SolidRun Bedrock R7000 Edge AI fanless computer combines an 8-core AMD Ryzen 7840HS/U Zen4 processor and up to three Hailo-8 AI accelerators for artificial intelligence applications in harsh environments with an industrial temperature range of -40ºC to 85ºC.
The industrial embedded system comes with up to 64GB DDR5 memory, supports up to three M.2 NVMe SSD, and offers various interfaces such as dual 2.5GbE and four 4K-capable display interfaces using HDMI and DisplayPort. Just like the previous generation mode, the AMD Ryzen Embedded V3000-powered Bedrock V3000 fanless industrial computer, the BedRock R7000 is offered in three different form factors including the 30W and 60W dissipation model with integrated heatsink fins, and a tile variant designed to be fastened to a cold plate on a chassis.
Bedrock R7000 Edge AI specifications:
- SoC – AMD Ryzen 7 7840HS or 7840U octa-core Zen4 processor clocked at 3.8 GHz / 5.1GHz (Turbo) with Radeon 780M GPU; Configurable TDP (cTDP): 35-54W (Note: SolidRun allows power control from 8W to 54W)
- System Memory – Up to 64GB dual-channel DDR5 ECC / non-ECC
- Storage – Up to 3x NVMe PCIe Gen4 x4 socket for SSDs
- AI Accelerators – Up to 3x Hailo-8 M.2 AI acceleration modules or up to 78 TOPS AI inference performance
- Display – 1x HDMI 2.1 up to 8Kp60, 3x DisplayPort 2.1 up to 8Kp60 (2x miniDP + 1x DP)
- Networking
- 2x 2.5 GbE ports via Intel I226 controller
- WiFI 6E and Bluetooth 5.3 via Intel AX210 module
- Optional 4G / 5G modem from Quectel
- Up to 4x external antennas
- USB – 1x USB 3.2 (10 Gbps) port, 3x USB 3.2 (5 Gbps) ports
- Console – Serial over mini USB port
- Misc
- AMI Aptio V BIOS on dual SPI flash for redundancy
- Cooling
- Liquid metal TIM (thermal interface material)
- 360º stacked heat pipes
- Dual-layer chimney effect heat exchanger
- Thermal coupling of all internal devices
- Power Supply – 12V-60V DC via 2-pin terminal block
- Dimensions
- 30W model: 45 mm (W) x 160 mm (H) x 130 mm (D) – 0.9 liter
- 60W model: 73 mm (W) x 160 mm (H) x 130 mm (D) – 1.5 liter
- Tile model: 29 mm (W) x 160 mm (H) x 130 mm (D) – 0.6 liter
- Enclosure – All aluminum enclosure enabling fanless cooling
- Temperature Range – Up to -40ºC to 85ºC range
SolidRun offers DIN-rail, wall, and tabletop mounting options, and supports Windows 10, Windows 11, Windows IoT, and (Ubuntu) Linux operating systems for the fanless Egde AI computer. If you don’t need AI acceleration, the company also designed the “BedRock 7000 Basic” without Hailo-8 modules and fewer M.2 NVMe sockets. Work-in-progress technical documentation is available, and most of the resources will be made available next month (August 2023).
The company expects the high-end Edge AI box PC to be found in various segments of the embedded market, including Industry 4.0, robotics, autonomous guided vehicles, healthcare, transportation, smart cities, retail, smart agriculture, defense, and utilities.
SolidRun did not provide pricing and availability information, but that’s probably because each BedRock R7000 embedded PC is made to order with a wide range of options to choose from (see diagram below). Additional information may be found on the product page and the press release.
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
The Hailo-8 modules seems very unneeded as there is a lot of AI acceleration in the AMD chip.
It has a dedicated AI engine(10 TOPS one), the iGPU which got a very big boost this generation for AI with dual-issue, WMMA and stuff(so it’s able to get like 16Tflops at FP16/BF16) and even the CPU with AVX512-VNNI and AVX512-BF16.
Keeping in mind that there could be a lot of models that aren’t suited to the Hailos architecture and might be far away from using most of the TOPs provided(if the benchmarks were even that honest to start with). It also relies completely on the internal memory, so it’s likely to be unusable for bigger models.
So I don’t see much of a reason to include them.