NVIDIA Unveils GB10 Superchip: A Leap Forward in AI PC Technology

NVIDIA's new GB10 Superchip for AI PCs is a total tease of its debut in gaming laptops, Mini-PC

TL;DR: NVIDIA's new GB10 Superchip, featuring the Blackwell GPU architecture and advanced multi-die packaging, powers the compact DGX Spark AI workstation. It delivers up to 31 TFLOPS FP32 and 1000 TOPS AI performance with 128GB unified memory, enabling scalable AI workloads and flexible deployment for AI PCs and developer systems. NVIDIA has provided more details on its new GB10 Superchip with a Blackwell GPU at the Hot Chips 2025 event, a new chip that will power multiple DGX AI Mini supercomputers. The first system announced powered by the new NVIDIA GB10 Superchip is the DGX Spark, which is the first step from the company into the "AI PC" segment, with NVIDIA partners also announcing their own GB10-powered "AI PC" platforms. NVIDIA used the event to detail its GB10 Superchip, and how it scales the Blackwell GPU architecture down into small-sized developer and workstation systems. The new GB10 Superchip has a large number of innovations from the datacenter mixed in with the Blackwell GPU architecture that is traditionally in gaming and AI GPU markets (GeForce RTX 50 series, GB100, GB200, GB300 AI GPUs). In order to create the GB10 Superchip, NVIDIA uses technologies from the datacenter like NVFP4, CUDA, SLANG, TensorRT, vLLM, CX-7 NIC, NVLINK C2C, TMEM, and more, crammed into a Mini-PC platform in a small form factor. This is possible through the use of a multi-die packaging technology, a very low-power C2C interface, and Unified Memory Architecture (UMA). The results of this allow GB10 Superchip to power the new DGX Spark Workstation, with key features and benefits including: * GB10 Grace Blackwell Superchip: Accelerates AI, Data Science, Compute, Rendering & Visualization * 128GB Coherent Unified System Memory: Works with Large AI models up to 200 billion parameters, fine-tune models of up to 70 billion parameters * ConnectX-7 Networking: Connect two DGX Spark systems together to work with models of up to 405b parameters * DGX Base OS and NVIDIA AI Software Stack: Seamlessly move workloads from DGX Spark to DGX Cloud or any accelerated data center or cloud infrastructure * Flexible deployment configurations: Configure as an AI Workstation or a network-connected personal AI cloud * Great Desktop Experience: Multi-head display support and flexible connectivity * Compact, power-efficient design: Easily fits on any desk, powered by a standard wall outlet In the photo above is the GB10 Superchip and what makes it tick, with the chip featuring two dielets -- an S-Dielet which contains the CPU, memory subsystem, and more, along with a G-Dielet which contains the Blackwell GPU cores -- with the two dielets packaged together using Advanced 2.5D packaging, fabbed on TSMC's new 3nm process node. NVIDIA uses an Arm-based Arch v9.2 processor with two clusters of 10 cores making 20 cores total, with each of the cores individually housing private L2 cache, and 16MB of L3 cache per cluster for 32MB in total. NVIDIA says that the GPU inside of the GB10 Superchip is an iGPU based on the NVIDIA Blackwell architecture, packing 5th Gen Tensor Cores, and support for DLSS 4 and Ray Tracing. The GPU is capable of up to 31 TFLOPS of FP32 compute performance, and up to 1000 TOPS for AI workloads. The memory subsystem inside of the NVIDIA GB10 Superchip SoC has support for a 256-bit LPDDR5X coherent memory (UMA) with up to 9400MT/s speeds, providing up to 301GB/sec of memory bandwidth, with capacities of up to 128GB. This memory system fabric is a high-performance coherent fabric that also supports CHI-E Coherency Protocol, with the GPU having access to the entire system bandwidth of 600GB/sec (Aggregate) over the C2X interface. The GB10 Superchip has PCIe, USB, Ethernet over PCIe, and supports 4 concurrent displays (3 x DP + 1x HDMI) at up to 4K 120Hz with DP Alt-mode support, and up to 8K 120Hz over HDMI 2.1. The GB10 Superchip also supports Dual Secure Root support, SROOT processor, OSROOT processor, and support for fTPM and discrete TPM, and a 140W TDP. NVIDIA also allows for multiple GB10 Superchips to be connected through its NVIDIA ConnectX technology to scale throughput, bandwidth, and DRAM capacities to support larger AI models. ConnectX NIC is connected to the GB10 SoC over the PCIe Gen5 x8 interface, with units communicating together over Ethernet. NVIDIA has said that the GB100 Superchip SoC is a successful collaboration with NVIDIA and MediaTek, as the CPU IP is provided by MediaTek. GB10 went under extensive performance modeling of GPU memory traffic into the memory subsystem from MediaTek. Okay... after all of that, what does it mean?NVIDIA is making a new class of computer designed, scalable, powerful, from the ground up for a totally new market. It's a tease of things to come... and here's why. We've been hearing about NVIDIA's new N1X and N1 processors for a while now with more on that in the links above, a purported 2026 release, with the new N1X chips using Arm-based CPU cores with Blackwell GPU cores. The biggest hurdle with the broader consumer market is that Arm processors require an Arm-capable operating system (OS). Windows on Arm is much better now than it used to be, but it's a far, far cry from Windows. You can't just run all of your software like you would on a x86-based Windows system, and good luck trying to play most games. If you're web browsing, even working on it, Arm-based Windows PCs aren't all that bad, but they're not ready for mass market. I'd like to think that Microsoft is working behind the scenes as much as it can with NVIDIA to make its Windows on Arm OS work wonders on the new N1X AI PC processors. Arm-based processors are going nowhere, there's more being made now than ever before, it's just not in the mass market like it is in servers. Arm processors can work wonders on operating systems like Linux, even Apple processors for macOS are Arm-based, but on Windows? Unless you've tried one, thinking it runs Windows and that it supports all of the software and games you're running now, it just won't work... and if it does (outside of Arm-based software, games, etc) it's run through emulation, and that's an entire other conversation. NVIDIA's new GB10 Superchip is a big splash in the AI PC market, it's powerful, but it supports everything a regular PC does -- HDMI 2.1 for up to 8K 120Hz, DP for up to 4K 120Hz, Ethernet, USB, PCIe ports, and more... all inside of a 140W chip? Not bad at all... something that would better fight AMD's new Strix Halo APUs, and its next-gen Medusa Halo APUs which will make for some powerful AI PC and gaming systems in the future. The thing is, NVIDIA has been working with Alienware on next-gen gaming laptops that use its new N1X processor, which would see an Arm-based CPU using a Blackwell-powered GPU, in a totally new gaming laptop in the market.

