTS
$THREADNVDAx ↗
THREAD / THE ARCHITECTURE OF PARALLEL COMPUTEOne task.
A thousand
possibilities.
Many lanes. One workload.
Explore the thinking behind parallel compute.
EXECUTIONReal browser workers
COMPARISONSerial → Parallel
APPROACHMeasured. Reproducible. Local.
INSIDE THE EXECUTION PIPELINE
01 Dispatch →02 Compute →03 Synchronize →04 Verify
01 / THE ARCHITECTURENot every problem
Not every problem
needs to stand in line.
02 / FROM AN IDEA TO EXECUTION
Built around parallel work.
Give it a worker.
A pool of browser workers pulls tasks from a shared queue. The main page stays responsive while they compute.
Bring it together.
Compare every result across both runs. More workers only mean something when the output still agrees.
03 / WORK YOU CAN WATCH
BROWSER CPU / NOT GPUThe parallel playground.
Same tasks. Different execution. Run the workload once with one worker, then again with your selected pool.
TASK QUEUE▧ Pending Running Complete
WORKER POOL
This test uses CPU Web Workers, not CUDA or GPU execution. Timing includes task dispatch and result collection, but excludes worker startup. Browser scheduling, background load, and device limits affect the result.
04 / LET THE RESULT SPEAK
Measured here. On your device.
ONE WORKER—ms
Serial baseline
WORKER POOL—ms
4 parallel workers
OBSERVED SPEEDUP—×
Serial time ÷ parallel time
OUTPUT INTEGRITY—
Run both modes to verify
Serial
No run yet
Parallel
No run yet
A larger worker pool is not guaranteed to be faster. Small jobs may spend more time coordinating than computing.
05 / THE COMPUTE CONNECTIONA shared belief
A shared belief
in parallel potential.
THREAD is an independent exploration of parallel work. NVDAx is the proposed quote asset for its token concept, reflecting the project’s focus on computation.
PROJECTTHREAD
PROPOSED QUOTENVDAx
MARKET STATUSNot deployed
No live token pair or liquidity pool is connected. A proposed quote asset is not backing, equity ownership, or an endorsement by NVIDIA. THREAD is not affiliated with NVIDIA.
What is actually running?
A deterministic integer mixing loop runs for each task. Both modes use the same inputs and iteration count. The serial run has one dedicated Web Worker; the parallel run distributes tasks across a pool. This is a small browser benchmark, not an AI model or GPU workload.
How is timing measured?
Workers are created and report ready before the clock starts. Timing uses performance.now() from the first task dispatch to receipt of the last result. It includes messaging and scheduling. Serial runs first, so thermal state and browser optimization can affect comparisons.
Why can more workers be slower?
Creating useful parallel work involves coordination. Messaging, available CPU resources, and the size of each task determine whether parallel execution helps. Repeat runs and treat this as an experiment rather than a hardware ranking.
Does this run on NVIDIA hardware?
No GPU is requested. Web Workers are scheduled by your browser and operating system on CPU resources. The NVDAx proposal is a separate token concept, not a dependency of the playground.
Where does my data go?
The workload is generated locally, and results stay in memory until you export them. No wallet, account, or external compute service is used. Reloading clears the current run.