01
Performance predictors
Latency and throughput models per workload class and GPU architecture.
Predict · Simyx
Simyx is a learned performance twin of NVIDIA GPUs. It estimates the latency, throughput and accuracy of candidate optimizations before they run, so every other product benchmarks only the most promising few.
Capabilities
01
Latency and throughput models per workload class and GPU architecture.
02
Estimates quality loss of precision changes before they run.
03
Caps GPU spend per run and allocates it to the best-ranked candidates.
04
Chooses some candidates to run for what the model will learn, not only for speed.
05
Estimate a change or a new GPU before you buy or build anything.
06
Fleet-level simulation for GPU planning.
Guided search
Simyx estimates each candidate's speed and accuracy before it runs. Only the top-ranked candidates use GPU time.
How it works
Turn each candidate and workload into trace-level features.
Score speed and accuracy with uncertainty.
Send the top-k and a few exploratory picks to the GPU.
Fold measured results back into the model.
Specification
| Item | Detail | Status |
|---|---|---|
| Used by | Profyle, Kernl, Quantra, Servex, Edgyn | Pilot |
| Predicts | Latency, throughput; accuracy next | Pilot · Next |
| Training data | Turbokern optimization-trajectory dataset | Pilot |
| Compute | RAPIDS on GPU for training and analytics | Pilot |
| Standalone what-if | Estimates without a full run | Planned |
| Capacity planning | Fleet right-sizing simulation | Research |
Place in the loop
Every benchmark any product runs trains Simyx, and every product asks Simyx what to run next.
FAQ
No. Simyx only decides what to measure. Nothing is kept without a real benchmark on the target GPU.
Predictions start less certain. Simyx sends more candidates to the GPU until it has learned the new architecture, which costs more compute early on.
Only anonymous performance features, and only if you allow pooling. Weights, prompts and eval data are never used.
A standalone what-if and capacity-planning tier is planned after the pilot.
Next step
In the pilot every run reports how many candidates Simyx pruned and what that saved.