NVIDIA Dissects Its GB10 Superchip For DGX AI PCs: 3nm With 20 ARM v9.2 CPU Cores, 1000 TOPS NVFP4 Blackwell GPU, LPDDR5x-9400 Memory Support, 140W TDP

NVIDIA has just detailed its GB10 Superchip with Blackwell GPU, which is being used to power several DGX AI Mini supercomputers. NVIDIA's DGX Spark, the first system to be announced with the GB10 Superchip, has been making headlines. The system is NVIDIA's foray into the "AI PC" segment, and ever since the announcement, several others have spawned their own GB10 "AI PC" platforms. Today at Hot Chips 2025, NVIDIA is giving a deep dive into its GB10 Superchip and how it scales the Blackwell architecture down to Mini developer and workstations. The idea behind DGX Spark was the design a Mini AI Supercomputer with the Blackwell architecture. To make this happen, NVIDIA developed the GB10 Superchip, which combines innovations from datacenters, such as NVFP4, CUDA, SLANG, TensorRT, vLLM, CX-7 NIC, NVLINK C2C, TMEM, and more, down to a Mini PC platform which utilizes a small form factor, made possible using multi-die packaging tech, a very low-power C2C interface, and Unified Memory Architecture (UMA). As a result, the DGX Spark Workstation was built, which offers the following key features and benefits: So let's dive into the specifications of the GB10 Superchip. First up, we have the SoC composition, which shows that the chip itself is composed of two dielets, an S-Dielet which houses the CPU, memory subsystem, etc, and a G-Dielet which houses the GPU core. These two dielets are packaged together using Advanced 2.5D packaging and are fabricated on TSMC's 3nm process technology. The CPU is based on the ARM Arch v9.2 architecture with 20 cores in total. There are 2 clusters of 10 cores each, and each core has a private L2 cache and a 16 MB L3 cache per cluster, so 32 MB in total. The GPU is based on the GB100 Blackwell architecture and is considered an iGPU since it is on the same package and silicon. It features 5th Gen Tensor Cores with DLSS 4 support and RTX Ray Tracing cores. It produces up to 31 TFLOPs of FP32 and 1000 TOPS of NVFP4 (FP4) compute for AI workloads. The GPU also gets an additional 24 MB of L2 cache. Moving into the memory system, the NVIDIA GB10 Superchip SOC features support for 256b LPDDR5x (UMA) with up to 9400 MT/s speeds, enabling up to 301 GB/s of raw bandwidth, and up to 128 GB of maximum capacities. The system fabric is a high-performance coherent fabric that offers support for CHI-E Coherency Protocol. The GPU has access to the entire system bandwidth of 600 GB/s (Aggregate) over the C2X interface. There's also 16 MB of System Level Cache, which serves as L4 for the CPU, and enables power-efficient data-sharing between the multiple engines on the SoC. The C2C interface is also high-bandwidth and low-power, enabled through NVIDIA's NVLINK architecture. On the connectivity side, NVIDIA's GB10 Superchip SoC offers PCIe, USB, Ethernet over PCIe, and drives up to 4 concurrent displays (3 DP + 1 HDMI) at up to 4K @120Hz with DP Alt-mode, and up to 8K @ 120Hz with HDMI 2.1a. Security features include Dual Secure Root support, SROOT processor, OSROOT processor, and support for both fTPM and discrete TPM. The whole chip has a TDP of 140W. Following is the block diagram of the NVIDIA GB10 Superchip SoC: Scalability is also another fun aspect of the GB10 Superchip. You can connect multiple GB10 chips through NVIDIA's ConnectX Technology and scale throughput, bandwidth, and DRAM capacities to support larger AI models. The ConnectX NIC is connected to the GB10 SoC using a PCIe Gen5 x8 interface, and the units communicate with each other using Ethernet. NVIDIA calls the GB10 Superchip SoC a successful collaboration between them and Mediatek since the CPU IP is from Mediatek. The chip underwent extensive performance modeling of GPU memory traffic into Mediatek's memory subsystem.