- Pre-order
Over 500,000 customers already trust us. With us you get:
same-day shipping for products marked “24h shipping”
our own service centre in Poland
30 days to return and free delivery in Poland from PLN 500
4.9/5 on TrustMate and 4.9/5 on Google (almost 4,000 reviews)
VAT invoice, leasing and instalments for businesses
For businesses
individual quote for larger quantities
VAT invoice
For schools and universities
orders based on requests for quotation
help with choosing kits for your lab
LattePanda Sigma 2 - local AI computer - Intel Core Ultra X7 358H - 64 GB RAM - 1 TB SSD - 180 TOPS - DFR1316
LattePanda Sigma 2: 3.5-inch x86 computer for local AI with Intel Core Ultra X7 358H, 64 GB LPDDR5X and up to 180 TOPS (Arc B390 GPU + NPU).
- 2 x Thunderbolt 4, 2 x 10Gb Ethernet, HDMI 2.1, 4 x M.2, OCuLink
- 1 TB NVMe drive with Ubuntu, OpenVINO and local models included
- RP2350 microcontroller (MicroPython) for sensors and actuators; power adapter included
We could not load the price.
The promotional price does not combine with quantity discounts.
A local AI computer that fits inside a robot enclosure
LattePanda Sigma 2 is a single-board x86 computer with an Intel Core Ultra X7 358H processor (Panther Lake). The processor has 16 cores up to 4.8 GHz, Intel Arc B390 graphics with 12 Xe cores and a built-in NPU. Together they deliver up to 180 TOPS of AI performance (INT8), and 64 GB of LPDDR5X-8533 memory lets you run quantized 35B-class MoE models locally.
The board is 3.5 inches (146 x 102 mm) and has two Thunderbolt 4 ports, two 10Gb Ethernet ports, four M.2 slots and an OCuLink connector. The 1 TB NVMe drive comes with Ubuntu, GPU and NPU drivers and a ready-made environment for local AI.
LattePanda Sigma 2 with a fan and a 1 TB NVMe drive installed
180 TOPS
total INT8 AI performance: GPU 122, NPU 50, CPU 8
64 GB
LPDDR5X-8533, 136.5 GB/s (theoretical), up to 59 GB for GPU
2 x 10 Gb/s
RJ45 Ethernet plus 2 Thunderbolt 4 ports (40 Gb/s)
4.5 W
average idle power draw; 38 W in a local LLM test
What sets this model apart
64 GB of shared memory and three AI engines on one board
The previous LattePanda Sigma had a Core i5-1340P processor and 32 GB of RAM. Sigma 2 has a Core Ultra X7 358H processor with Arc B390 graphics and an NPU, plus twice the memory (64 GB LPDDR5X). According to the manufacturer, it generates text in a local LLM test almost twice as fast, at an average draw of 38 W instead of 35 W. An RP2350 microcontroller for hardware control is included as well.
59 GB
of memory you can allocate to the GPU for model weights and context memory
4 x M.2
3 M-Key slots (PCIe 5.0 x2, 4.0 x4, 4.0 x1) and an E-Key slot for a wireless module
2 x TB4
Thunderbolt 4 and OCuLink PCIe 4.0 x2 for an external graphics card
Sigma 2 mainboard without cooling: processor, memory and M.2 slots on the underside
Local language models and agents without the cloud
The 1 TB NVMe drive (WD SN7100) has Ubuntu, GPU and NPU drivers, OpenVINO Model Server, OpenClaw and local models preinstalled. Once you start the local inference API, agents and your own applications use the model on the board, with no cloud API key.
- Frameworks: Intel OpenVINO (CPU, GPU, NPU), Ollama, llama.cpp
- Models: language, vision-language, YOLO detection, speech processing
- Expansion: AI accelerators in M.2 M-Key, external GPU via OCuLink or Thunderbolt 4
- Systems: Windows 11 and Ubuntu 24.04 LTS or newer
Ports: Thunderbolt 4, 10GbE, HDMI 2.1 and OCuLink
The board edge carries a full set of connectors like a workstation, and the underside has four M.2 slots. You can output video from HDMI 2.1 (up to 4096 x 2304 @ 60 Hz), from Thunderbolt 4 (DisplayPort 2.1, up to 7680 x 4320 @ 60 Hz) or from eDP 1.5 to a built-in panel.
