Recommended Free Tools
In one reported ImgIng cold start, a 5,955,745-byte ONNX Runtime WebAssembly file took about 13.2 seconds to download, while a 44,279,201-byte AI model finished in 4.188 seconds. The runtime request started only after the model download and its SHA-256 check were complete. The result is a useful case study in how request order and delivery speed can outweigh file size—but it is not a general benchmark for browsers or ONNX Runtime.
What happened in the cold start?
Lank_M, who says they own ImgIng’s model-loading path, reported testing the site’s quick background-removal tier using the ISNet INT8 model. The test was run on September 28, 2026, in a fresh browser context on an M4 Mac with 16 GB of memory and Chromium 149. The author called TYBG.segment(img, 'quick', onProgress) directly, logged network requests and progress events, and used a bottle image as input. Inputs were resized to 1024×1024.
As an Amazon Associate I earn from qualifying purchases.
The WebGPU test used Metal. Its 44,279,201-byte model finished downloading at 4,188 ms, and SHA-256 verification finished at 4,207 ms. Session setup began at 4,250 ms. Only then did the browser request ort-wasm-simd-threaded.asyncify.wasm, an asset reported as 5,955,745 gzipped bytes and served from ImgIng’s own origin. That request finished at about 13.2 seconds. Inference began at 13,725 ms, and the result returned at 14,594 ms.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteIn this implementation, the model and runtime downloads did not overlap: the runtime request followed the model download and hash check. The request sequence—not just the amount of data—therefore shaped the cold-start time.
#1 Best Overall
Why did the smaller file take longer?
File size alone does not determine download duration. In separate curl checks that evening, Lank_M reported the runtime file taking 9.19 seconds and 10.78 seconds across two tries, or about 552–648 KB/s. The model took 4.20 seconds, at about 10.5 MB/s. The files were delivered at very different observed rates, and the runtime was requested later in the sequence.
Those transfer figures describe one ordinary home connection, the hosts involved, and one evening. They do not establish that a small runtime file will usually download more slowly than a larger model. Network conditions, location, time, origin performance, and cache state can all change the result.
Rank #2
Did the WebGPU path still need the WASM file?
In the tested build, yes: Lank_M reported that the WebGPU execution provider was compiled into the WASM asset, so the WebGPU run requested it too. That is a detail of this particular implementation, not a rule for every ONNX Runtime Web build. The runtime components required depend on the selected execution provider and how the application is configured.
Free tools Windows power users keep installed
One-click scans. No signup required.
In a separate run without a usable GPU adapter, the page attempted WebGPU and then showed a switch-to-WASM message at 13,551 ms. The call completed at 16,015 ms. These are reported observations from the same site and test setup, not controlled comparisons of browser fallback behavior generally.
What changed with a different model or a warm cache?
Professional-tier model
Lank_M reported the same ordering in a WebGPU run of ImgIng’s professional tier, using the BEN2 FP16 model, reported at 219,121,675 bytes: the model completed at 20.5 seconds, the runtime at 29.8 seconds, and the full call at 34.8 seconds. These timings are specific to that run.
Warm cache
On a warm cache, the runtime was not requested again under its Cache-Control: max-age=31536000, immutable setting, and the model was served from Cache Storage. In a new page, the reported call took 930 ms with WebGPU or 2,643 ms without a usable adapter. Those figures are site-specific observations, not promises about warm-cache performance elsewhere.
Rank #4
How to investigate a cold start in your own app
The useful first step is to inspect a cold-load Network waterfall, then follow the runtime request as well as the model. Lank_M’s recommendation was: “If you ship in-browser inference, open the Network waterfall on a cold load. Don’t stop at the biggest file. Check when the runtime .wasm request starts relative to the model, and how fast your own origin serves it.”
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Check request order. Note when the model request starts and ends, when any integrity check completes, and when session setup triggers the runtime request. In ImgIng’s reported flow, the runtime download began after the model and hash check.
- Check the asset’s origin and transfer. Identify where the runtime file is served from and compare its observed delivery time with the model’s. The report describes different delivery rates; it does not establish that moving, preloading, or changing either asset will improve another deployment.
- Separate cold and warm behavior. Record whether the runtime is fetched again or served from cache, and where the model comes from. A repeat page load can have a very different critical path from a fresh context.
- Verify the selected runtime components. Confirm which execution provider and WASM assets the deployed build actually requires. Do not assume the tested build’s WebGPU/WASM arrangement applies to yours.
- Repeat under representative conditions. A single waterfall is evidence about that run, not a stable network benchmark. Compare results from the environments and cache states that matter to your users.
The report does not compare fixes or establish which optimization works best. Treat sequencing, asset delivery, caching, and runtime composition as questions to test against your own build rather than as guaranteed remedies.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
