
LattePanda Sigma 2 是一款紧凑型单板服务器/计算机,搭载 Intel Core Ultra X7 358H Panther Lake-H 处理器,配备 Intel Arc B390 显卡和 50 TOPS NPU,面向本地 AI 工作负载的合计性能最高可达 180 TOPS。
该板最高支持 64GB LPDDR5X-8533 RAM,提供三个 M.2 M-Key 插槽、一个 M.2 E-Key 插槽和一个用于存储、无线和加速器扩展的 OCuLink 端口,同时还有两个 Thunderbolt 4 端口、双 10GbE、USB 3.2 Gen 2 和 USB 2.0 端口,以及 HDMI 2.1、DisplayPort 2.1 和 eDP 1.5 显示接口。
SoC – Intel Core Ultra X7 Processor 358H
CPU – 16 核(4P + 8E + 4LPE)@ 1.9/1.5/1.5 GHz(基础频率)、4.8/3.5/3.5 GHz(睿频频率)
缓存 – 18 MB Intel Smart Cache
GPU – 12 核 Intel Arc B390 GPU @ 2.5 GHz,最高 122 TOPS(INT8)
AI 加速器 – 50 TOPS NPU;CPU + GPU + NPU 合计:180 TOPS
处理器基础功耗(PBP)– 25 瓦
CPU – 16 核(4P + 8E + 4LPE)@ 1.9/1.5/1.5 GHz(基础频率)、4.8/3.5/3.5 GHz(睿频频率)
缓存 – 18 MB Intel Smart Cache
GPU – 12 核 Intel Arc B390 GPU @ 2.5 GHz,最高 122 TOPS(INT8)
AI 加速器 – 50 TOPS NPU;CPU + GPU + NPU 合计:180 TOPS
处理器基础功耗(PBP)– 25 瓦
系统内存 – 64GB LPDDR5X-8533 RAM;带宽 136.5 GB/s
存储 – 预装 1TB NVMe
可通过 M.2 插槽或 OCuLink 连接器添加更多存储设备
可通过 M.2 插槽或 OCuLink 连接器添加更多存储设备
显示接口 – HDMI 2.1 最高 4096 x 2304 @ 60Hz;通过 USB Type-C 的 DP 2.1 最高 7680 x 4320 @ 60Hz;eDP 1.5 接口最高 3840 x 2400 @ 120Hz
HDMI 2.1 最高 4096 x 2304 @ 60Hz
通过 USB Type-C 的 DP 2.1 最高 7680 x 4320 @ 60Hz
eDP 1.5 接口最高 3840 x 2400 @ 120Hz
音频 – 3.5mm 麦克风与耳机组合插孔;通过 HDMI 或 DP 输出数字音频
3.5mm 麦克风与耳机组合插孔
通过 HDMI 或 DP 输出数字音频
网络 – 通过 Realtek RTL8127 控制器的 2 个 10GbE RJ45 端口
可选通过 M.2 Key-E 模块提供 WiFi 和蓝牙
通过 Realtek RTL8127 控制器的 2 个 10GbE RJ45 端口
可选通过 M.2 Key-E 模块提供 WiFi 和蓝牙
USB/Thunderbolt – 2× Thunderbolt 4(40 Gbps)Type-C 端口;2× USB 3.2 Gen 2(10 Gbps)和 2× USB 2.0 端口
2× Thunderbolt 4(40 Gbps)Type-C 端口
2× USB 3.2 Gen 2(10 Gbps)和 2× USB 2.0 端口
扩展 – 3× M.2 M-Key 插槽(PCIe 4.0 x1、PCIe 4.0 x4、PCIe 5.0 x2);M.2 E-Key 插槽(PCIe 4.0 x1、CNVio3、USB2.0);OCuLink PCIe 4.0 x2;用于底层控制的 30+4-pin GPIO 排针
3× M.2 M-Key 插槽(PCIe 4.0 x1、PCIe 4.0 x4、PCIe 5.0 x2)
M.2 E-Key 插槽(PCIe 4.0 x1、CNVio3、USB2.0)
用于底层控制的 30+4-pin GPIO 排针
其他 – Raspberry Pi RP2350 协处理器,用于控制 GPIO 排针
电源 – 通过 USB Type-C(USB PD)端口输入 20V;通过 5.5 x 2.5mm 电源插孔输入 15~20V DC
