iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
In one reported cold run of ImgIng’s browser-based background remover, a 5,955,745-byte ONNX Runtime WebAssembly file finished downloading about nine seconds after a 44,279,201-byte model. The runtime request started only after the model download and its SHA-256 check were complete, so the smaller file became the longer wait. That result describes one site build, machine, browser, workload, and evening—not a general rule about browser AI.
What happened in the cold start
Lank_M, who says they own ImgIng’s model-loading path, reported the test on September 28, 2026. It used the quick background-removal tier with the ISNet INT8 model in a fresh browser context on an M4 Mac with 16 GB of memory and Chromium 149. The test called TYBG.segment(img, 'quick', onProgress) directly, logged network requests and progress events, and used a bottle image resized to 1024×1024. The comparison included a WebGPU-on-Metal run and a Chromium run without a usable GPU adapter. Lank_M’s report
In the cold WebGPU run, the model download completed at 4,188 ms and its SHA-256 verification at 4,207 ms. Session setup began at 4,250 ms, then triggered a request for ort-wasm-simd-threaded.asyncify.wasm, a 5,955,745-byte gzipped asset 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.
The critical detail is the sequence: the runtime request began after the model download and hash check. In this implementation, the two downloads did not overlap. The model was larger, but its delivery was faster and it started first; the runtime was smaller, but it came later and took longer to arrive.
#1 Best Overall
- EVOLUTION AMD RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
Why did the smaller file take longer?
File size alone does not determine the time a dependency adds to a cold start. The relevant questions are when each request begins, how quickly its host delivers it, and whether it can be fetched alongside other required assets. Here, the runtime request was on the critical path after model verification, and the report describes slower delivery from ImgIng’s origin than from the model’s CDN in that run.
In separate curl checks that evening, Lank_M reported runtime-file download times of 9.19 and 10.78 seconds, or roughly 552–648 KB/s, while the 44 MB model took 4.20 seconds at about 10.5 MB/s. These measurements came from one ordinary home connection on one evening. They do not establish that the same origin or file will be slower elsewhere, at another time, or on another network.
Rank #2
- Built for Local AI Development: AMD Ryzen AI Halo is designed for local AI development and inference, featuring 128GB unified memory and support for up to 200B parameter models to build and run intensive AI workloads locally.
- 128GB Unified Memory: Features 128GB LPDDR5x unified memory at 8000 MT/s with 256 GB/s memory bandwidth, providing a shared memory pool across the CPU, GPU, and NPU to support larger AI models.
- AMD Ryzen AI Max+ 395 Processor: Features 16 cores, 32 threads, and Zen 5 architecture, paired with AMD Radeon 8060S integrated graphics featuring 40 RDNA 3.5 compute units and an AMD XDNA 2 NPU with up to 50 TOPS.
- Linux AI Developer Platform: Purpose-built for Linux-based AI development with full AMD ROCm software support and preloaded tools, models, and workflows optimized for local AI development.
- Compact, Connected Design: Includes a 2TB M.2 SSD, 10GbE LAN, Wi-Fi 7, Bluetooth 5.4, USB-C connectivity, and HDMI 2.1b.
A later cold run placed the runtime completion at +34.2 seconds and the model at +4.3 seconds, but the machine was sharing resources with other jobs. Lank_M cautioned that only the ordering—not those later timings—should be taken from that run.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsWhat the GPU and fallback runs show
The reported WebGPU run still fetched the WASM asset because, according to Lank_M, the tested build included the WebGPU execution provider in that runtime file. That is a detail of the configuration tested, not evidence that every ONNX Runtime Web build or WebGPU setup requires the same asset. Runtime components and dependencies need to be checked for the specific build and execution provider.
Rank #3
- EVOLUTION RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
In the run without a usable GPU adapter, the page attempted WebGPU, displayed a switch-to-WASM message at 13,551 ms, and returned a result at 16,015 ms. This illustrates the tested page’s fallback behavior; it does not establish how other applications handle adapter failures or how their fallback changes startup time.
Warm cache changes the picture
Lank_M reported that on a warm cache the runtime was not requested again under its Cache-Control: max-age=31536000, immutable policy, while the model came from Cache Storage. A new page then completed the call in 930 ms with WebGPU or 2,643 ms without a usable adapter. These are site-specific observations; the report does not establish that every browser, cache configuration, or update cycle will behave the same way.
How to diagnose your own cold start
Use a genuinely cold-load Network waterfall rather than inferring bottlenecks from asset sizes. Lank_M’s advice is: “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.
- Reproduce a cold load. Use a browser context without the relevant cached runtime and model assets, and record the browser, device, workload, and execution path so the conditions are clear.
- Find the runtime request. In the Network panel, locate the ONNX Runtime
.wasmrequest. Note its start time relative to model download completion, any verification, and session setup. - Check delivery. Record the response size, duration, and serving origin. Compare actual observed behavior rather than assuming a larger file must be slower.
- Compare cold and warm behavior. Check whether the runtime is fetched again, whether the model is served from a browser cache or Cache Storage, and how those states affect the call.
- Confirm build requirements. Verify which runtime assets the selected build and execution provider require, including the path used when the expected GPU adapter is unavailable.
If the waterfall shows the runtime request starting only after model transfer and verification, investigate whether the sequence is required and whether those downloads can safely overlap in your implementation. If the runtime begins early but remains slow, investigate the serving origin and its delivery under your users’ network conditions. These are diagnostic paths to test, not fixes whose benefit is established by this report.
What this case does—and does not—prove
The report’s professional-tier run showed the same ordering: the 219,121,675-byte BEN2 FP16 model completed at +20.5 seconds, the runtime at +29.8 seconds, and the whole call at 34.8 seconds. This is another observation from the same site and reporting context, not an independent benchmark.
The useful lesson is narrow but practical: a cold-start waterfall can reveal a late, slow dependency that file-size rankings hide. This case does not identify a universally best hosting arrangement, preload strategy, request sequence, or execution provider. Its timings cannot predict yours; inspect the requests and cache state in your own deployment.
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.
Recommended Free Tools
