HP EliteBook X G1a: Fine for CAD, Weak for LLMs

The HP EliteBook X G1a is a good fit for school, coding and light-to-moderate CAD, but it is the wrong laptop for medium local LLMs. The reason is that the 14-inch EliteBook X G1a is not the Strix Halo machine its “G1a” name and AI branding suggest. Spec listings and reviews show it on Strix Point chips (Ryzen AI 7 Pro 360 with Radeon 880M, or Ryzen AI 9 class parts with the 890M), with 32 to 64 GB of soldered LPDDR5X. If you need 128 GB of shared memory, you want HP’s sibling, the ZBook Ultra G1a. Confirm the exact SKU in HP’s QuickSpecs before ordering, because HP sells several configurations under similar names.

Skip to the picks

What “medium” LLMs actually need

Local model size is set by memory first and bandwidth second. A rough rule for 4-bit quantised weights is about 0.6 GB per billion parameters, plus a few GB for context.

• 7B to 14B models need about 5 to 10 GB. Any 16 GB machine handles them, and the EliteBook is comfortable here.
• 27B to 32B models need roughly 18 to 24 GB. A 64 GB EliteBook can hold them, but you give up RAM for the OS, IDE, CAD and a game engine at the same time.
• 70B models need roughly 40 to 45 GB. This is where a 64 GB laptop with an integrated GPU runs out of room once everything else is open.

Speed is the second limit. Token generation is bound by memory bandwidth, and a 128-bit LPDDR5X bus like the Strix Point’s gives well under half the bandwidth of the 256-bit bus on Strix Halo. Expect slow output from anything over about 14B on the EliteBook. No independent lab has published local-LLM tokens-per-second figures for this model that this page could verify, so treat that as an inference from bus width, not a measurement.

Where the EliteBook is the right call

Reviewers describe it as a polished, light business machine. TechRadar’s review says it lasts a working day on one charge, and the 2.8K OLED panel is a highlight. StorageReview covers the same OLED model, with the 74.5 Wh battery and 100 W USB-C adapter. Itdaily’s verdict, “solid, but better on paper”, is the caution worth keeping: its real-world performance does not match what the AI-PC branding implies. Those are reviewers’ reports; this site has not handled the laptop.

For the other items on a student’s list:

• School and coding: more than enough. Eight Zen 5 cores, 32 GB or more of RAM and fast NVMe storage compile and run IDEs, Docker and WSL2 without trouble.
• Game development: fine for 2D, small 3D projects and Godot or Unity work. Unreal Engine 5 with large levels will stress an 880M or 890M, and shader compiles will be slow.
• CAD: fine for Fusion 360, Onshape, and small-to-medium SolidWorks or Inventor parts. Large assemblies, heavy rendering and viewport-intensive work favour a discrete GPU with certified drivers. Check your school’s software list; some programs only certify specific GPUs. For a related look at the same question, see HP Omen 16 Slim: Enough for CAD Work and Gaming?.

Comparison

Who should buy which

Buy the ZBook Ultra G1a if running 30B-plus models locally is something you will do regularly, not occasionally. Its design is what makes that possible: memory the GPU can address in bulk. It is the pick for this combination of CAD, coding, game development and LLMs in a single 14-inch machine.

Choose the EliteBook X G1a if local AI means 7B to 14B models for coding assistance and the rest of the day is classes, code and light CAD. In that case the lighter build and battery life matter more than a memory pool you would rarely use. The EliteBook does not deserve a skip, because it is a capable laptop; it simply isn’t what the “medium LLM” requirement needs.

Choose a discrete-GPU workstation like the ThinkPad P1 if CAD is the main workload and you want CUDA for AI work. Its limit is VRAM: small models run quickly, but anything past mid-teens of billions of parameters spills into system RAM and slows sharply. Because the memory is soldered on all of these machines, buy the capacity you will need in three years; it cannot be upgraded later. Pairing a smaller laptop with a desktop or cloud GPU for occasional large-model work is also reasonable. For smaller models on modest hardware, see Phi-3 Mini on a Low-End Laptop: Use Q4.