- USB: 2 x Thunderbolt 4 (USB-C, 40 Gb/s), 2 x USB 3.2 Gen 2 (10 Gb/s), 2 x USB 2.0, USB 2.0 header
- Network: 2 x 10Gb Ethernet RJ45 (RTL8127); wireless connectivity via an M.2 E-Key module (not included)
- M.2: M-Key 2280 PCIe 5.0 x2, M-Key 2280 PCIe 4.0 x4, M-Key 2230/2280 PCIe 4.0 x1, E-Key 2230 (PCIe 4.0 x1, USB 2.0, CNVio3)
- OCuLink: SFF-8611, PCIe 4.0 x2, for an external graphics card
- Headers: RS-232/485 (9-pin COM), 34-pin GPIO, front panel, audio, 2 x 12 V fan, 40-pin eDP, I2C touch
Rear edge: USB 3.2, two 10GbE RJ45 ports and HDMI 2.1
Active cooling included: a 12 V fan installed at the factory
RP2350 for sensors and actuators
Alongside the Intel processor runs an RP2350 microcontroller with factory MicroPython firmware. Intel runs the model and the application, while the RP2350 reads sensors and drives actuators via I2C, SPI, UART and analog inputs on the GPIO header. Programming is made easier by separate MCU_BOOT and MCU_RST buttons, and you can enable automatic start on power-up with a DIP switch.
A close look at Sigma 2
Is it the right choice for you?
A good choice if:
- You want to run language models and agents locally, with 64 GB of memory shared with the GPU
- You are building an AI device or robot and need x86, Thunderbolt 4, 10GbE and GPIO on one board
- You are setting up a homelab or edge server that, according to the manufacturer, draws an average of 4.5 W at idle
- You plan to expand with M.2 accelerators or an external GPU via OCuLink
Consider another option if:
- You do not need AI performance or 10GbE and an x86 computer for ordinary tasks is enough: see LattePanda Sigma (i5-1340P)
- You need a small board for simple projects, in which case choose LattePanda IOTA (N150)
Sigma 2 vs the previous LattePanda Sigma
| LattePanda Sigma 2 | LattePanda Sigma | |
|---|---|---|
| Processor | Intel Core Ultra X7 358H, 16 cores | Intel Core i5-1340P |
| Graphics and NPU | Intel Arc B390 (12 Xe) + NPU, up to 180 TOPS | Intel Iris Xe |
| RAM | 64 GB LPDDR5X-8533 | 32 GB |
| Drive in the kit | NVMe 1 TB with Ubuntu | NVMe 500 GB |
| Power draw (LLM test / idle) | 38 W / 4.5 W | 35 W / 5 W |
| Microcontroller | RP2350 (MicroPython) | ATmega32U4 |
| See | this page | LattePanda Sigma |
Sigma 2 data and the power draw comparison come from the manufacturer.
Fun fact
Why does an x86 board need a second processor?
Since its first generation, LattePanda has combined an x86 computer with a microcontroller. In Sigma 2 this is the RP2350, which handles sensors and actuators, while Intel takes care of the AI model and the application. The manufacturer describes this duo as the foundation for robots, interactive devices and AI systems with sensors.
Source: DFRobot: LattePanda Sigma 2 (One Board for AI and Hardware Control)
Frequently asked questions
How large a language model can I run on Sigma 2?
The manufacturer cites quantized 35B-class MoE models: you can allocate up to 59 GB of the 64 GB memory to the GPU for model weights and conversation context memory.
Is WiFi included in the kit?
No. The board has an M.2 E-Key slot (PCIe 4.0 x1, USB 2.0, CNVio3) for a WiFi and Bluetooth card, which you need to buy separately. For wired networking you have two 10Gb Ethernet ports.
Can I connect an external graphics card?
Yes, via OCuLink (PCIe 4.0 x2) or Thunderbolt 4. The manufacturer suggests this as a way to increase AI performance, alongside accelerators in the M.2 M-Key slots.
What system is on the drive?
Ubuntu with a ready-made environment: GPU and NPU drivers, OpenVINO Model Server, OpenClaw and local models. The board also works with Windows 11, but you install it yourself.
What should I use to power the board?
The board requires DC from 15 V to 20 V at a power of at least 120 W or USB-C PD 20 V at a power of at least 100 W. A power adapter with a DC cable and AC cables (EU, US, UK) are included in the kit.
What is the RP2350 on the board for?
It is a microcontroller with MicroPython firmware that reads sensors and drives actuators via I2C, SPI, UART and analog inputs on the GPIO header, while the Intel processor handles the model and the application.
Kit contents
- 1 x LattePanda Sigma 2 (Core Ultra X7 358H, 64 GB)
- 1 x Active fan (installed)
- 1 x M.2 NVMe WD SN7100 1 TB drive with Ubuntu (installed)
- 1 x Metal bottom plate
- 3 x Thermal pads for M.2 drives
- 1 x Power adapter with DC cable
- 3 x AC cables (EU, US, UK)
- 1 x RTC battery (installed)
- 1 x User manual
No enclosure and no WiFi/Bluetooth module.