通过 USB Type-C(USB PD)端口输入 20V
通过 5.5 x 2.5mm 电源插孔输入 15~20V DC
尺寸 – 146 × 102 mm
该公司表示,该板支持 Windows 11 或 Ubuntu 24.04 LTS 或更高版本。不过,文档显示 Ubuntu 26.04 预装在 1TB SSD 上,并包含所有驱动、用于服务本地 AI 模型(Qwen3.6-35B-A3B INT4)的 OVMS(OpenVINO Model Server),通过端口 8080 上的 OpenAI 兼容 API 提供服务,同时可通过端口 18789 上的浏览器仪表板访问 OpenClaw。LattePanda 解释称,最多可分配 59GB 显存用于 AI 工作负载,为 35B 级量化 MoE 模型和长上下文 AI 代理所需的 KV cache 提供充足空间,Qwen3.6-35B-A3B 模型在本地运行速度约为每秒 30 个 token。
Raspberry Pi RP2350 微控制器进一步增强了 Sigma 2 板(以及由此延伸的 AI 代理)与传感器和执行器交互的能力。该微控制器默认运行 MicroPython 固件,但如果你更喜欢,也可以切换到 Arduino 编程。上述文档提供了更多技术细节和说明,涵盖从底层(GPIOs/BIOS)到 AI 代理和模型的操作。
在 BIOS 中配置功耗限制
我们过去介绍过其他一些 Intel Core Ultra X7 Processor 358H 平台,例如 UP Xtreme PTL Edge AI 迷你 PC、ADLINK COM-HPC-mPTL COM-HPC 小型模块化计算机,以及 ASRock Industrial NUC(S) Ultra 300 BOX 系列。这些都是面向工业和 B2B 客户的完整系统或模块,而 LattePanda Sigma 2 则采用紧凑的 SBC 外形,并向所有人提供。
也就是说,只要你有足够的资金购买,因为它并不便宜。LattePanda Sigma 2 目前以 1TB SSD/64GB RAM 配置提供预订,价格为 $2,599,预订期截至 11 月中旬。之后价格将为 $2,799。
Jean-Luc 于 2010 年创办了 CNX Software,起初只是兼职项目,之后辞去了软件工程经理的工作,并在 2011 年晚些时候开始全职撰写每日新闻和评测。
支持 CNX Software!可通过加密货币捐赠、在 Patreon 上成为赞助者,或在 Amazon 或 Aliexpress 上购买商品。我们还在文章中使用联盟链接,如果你点击这些链接并完成购买,我们会获得佣金。
本文来源:https://www.cnx-software.com/2026/10/29/2600-lattepanda-sigma-2-single-board-server-features-180-tops-intel-core-ultra-x7-358h-soc-for-local-ai-applications/
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
The LattePanda Sigma 2 is a compact single board server/computer powered by an Intel Core Ultra X7 358H Panther Lake-H processor with Intel Arc B390 graphics and a 50 TOPS NPU delivering up to 180 TOPS of combined performance for local AI workloads.
The board features up to 64GB of LPDDR5X-8533 RAM, three M.2 M-Key slots, one M.2 E-Key slot, and an OCuLink port for storage, wireless, and accelerator expansion, as well as two Thunderbolt 4 ports, dual 10GbE, USB 3.2 Gen 2 and USB 2.0 ports, and HDMI 2.1, DisplayPort 2.1, and eDP 1.5 display interfaces.
