This summary of the video was created by an AI. It might contain some inaccuracies.
00:00:00 – 00:21:04
The video centers around an in-depth examination of the LattePanda Sigma, a highly capable single board computer with notable hardware specifications, including an Intel i5-1340p CPU, up to 32GB of LPDDR5 memory, multiple M.2 slots, and Thunderbolt 4 ports. This board aims to bridge the gap between conventional maker boards and Intel NUC-like performance. Despite its advanced features such as in-band ECC memory and extensive I/O options, the board's high price point raises questions about its competitiveness compared to alternatives like the Intel NUC 13 Pro, Topton MiniPC, and Minisforum UM760.
The technical intricacies of setting up and optimizing the LattePanda Sigma are closely examined, including the challenges of enabling in-band ECC and configuring Thunderbolt DAS storage systems. The presenter also explores the performance impacts of in-band ECC, particularly regarding GPU-intensive tasks, and the efficiency of their experimental setups using Proxmox, Unraid, and Jellyfin.
Network capabilities are tested, showing impressive speeds but highlighting limitations such as increased power consumption and cable constraints with Thunderbolt networking. The video concludes that while the LattePanda Sigma offers robust performance and versatility, its high cost may deter potential buyers, suggesting more affordable mini PC alternatives could provide better value.
00:00:00
In this part of the video, the presenter introduces the LattePanda Sigma, a powerful single board computer featuring an Intel i5-1340p with 12 cores and 16 threads, 16GB of DDR5 memory with ECC support, three M.2 slots, and two Thunderbolt 4 ports. This is a significant upgrade from the earlier LattePanda v1, which was based on x86 architecture and intended to compete with the Raspberry Pi 4. The segment highlights the strong competition in the market for mini PCs, especially from Asian manufacturers like Topton and Minisforum. Additionally, the presenter shifts focus to today’s sponsor, Brilliant.org, explaining how it offers fun and interactive courses on complex subjects like math and computer science, and encourages viewers to try a free 30-day trial. The latter part circles back to the key features that make the LattePanda Sigma stand out among similar Intel-based mini PCs.
00:03:00
In this part of the video, the presenter evaluates the LattePanda Sigma, describing it more as an Intel NUC rather than a Raspberry Pi. The focus is on the board’s Input/Output (I/O) capabilities, which include multiple USB ports, Thunderbolt 4 ports, and Ethernet ports, among others. Notably, the Sigma features a 20-pin GPIO connected to an Arduino-compatible ATMega microcontroller, blending x86 performance with typical ARM-based maker board features. The host emphasizes that despite the built-in Arduino chip, its functionality is isolated from the Intel CPU portion. Additional I/O includes eDP and touch panel outputs, NVMe slots, and SATA ports. The onboard RAM is non-upgradable, with models offering either 16GB or 32GB of high-speed LPDDR5 memory with in-band ECC support, which differs from the on-die ECC found in DDR5.
00:06:00
In this part of the video, the speaker discusses the in-band ECC which uses existing memory channels to make memory modules cheaper, though it reduces performance, especially in GPU-intensive tasks. They review an Asrock industrial NUC with Intel i7-1360p, noting significant performance impacts when in-band ECC is enabled. Intel claims this ECC is only supported on Chrome designs and not Windows, with limited Linux support for 13th gen chips.
The speaker then outlines plans to use Thunderbolt ports on the board to create a compact NAS with a Terramaster Thunderbolt DAS enclosure. They replace the inefficient RAID controller with an ASMedia ASM1166 card to improve power efficiency and drive management. Using a 12V power supply and a Y-splitter, they power both the NAS and the board.
Challenges encountered include the DAS enclosure not powering on automatically after power loss. In terms of software setup, Unraid failed to recognize the hard drives, but Proxmox worked after multiple reboots and re-connections, suggesting potential issues with the Linux kernel or the DAS enclosure.
00:09:00
In this part of the video, the speaker explores various technical issues and experimental setups involving Proxmox, Unraid, and Thunderbolt DAS. They express doubts about using Thunderbolt for permanent storage due to its lack of robustness compared to PCIe. The speaker then decides to run Unraid in a virtual machine on Proxmox, passing through necessary hardware like the Unraid USB drive, SATA controller, and Intel Xe graphics card for hardware transcoding in Jellyfin. Initial setup shows Unraid detecting the drives and creating the storage array, but there’s an issue with the DAS fans not adjusting speeds, leading to overheating drives. The speaker then focuses on hardware video transcoding, setting up Jellyfin in Docker, and benchmarks the performance with a 4K HDR HEVC copy of Dune. They achieve impressive results with VPP tone mapping at 82 FPS and software tone mapping at 102 FPS, alongside running various QuickSync benchmarks, demonstrating significant performance from the 7th gen Intel Xe graphics chip.
00:12:00
In this part of the video, the speaker addresses the complexities of enabling in-band ECC on a LattePanda board, starting with the difficulty of finding the BIOS option and updating the firmware using a ZIP file from GitHub. Despite a successful update, the OS did not show ECC support, requiring the speaker to patch the Linux kernel. They followed a guide, modified the EDAC driver, compiled the kernel, and rebooted the machine to finally enable ECC. However, enabling in-band ECC impacts performance, especially in GPU tasks using slower system memory. Performance benchmarks with ECC disabled show notable improvements. Additionally, the speaker tested Thunderbolt networking by connecting their MacBook to the LattePanda Sigma, creating a new networking device called `thunderbolt0` through a forum guide.
00:15:00
In this part of the video, the presenter discusses network configurations and testing results of their setup. Proxmox is assigned a static IP, and a static IP is also set on their Mac. Network speed tests using iperf3 show 18 Gbps for uni-directional and around 13 Gbps for bi-directional, surpassing 10 Gbps speeds. However, Thunderbolt networking has limitations with cable length, requiring costly optical cables for distances over two meters.
The presenter then examines power consumption, noting the board’s efficiency, drawing as little as 3.6W at idle but rising to 65W during intensive tasks like compiling the Linux kernel. Thunderbolt DAS configuration causes the board to consume more power, being stuck at C2 state and drawing 15W even at idle. Alternatively, a non-Thunderbolt setup with the same components consumes only 8.3W at idle and performs more reliably. The presenter suggests that a drive enclosure based on M.2 might be a better option than Thunderbolt due to these power and reliability concerns.
00:18:00
In this part of the video, the speaker shifts focus from experimental use to discussing the price and practicality of the LattePanda Sigma. The 16 GB model starts at $579, excluding storage and WiFi, while the 32 GB model is $629. Import costs in Germany heighten this significantly. Although the LattePanda Sigma’s performance and extensive I/O options are impressive, its high price makes it less competitive. Alternatives such as the Intel NUC 13 Pro or various other MiniPCs offer comparable or better specs at lower prices. For example, a Topton MiniPC costs approximately 296€ and a Minisforum UM760 with superior performance is available for 419€. Ultimately, the speaker believes the LattePanda Sigma’s versatility is outweighed by its cost, suggesting that consumers might find better value in cheaper alternatives and possibly adding an Arduino separately.
00:21:00
In this part of the video, the speaker expresses gratitude towards their Patrons, acknowledging their support.