Reviewed by Jonathan West · Updated Oct 16, 2026

GPD Pocket 4 Specs, Price, and Local AI Use

What the Ryzen AI 9 HX 370 handheld costs and which local AI models it can hold.

Reviewed by Jonathan West · Updated Oct 16, 2026

The GPD Pocket 4 is an 8.8-inch handheld Windows mini laptop. With the AMD Ryzen AI 9 HX 370 and 32GB or 64GB of memory, it can run small local AI models. The flagship configuration pairs a 12-core Central Processing Unit (CPU) and a Neural Processing Unit (NPU) rated at up to 50 Trillion Operations Per Second (TOPS) with integrated AMD Radeon 890M graphics. The fan is loud under load, the battery is short when gaming, and the keyboard is cramped.

Reporting from Liliputing indicates the convertible device features a screen that folds over the keyboard for tablet use. At launch, Liliputing reported that backers were expected to receive units starting in January 2025. Hardware testing from Notebookcheck recorded dimensions of 22.2 by 206.8 by 144.5 millimeters, an overall chassis weight of 777 grams, and a power supply weighing 202 grams, though initial launch materials listed the weight at 770 grams.


GPD Pocket 4 Specs and Price

The GPD Pocket 4 combines modular physical ports, high-resolution display hardware, and mobile AMD processors starting at an entry price of $895. Launch tiers reported by Notebookcheck set the base configuration with an AMD Ryzen 7 8840U processor at $895, or $829 during its initial Indiegogo campaign. The flagship edition featuring the AMD Ryzen AI 9 HX 370 processor, 64GB of RAM, and a 2 Terabyte (TB) Solid-State Drive (SSD) launched at $1,466.

In its review, the Notebookcheck review referenced its tested 32GB configuration at approximately $1,300 in the evaluation text, $1,400 on its technical specification sheet, and $1,475.95 in a linked Amazon marketplace listing. A subsequent German retail listing for the same 32GB and 2TB configuration listed the price at roughly 2,143 euros. Prices vary by region and time.

In its overall evaluation, Notebookcheck awarded the device a rating of 80 percent, categorizing it as good and calling it a real laptop replacement with just a few compromises.

  • Chassis and Dimensions: Measures 22.2 by 206.8 by 144.5 millimeters; weighs 777 grams with a 202-gram power supply (initial launch specifications cited 770 grams).
  • Processor Choices: AMD Ryzen AI 9 HX 370 (12 cores, 24 threads; 4 Zen 5 cores at 5.1 GHz plus 8 Zen 5c cores at 3.3 GHz), AMD Ryzen AI 9 HX 365, or AMD Ryzen 7 8840U.
  • System Memory: 32GB of soldered dual-channel LPDDR5x-7500 on reviewed configurations; 64GB available on the flagship tier.
  • Onboard Storage: 2TB M.2 2280 SSD on tested models; drive is user-replaceable and subject to thermal throttling under extended loads.
  • Launch Pricing Structure: Started at $895 ($829 Indiegogo price) for the Ryzen 7 8840U; $1,466 for the top-tier Ryzen AI 9 HX 370 with 64GB RAM and 2TB storage.

Local AI Execution on the AMD Ryzen AI 9 Processor

The GPD Pocket 4 can run small, quantized local language models directly on device by using unified system memory and the AMD Ryzen AI 9 HX 370 processor. The HX 370 silicon contains an AMD XDNA2 Neural Processing Unit rated at up to 50 TOPS alongside integrated AMD Radeon 890M graphics. The mini laptop lacks a discrete GPU, so integrated graphics rely on shared system memory.

Small quantized open-weight models fit in the 32GB configuration on paper when run with Ollama or LM Studio. According to our site guides, Llama 3.1 8B quantized to 4-bit precision fits in about 6GB of VRAM. Similarly, Google Gemma models at 4-bit 9B precision require about 6GB of memory, while compact Gemma 1B to 4B builds execute acceptably on CPU cores alone. Microsoft Phi models also run without discrete accelerators; Phi-2 and Phi-3 mini function on a CPU using 8GB to 16GB of RAM, while Phi-4 (14B) at 4-bit precision occupies roughly 8GB of memory and executes on a CPU equipped with 16GB to 32GB of RAM.

Our advice is to choose the 64GB memory tier if you plan to experiment with larger open-source language models. In our view, a 70B model quantized to 4-bit precision requires roughly 40GB of memory and will not fit inside the 32GB configuration. Because the integrated Radeon 890M GPU draws directly from unified LPDDR5x system memory, users should treat discrete VRAM allocation figures as rough guidelines and test with their own prompts. Detailed configuration steps appear in our guides for how to run Llama locally, how to run Gemma locally, and how to run Phi locally.