SoC – Intel Core Ultra X7 Processor 358H CPU – 16-core (4P + 8E + 4LPE) @ 1.9/1.5/1.5 GHz (base frequency), 4.8/3.5/3.5 GHz (Turbo frequency) Cache – 18 MB Intel Smart Cache GPU – 12-core Intel Arc B390 GPU @ 2.5 GHz up to 122 TOPS (INT8) AI accelerator – 50 TOPS NPU; CPU + GPU + NPU combined: 180 TOPS Processor Base Power (PBP) – 25 Watts
CPU – 16-core (4P + 8E + 4LPE) @ 1.9/1.5/1.5 GHz (base frequency), 4.8/3.5/3.5 GHz (Turbo frequency)
GPU – 12-core Intel Arc B390 GPU @ 2.5 GHz up to 122 TOPS (INT8)
AI accelerator – 50 TOPS NPU; CPU + GPU + NPU combined: 180 TOPS
System Memory – 64GB LPDDR5X-8533 RAM; 136.5 GB/s bandwidth
Storage 1TB NVMe pre-installed Additional storage devices can be added through M.2 sockets or the OCuLink connector
Additional storage devices can be added through M.2 sockets or the OCuLink connector
Display Interfaces HDMI 2.1 up to 4096 x 2304 @ 60Hz DP 2.1 via USB Type-C up to 7680 x 4320 @ 60Hz eDP 1.5 connector up to 3840 x 2400 @ 120Hz
DP 2.1 via USB Type-C up to 7680 x 4320 @ 60Hz
eDP 1.5 connector up to 3840 x 2400 @ 120Hz
Audio 3.5mm microphone and headphone combo jack Digital audio output via HDMI or DP
3.5mm microphone and headphone combo jack
Networking 2x 10GbE RJ45 ports via Realtek RTL8127 controllers Optional WiFi and Bluetooth via M.2 Key-E module
2x 10GbE RJ45 ports via Realtek RTL8127 controllers

Optional WiFi and Bluetooth via M.2 Key-E module
USB/Thunderbolt 2× Thunderbolt 4 (40 Gbps) Type-C ports 2× USB 3.2 Gen 2 (10 Gbps) and 2× USB 2.0 ports
2× USB 3.2 Gen 2 (10 Gbps) and 2× USB 2.0 ports
Expansion 3x M.2 M-Key sockets (PCIe 4.0 x1, PCIe 4.0 x4, PCIe 5.0 x2) M.2 E-Key socket (PCIe 4.0 x1, CNVio3, USB2.0) OCuLink PCIe 4.0 x2 30+4-pin GPIO header for hard control
3x M.2 M-Key sockets (PCIe 4.0 x1, PCIe 4.0 x4, PCIe 5.0 x2)
M.2 E-Key socket (PCIe 4.0 x1, CNVio3, USB2.0)
Misc – Raspberry Pi RP2350 co-processor for GPIO header control
Power Supply 20V via USB Type-C (USB PD) port 15~20V DC via 5.5 x 2.5mm power jack
The company says the board supports Windows 11 or Ubuntu 24.04 LTS or later. However, the documentation reports Ubuntu 26.04 is pre-installed on the 1TB SSD with all drivers, OVMS (OpenVINO Model Server) serving the local AI model (Qwen3.6-35B-A3B INT4) through an OpenAI-compatible API on port 8080, and OpenClaw accessible on a browser dashboard on port 18789. LattePanda explains that up to 59GB of VRAM can be allocated for AI workloads, providing enough room for 35B-class quantized MoE models and the KV cache required by long-context AI agents, and the Qwen3.6-35B-A3B model runs locally at around 30 tokens per second.
The Raspberry Pi RP2350 microcontroller further enables the Sigma 2 board, and by extension AI agents, to interact with sensors and actuators. The microcontroller runs MicroPython firmware by default, but you can also switch to Arduino programming if you prefer. The aforementioned documentation provides more technical details and instructions to work from a low level (GPIOs/BIOS) up to AI agents and models.
Configuring the power limits in the BIOS
We covered a few other Intel Core Ultra X7 Processor 358H platforms in the past, like the UP Xtreme PTL Edge AI mini PC , ADLINK COM-HPC-mPTL COM-HPC Mini computer-on-module, and ASRock Industrial NUC(S) Ultra 300 BOX Series . These are complete systems or modules that target industrial and B2B customers, and the LattePanda Sigma 2 is offered in a compact SBC form factor and is available to anybody.
That is, if you have enough funds to cover the purchase, since it’s not exactly cheap. The LattePanda Sigma 2 is currently up for pre-order for $2,599 in its 1TB SSD/64GB RAM configuration until mid-November. After that, the price will be $2,799.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies , become a Patron on Patreon, or purchase goods on Amazon or Aliexpress . We also use affiliate links in articles to earn commissions if you make a purchase after clicking on those links.