Specifications
| Parameter | Value |
|---|---|
| Manufacturer | LattePanda (DFRobot) |
| Manufacturer code | DFR1316 |
| Type | x86 single-board computer for local AI, 3.5-inch format |
| Processor | Intel Core Ultra X7 358H (Panther Lake), 16 cores, 16 threads, up to 4.8 GHz |
| Graphics | Intel Arc B390, 12 Xe cores, up to 2.5 GHz, up to 122 TOPS INT8 |
| NPU | Intel NPU, up to 50 TOPS INT8 |
| Total AI performance | up to 180 TOPS INT8 (GPU + NPU + CPU) |
| Microcontroller | RP2350 with MicroPython firmware |
| RAM | 64 GB LPDDR5X-8533 dual-channel, theoretical bandwidth 136.5 GB/s, up to 59 GB for GPU |
| Drive | M.2 NVMe 1 TB WD SN7100 (in the kit, with Ubuntu) |
| Video outputs | HDMI 2.1 (up to 4096 x 2304 @ 60 Hz), DisplayPort 2.1 via USB-C/Thunderbolt 4 (up to 7680 x 4320 @ 60 Hz), eDP 1.5 4-lane (up to 3840 x 2400 @ 120 Hz) |
| USB | 2 x Thunderbolt 4 (40 Gb/s), 2 x USB 3.2 Gen 2 (10 Gb/s), 2 x USB 2.0, USB 2.0 header |
| Network | 2 x 10Gb Ethernet RJ45 (Realtek RTL8127) |
| M.2 | M-Key 2280 PCIe 5.0 x2; M-Key 2280 PCIe 4.0 x4; M-Key 2230/2280 PCIe 4.0 x1; E-Key 2230 (PCIe 4.0 x1, USB 2.0, CNVio3) |
| OCuLink | SFF-8611, PCIe 4.0 x2 |
| Audio | 3.5 mm combo jack, front panel audio header |
| Headers | 34-pin 2.54 mm GPIO; 9-pin RS-232/485; 9-pin front panel; 4-pin USB 2.0; 2 x 4-pin 12 V fan; 40-pin FPC eDP; 6-pin FPC touch (I2C) |
| Power supply | DC 15 V to 20 V, min. 120 W (5.5 x 2.5 mm jack, 2 mm 8-pin connector); USB-C PD 20 V, min. 100 W |
| RTC battery | CR1220 (3 V) socket and 2-pin connector; battery installed |
| Security | TPM 2.0 in the processor |
| Operating systems | Windows 11; Ubuntu 24.04 LTS or newer (Ubuntu installed on the drive) |
| AI software | OpenVINO, OpenVINO Model Server, Ollama, llama.cpp, OpenClaw |
| Operating temperature | 0 °C to 40 °C, humidity 0 to 80 % |
| Board dimensions | 146 x 102 mm (3.5 inches) |
Useful links
| Name | Value |
|---|---|
| Voltage to | 15.0 V |
| Voltage from | 20.0 V |
| Voltage nominal | 15-20 V |
| UC - Microcontroller | Intel Core Ultra X7 358H |
| UC - RAM | 64 GB |
| UC - core | 16. Hexadeca Core |
| UC - Digital pins | 14 |
| UC - External memory | SSD 1 TB |
| UC - freq | 4.8 GHz |
| UC - Ethernet | yes |
| UC - WiFi | Nie |
| UC - USB | 6x USB |
| UC - microSD | Nie |
| UC - Bluetooth | Nie |
| UC - HDMI | yes |
| UC - I2C | yes |
| UC - UART | yes |
| UC - Connector |
|
| UC - SPI interface | yes |
| UC - LCD interface | yes |
| UC - Camera interface | no |
| UC - Linux | yes |
| UC - Android | no |
| UC - Windows | yes |
| Width | 25 cm |
| Height | 20 cm |
| Depth | 9 cm |
| Weight | 2 kg |
GLS
Shipping in 24 hours
from €13.28
The full delivery terms are on the page Delivery and payment.
30 days to return, without giving a reason
If the product does not meet your expectations, you have 30 days from receiving the parcel to withdraw from the contract. The product only has to be complete. The right of return belongs to consumers (a purchase unrelated to business activity).
Complaints
If the product turns out to be faulty, you can make a claim under the guarantee or the statutory warranty. We handle claims without delay, at the latest within 14 working days of receiving the report together with the goods.
Be the first to ask a question about this product!
Instructions and documents
GPSR data
Products from the same category
See all
