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
Local AI can be slow on a legal PDF for several different reasons: extracting its text, OCRing scanned pages, preparing a long prompt, loading the model, or generating an answer. Find which stage is taking the time before changing model settings or buying hardware. OCR can be far slower than ordinary text extraction, while long contexts and concurrent requests can increase memory use.
Find out which stage is slow
Time the workflow in separate parts: opening the document, extracting text, running OCR, loading the model, waiting for the first token, and generating the response. These measurements help distinguish a document-preparation delay from slow model inference. If the application spends most of its time processing a scanned PDF before generation begins, a faster GPU may not fix that wait.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
MINISFORUM MS-02 Ultra Workstation Mini PC, Intel Core Ultra 9 285HX (24C/24T, up to 5.5GHz), PCIe... | $1,659.00 | Buy on Amazon |
| 2 |
|
GMKtec EVO-X2 AI Mini PC Ryzen Al Max+ 395 Superchip 128GB LPDDR5X 2TB SSD | $3,649.99 | Buy on Amazon |
There is no single bottleneck or ideal configuration for every legal workflow. Results depend on the PDF, model, runtime, hardware, and task.
Check whether the PDF needs OCR
Use embedded text when it is available
A text-based PDF may have usable embedded text that can be extracted directly. A scanned or image-only page generally needs optical character recognition (OCR) before a language model can work with its contents. Check pages rather than assuming the whole file requires OCR.
#1 Best Overall
- High-Performance AI Processor：The MS-02 Ultra features an Intel Core Ultra 9 285HX (24C/24T, up to 5.5 GHz, 13 TOPS NPU), delivering fast and efficient performance for AI inference, algorithm development, and media workloads. A PCIe x16 expansion slot supports desktop-class GPU upgrades for advanced model training and accelerated computing tasks. It's ideal for creators, engineers, and teams handling intensive parallel workloads.
- 4 × M.2 PCIe 4.0 + 4 × DDR5 SODIMM slots:Four DDR5 SODIMM slots support up to 256 GB of memory, while ECC helps maintain data integrity in mission-critical environments. Four PCIe 4.0 M.2 slots support up to 24 TB of storage, supporting RAID 0/1/5/10, combining high-speed performance with data protection. It allows for the creation of independent scratch disks, media libraries, and project drives, providing high-throughput for production workflows.
- PCIe & USB 4.0 v2: Up to three PCIe slots can be equipped, including a dual-slot x16 GPU. The main slot supports PCIe 5.0, meeting the needs of high-bandwidth creative and computing workloads. USB 4.0 v2 (80Gbps) supports high-bandwidth external storage and displays.
- Ultra-fast Networking: Wi-Fi 7 further enhances wireless performance with next-generation speeds and low-latency stability. Intelligent bandwidth switching optimizes throughput in different network environments, ensuring optimal performance for enterprise or local networks. Dual 25GbE ports (providing up to approximately 3.125 GB/s bandwidth, about 25 times faster than traditional 1GbE), enabling seamless large-scale file transfers and parallel computing. 10GbE and 2.5GbE ports, with support for Intel vPro technology, ensure enterprise-grade remote management and deployment flexibility.
- Server-grade thermal architecture: Utilizing a dedicated CPU/GPU airflow design, equipped with a 6-pipe dual-fan cooler, it maintains stable performance even under sustained loads, delivering up to 140W Turbo power while maintaining a 100W TDP, and operating with noise levels as low as 36 dB. An integrated 350W power supply ensures stable and reliable output for demanding computing tasks and fully loaded extended configurations.
PyMuPDF says OCR is roughly one thousand times slower than standard text extraction. Its OCR guidance recommends doing OCR only when needed and retaining the resulting TextPage so later text extraction or searches can reuse it. If you analyze the same file repeatedly, reusing a validated OCR result can avoid repeating that work.
Verify OCR output against the page
OCR is not a guarantee of faithful document structure. PyMuPDF notes that Tesseract output does not preserve original font styling and does not recognize vector graphics. Compare extracted text with the page image when exact wording, tables, stamps, handwritten notes, or layout matter. A missed word or misread character can be more consequential in a legal document than a modest delay.
