Gigabyte AI Top Atom mini Ai Workstation Launched with NVIDIA’s Latest APU Based on the latest Grace Blackwell architecture, it has 20 powerful ARM cores & a powerful GPU.
Ever since Nvidia launched its superchip GB10, everyone has been surprised because a company known for making GPUs had introduced a powerful APU. Nvidia has completely captured the GPU market; at present, Nvidia possesses such advanced technology that you can’t even imagine.
Nvidia has advanced and made GPU technology more efficient, which is why the company is at the top today. When Nvidia designed the GB10 chip in collaboration with Mediatek, it was because instead of designing its own CPU cores, it proved better for Nvidia to develop ARM cores through Mediatek.
With this GB10 Superchip, Nvidia announced and launched its mini AI workstation/supercomputer DGX Spark, but other OEM partners like MSI, ASUS, Acer, Lenovo, Dell, HP, and all other brands have launched their own mini AI workstations with the GB10 Superchip, such as Acer Veriton GN100, Asus Ascent GX10, and Dell Pro Max AI, and now Gigabyte has also launched the AI Top ATOM globally with this superchip. Let’s see what it all offers you.
Table of Contents
Gigabyte AI Top Atom Design:
When it comes to design, all mini AI workstations that come with the Nvidia GB10 Superchip have a similar design. The Gigabyte AI Top ATOM is extremely compact with just a 1-liter chassis volume, approximately 150 × 150 × 50.5 mm, and its total weight is around 1.2 kg. To improve the thermal performance of this AI Atom TOP mini AI workstation, simple air vents are provided at the front and rear. In terms of other specs and ports, all of these Nvidia GB10-based Mini AI Workstations are also similar, because Nvidia has made the same specs mandatory for the GB10 Superchip in all of them. These mini AI workstations have been developed only for professional users, not for gaming, so you will not find RGB lighting in them.
CPU/GPU:
When we talk about performance, the Gigabyte AI Top ATOM comes with NVIDIA’s latest Grace Blackwell architecture-based, efficient yet powerful GB10 APU. This superchip features 20 special ARM cores designed by Mediatek, which are both powerful and power-efficient (10x Cortex-X925 + 10x Cortex-A725), meaning 2 clusters of 10 cores each, with 16MB L3 cache per cluster. Notably, each core has its own private L2 cache reserved. The NVIDIA GB10’s S-dielet and G-dielet are fabricated on advanced 3nm technology. Internally, the GB10 die is divided into two parts, thanks to the new Grace Blackwell architecture—S-dielet and G-dielet. The S-dielet contains Mediatek’s CPU cores, memory subsystems, etc., while the G-dielet houses GPU cores, including fifth-generation Tensor Cores with FP4 (4-bit) precision. The NVIDIA GB10 Grace Blackwell superchip is capable of delivering up to 1 petaFLOP of AI performance (1,000 TOPS) by utilizing FP4 and sparsity.
Image: Nvidia
RAM/Storage:
The Gigabyte AI Top ATOM comes with the same superfast massive 128 GB LPDDR5X unified system memory as other GB10-based AI workstations, along with a 256-bit memory interface, offering 273 GB/s memory bandwidth. This memory will be shared between the CPU and IGPU, though the exact amount of shared memory has not been clarified yet. This massive memory enables local handling of large generative models—up to 200 billion parameters. The Gigabyte Ai Top ATOM will also offer up to 4TB Gen 5 NVMe SSD options.
OS/Software:
Gigabyte Ai Top ATOM Mini AI Workstations provide you with server-grade performance despite their small size. Similar to the Nvidia DGX Spark, they come with Nvidia’s AI software stack and DGX Base OS, which supports commonly used developer tools such as PyTorch, Jupyter, and Ollama.
IO Ports:
On the rear side of the Gigabyte AI Top ATOM Mini AI Workstations, you get a power button, next to which is a 40 Gbps USB 3.2 Gen 2×2 Type-C (PD IN), alongside three USB 3.2 Gen 2×2 Type-C ports (up to 20 Gbps), one HDMI 2.1a port, one RJ-45 connector, 10 GbE, and finally NVIDIA ConnectX-7 ports. As for wireless connectivity, it comes with Wi-Fi 7 and Bluetooth 5.3. Just as you can connect two machines in an NVIDIA DGX Spark to double its computing power, similarly, in the Gigabyte AI Top ATOM, you can connect via ConnectX-7 Smart NIC to double memory and compute capacity, enabling workloads on models with up to 405 billion parameters.
Image: Gigabyte
Opinion:
If you really want a powerful, power-efficient mini AI supercomputer, then the Gigabyte AI Top ATOM will prove to be the best option for you because the Nvidia GB10 chip is truly powerful, thanks to Nvidia’s new Grace Blackwell architecture. You can experience up to 1 petaflop of AI performance. The special feature of this Gigabyte AI Top ATOM is that it offers up to 4TB Gen 5 NVMe options, which you won’t find in other GB10 machines. Regarding the price of the Gigabyte AI Top ATOM, the basic model with 1TB NVMe costs $3,500*, the 4TB NVMe model costs $3,900*, and the 4TB Gen 5 NVMe Gigabyte AI Top ATOM is available for $4,000*.