  • Llama 3.1 8B: Requires roughly 6GB of memory at 4-bit quantization.
  • Gemma Architectures: 4-bit 9B builds require roughly 6GB of memory; smaller 1B to 4B models run acceptably on CPU cores alone.
  • Phi Model Family: Phi-2 and Phi-3 mini operate on 8GB to 16GB of system RAM; 14B Phi-4 at 4-bit fits into 8GB of memory and runs on CPU with 16GB to 32GB of RAM.
  • Llama 3.1 70B Boundary: Demands roughly 40GB of memory at 4-bit precision; our advice is that it will not fit in 32GB; consider the 64GB configuration.

Display Quality, Connectivity Ports, and Modular Expansion

The GPD Pocket 4 features an 8.8-inch In-Plane Switching (IPS) touchscreen with a 16:10 aspect ratio and a 144 Hertz (Hz) refresh rate. Laboratory measurements recorded by Notebookcheck confirmed a resolution of 2560 by 1600 pixels across 343 Pixels Per Inch (PPI), paired with a glossy screen coating. Peak panel brightness reached 568.8 candelas per square meter (cd/m2) with an average across the panel of 523.6 cd/m2, an internal contrast ratio of 1278:1, and color gamut coverage reaching 92.7 percent of Display P3 and 99.6 percent of sRGB.

The display panel includes no support for High Dynamic Range (HDR), Organic Light Emitting Diode (OLED) contrast levels, Variable Refresh Rate (VRR), or AMD FreeSync adaptive sync technologies. Other hardware includes a 5 Megapixel (MP) camera supporting 1440p video at 30 frames per second without a physical privacy shutter, dual 2-Watt stereo speakers delivering loud volume output, an Intel AX210 wireless card providing Wi-Fi 6E, and Bluetooth 5.3.

The fixed rear chassis houses a 2.5 Gigabits per second (Gbps) Registered Jack 45 (RJ-45) Ethernet port, a 10 Gbps USB-C 3.2 Gen 2 port with DisplayPort and Power Delivery, a 40 Gbps USB-C 4.0 port with DisplayPort and Power Delivery, and a dedicated microSD card slot. The left edge features a High-Definition Multimedia Interface (HDMI) 2.1 port, a USB-A 3.2 Gen 2 port, and a reset button, while the right side carries a USB-A 2.0 port and a 3.5 millimeter headset jack. Optional rear modules screw into the rear to add an extra microSD card slot, an RS232 serial interface (Recommended Standard 232), a Keyboard Video Mouse (KVM) module, or 4G Long Term Evolution (LTE) cellular connectivity.

  • Display Specifications: 8.8-inch 16:10 IPS panel at 2560 by 1600 resolution (343 PPI), 144 Hz refresh rate, 568.8 cd/m2 peak brightness, 92.7% Display P3, and 99.6% sRGB.
  • Fixed Ports: One RJ-45 Ethernet port (2.5 Gbps), one USB-C 4.0 port (40 Gbps), one USB-C 3.2 Gen 2 port (10 Gbps), one USB-A 3.2 Gen 2 port, one USB-A 2.0 port, HDMI 2.1, microSD, and a 3.5 mm audio jack.
  • Modular Expansion Options: Screw-in rear modules add an RS232 serial port, KVM console access, a secondary microSD reader, or 4G LTE cellular data.
  • Wireless and Multimedia: Intel AX210 Wi-Fi 6E, Bluetooth 5.3, dual 2-Watt stereo speakers, and a 5 MP 1440p webcam lacking a physical privacy cover.

Hardware Tradeoffs and Bottlenecks of the GPD Pocket 4

The Notebookcheck review calls the processor underclocked, with a power limit of 28 Watts PL2 and 25 Watts PL1. The review measured the CPU at roughly 15 to 20 percent slower than the average laptop with the same chip.

The GPD Pocket 4 scored 16,546 points in Cinebench R23 multi-core against an average of 20,658 for this CPU, alongside a single-core score of 2,006 points. In Geekbench 6.7 testing, the mini laptop posted a multi-core score of 13,426 points compared to an average of 14,644 for this CPU. Notebookcheck also measured the GPD Win 4 (Ryzen 7 6800U) at 10,546 in Cinebench R23 multi-core and the GPD Pocket 3 at 3,634. The M.2 2280 SSD also throttles under load.