Reduce unnecessary prompt work
Send the passages needed for the question
A long file can lead to a long request. When your application supports it, retrieve or select the relevant passages and retain page numbers or section references so you can check the answer against the source. This is a workflow to validate on your own files, not a universally proven chunk size or overlap setting for dense legal text.
There is no established chunk size that works best for every contract, case file, or statute. Test retrieval on representative documents and questions: too little surrounding text can omit a definition or exception, while sending much more text than needed increases the request the model must handle.
Set context for the task, not for the maximum
In its FAQ, Ollama documents a default context window of 2048 tokens and describes changing it with /set parameter num_ctx or an API option. A larger context can accommodate more input, but it is not free: context allocation uses memory, and parallel requests increase total allocation. Ollama says requests may queue if memory is insufficient.
Set the context to what the task requires and observe memory use and queueing. Maximizing context is not a general speed fix; the appropriate value depends on the model, workload, and available memory.
Check model placement and hardware fit
Confirm what the runtime is using
Model loading and generation depend on the runtime and hardware. If you use Ollama, its FAQ documents ollama ps for inspecting model placement. Check whether the intended accelerator is being used, whether the model fits in its memory, and whether the system is under memory pressure. These checks can reveal a placement or capacity issue, but do not assume GPU use is the cause of every slowdown.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Rank #2
- 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.
For Windows, Microsoft’s Windows ML overview describes CPU, GPU, and NPU execution providers. It says discrete GPUs generally offer maximum performance for high-throughput generative-AI workloads, while NPUs are suited to sustained, battery-efficient inference. Microsoft also cautions that results vary by hardware configuration and model. This is platform guidance, not a benchmark of every local AI runtime or legal task.
Windows AI APIs, Foundry Local, and Windows ML are distinct options with different model and device support. Microsoft’s overview advises choosing based on the task and platform requirements.
Account for backend-specific limits
The llama.cpp SYCL backend documentation lists supported Intel GPU families and warns that Intel integrated GPUs with fewer than 80 execution units will likely be too slow for practical use with that backend. It also notes that device memory limits model size. This warning applies to the documented SYCL backend; it should not be generalized to other runtimes or hardware.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Consider a smaller or quantized model carefully
Quantization reduces weight-storage requirements, which can help when memory is a constraint. Microsoft’s Windows ML efficiency guide gives a per-weight comparison of four bytes for FP32 and one byte for INT8, while noting that actual savings depend on model structure and quantization method. That storage comparison does not establish a particular speed increase or legal-domain accuracy.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Compare the current model with a smaller or more quantized option using the same representative legal documents and questions. Check both latency and answer quality, including citations, exceptions, defined terms, and exact wording where relevant. A smaller memory footprint alone does not show that a model is suitable for your work.
Use hardware examples as context, not a buying prescription
The Council of Bars and Law Societies of Europe’s 2026 guide to local inference for lawyers gives an illustrative setup of approximately €2,000 using September 2025 prices: a fast CPU, 128 GB of RAM, and multiple lower-cost GPUs totaling 24 GB of VRAM. The guide describes it as capable of running 20–40B text-only models at a comfortable speed. These are guide estimates based on those dated prices, not a current quotation or universal performance guarantee; the guide also warns that RAM prices are volatile.
The guide discusses memory capacity and bandwidth as relevant factors, but its example does not establish which hardware will help a particular slow workflow. Before shopping, identify whether your delay is in OCR, memory, model placement, or generation, then check model fit, runtime and operating-system support, power, and cost. No single GPU recommendation follows from the evidence for all legal-document users.
Quick Recap
A practical troubleshooting order
- Time the stages. Record document load, text extraction, OCR, model load, time to first token, and generation separately.
- Inspect the PDF. Determine which pages have usable text. OCR only pages that need it, reuse validated OCR output, and compare important text with the page image.
- Inspect runtime and memory. Check model placement and resource use; in Ollama, use
ollama ps. Note whether requests are queued or memory is insufficient. - Right-size the request. Use the context the task needs and provide relevant passages with their page or section references when your application allows it.
- Compare model choices. Test a smaller or quantized model against the current one on the same legal tasks, judging both speed and answer quality.
- Consider hardware last. Match any upgrade to the measured bottleneck, model size, accelerator support, memory requirements, and practical constraints.
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.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems
