01
One-command profiling
Capture timeline and kernel traces on DGX, HGX and Jetson with a single run.
Profile · Profyle
Profyle captures Nsight and CUPTI traces on the target GPU, classifies each bottleneck and attributes its cost to specific kernels and layers. It is the front door of the Turbokern loop.
Capabilities
01
Capture timeline and kernel traces on DGX, HGX and Jetson with a single run.
02
Learned models label each hotspot as compute, memory, launch or serving bound.
03
Latency and throughput cost assigned to individual kernels and layers.
04
Each kernel placed on the roofline, with a plain-language root cause.
05
A ranked, machine-readable list the optimizers consume directly.
06
Re-profiles on every model or GPU change and flags drift.
Opportunity map
Profyle reads the timeline so a person does not have to. It attributes time to kernels and layers, classifies why each is slow, and estimates what fixing it is worth.
The output is a machine-readable map that Kernl, Quantra and Servex act on directly.
| Rank | Bottleneck | Class | Share |
|---|---|---|---|
| 1 | Attention, decode | Memory-bound | 41% |
| 2 | MLP GEMMs | Low tensor-core use | 23% |
| 3 | Launch gaps | CPU-bound launch | 9% |
How it works
Run the workload under Nsight Systems, Nsight Compute and CUPTI on the target GPU.
Trace features go through the bottleneck classifier.
Cost is assigned to kernels and layers and ranked.
The opportunity map goes to Kernl, Quantra and Servex.
Specification
| Item | Detail | Status |
|---|---|---|
| Input | Model or workload, target GPU, representative inputs | Pilot |
| Trace sources | Nsight Systems, Nsight Compute, CUPTI | Pilot |
| Targets | H100, H200 (pilot); Blackwell, Jetson Orin | Pilot · Planned |
| Output | Opportunity map (JSON), HTML report | Pilot |
| Workload classes | LLM inference first; vision and speech next | Next |
| Narrative summaries | Natural-language root-cause write-ups | Planned |
Place in the loop
Every optimizer starts from Profyle's map, and every profile adds to the dataset Simyx learns from.
FAQ
No. Profyle brings its own capture tooling for the GPU it runs on.
Yes. Many teams will start with a profile to see what optimization could be worth before running anything else.
Profyle runs on a separate replica or on our GPUs, not on production traffic.
Profyle reports CPU launch gaps, data loading and serving queues too, and says when the GPU is not the limit.
Next step
Start the pilot with Profyle and get a ranked opportunity map for your model on your GPU.