The internal cooling fan produces loud, noticeable fan noise when gaming and emits a high-pitched tone under heavy processing loads, though pressing the Plus key allows users to toggle between Low and High fan profiles. The chassis incorporates a 45 Watt-hour (Wh) battery (listed as 44.8 Wh in launch documents) that delivers short operating runtimes during gaming and average overall battery longevity. Furthermore, the chiclet keyboard layout is cramped, nearly identical to the Pocket 3, and omits a Caps Lock Light-Emitting Diode (LED). The compact touchpad measures just 6.3 by 3.3 centimeters, and placing the mouse buttons on the opposite edge of the deck forces users to operate the controls with both hands.

  • Processor Underclocking: Configured at 28W PL2 and 25W PL1, running 15 to 20 percent behind the average laptop using the Ryzen AI 9 HX 370.
  • Cooling and Acoustics: Cooling fans produce loud, high-pitched acoustic noise under load, switchable between Low and High profiles via the Plus key.
  • Storage Bottlenecks: The user-replaceable 2TB M.2 2280 SSD suffers from thermal throttling during sustained data transfers.
  • Input Compromises: Cramped keyboard lacks a Caps Lock LED, while the 6.3 by 3.3 cm trackpad separates mouse buttons onto the opposite side of the deck.
  • Battery Endurance: The 45 Wh battery pack exhibits short runtime when gaming and average longevity.

Purchaser Criteria and Alternative Hardware Options

In our view, the GPD Pocket 4 is worth buying if you need physical Ethernet, RS232 serial interfaces, or KVM capabilities inside an 8.8-inch pocketable form factor. It also suits you if your workflow requires running small local AI models on the go. At Layer3Labs, we build custom agent integrations and workflow automations, and we do not sell hardware or software licenses.

If you need portable artificial intelligence performance without cramped keyboards, compare options in our roundup of the best AI laptops, see small desktop machines in our guide to the best AI mini PCs, or examine alternative form factors in our analysis of the best pocket AI devices.

Calculate your exact memory footprints with our local AI hardware calculator before choosing between the 32GB and 64GB configurations of the GPD Pocket 4.

  • Target Audience: Users who require hardware RS232, KVM access, and 2.5 Gbps Ethernet in a pocket chassis.
  • Who Should Look Elsewhere: Users needing long battery life when gaming, a full-size keyboard, or 70B models on the 32GB configuration.

Frequently Asked Questions

  • The GPD Pocket 4 is an 8.8-inch convertible handheld Windows mini laptop featuring a touchscreen that folds over the keyboard for tablet use. Weighing 777 grams, the device runs on AMD mobile processors, with flagship models pairing the 12-core AMD Ryzen AI 9 HX 370 CPU with up to 64GB of RAM and modular rear hardware ports.
  • The GPD Pocket specs for the reviewed unit include an AMD Ryzen AI 9 HX 370 processor (12 cores, 24 threads), Radeon 890M integrated graphics, an XDNA2 NPU rated at up to 50 TOPS, 32GB of LPDDR5x-7500 RAM, and a user-replaceable 2TB M.2 2280 SSD. It features an 8.8-inch 144 Hz 2560x1600 IPS touchscreen, a 45 Wh battery, Wi-Fi 6E, Bluetooth 5.3, and rear ports including RJ-45 Ethernet and USB-C 4.0, plus optional KVM or RS232 modules.
  • The GPD Pocket 4 launched with an entry-level AMD Ryzen 7 8840U configuration priced at $895 ($829 on Indiegogo), while the flagship AMD Ryzen AI 9 HX 370 edition with 64GB of RAM and 2TB SSD was priced at $1,466. Notebookcheck noted testing unit pricing between $1,300 and $1,475.95 on Amazon, while one German retailer listing showed roughly 2,143 euros. Prices vary by region and time.
  • Yes, the GPD Pocket 4 can run small local AI models such as 4-bit Llama 3.1 8B, Gemma 9B, and Microsoft Phi-4 (14B) using its AMD Ryzen AI 9 HX 370 processor, integrated Radeon 890M graphics, and shared LPDDR5x RAM. However, larger 40GB models like 4-bit Llama 3.1 70B will not fit inside 32GB of RAM, so 64GB is the configuration to consider. The device relies on shared system memory with integrated graphics rather than a discrete GPU.

Assess Your Edge Hardware and AI Deployments

Book a consultation to discuss where local AI fits your team's workflows.

Book a Consultation