NVIDIA today announced what is arguably its most important product of the year, one that—if done right—could help it break into the consumer PC processor market. NVIDIA envisions a change in the PC user interface paradigm. For over 40 years, the PC human interface contract involved you interacting with software through their toolbars, dialog boxes, and work areas, to get work done. NVIDIA sees AI agents fundamentally change the way consumers interact with PCs. Imagine you opened up Photoshop, and instead of manually running a complex workflow, simply prompted it your agent, and it did the rest. NVIDIA thinks the processor of the future should be able to not just run your applications, perform its core function as a general purpose CPU, but also run large AI models and agents entirely locally, while also fulfilling consumer-segment needs as an entertainment and gaming powerhouse. Enter, the NVIDIA RTX Spark.
NVIDIA RTX Spark is a heavily modified version of the NVIDIA GB10 Superchip running the DGX Spark AI development and local inferencing machine. While the DGX Spark can be considered a highly compacted workstation or AI accelerator; the NVIDIA RTX Spark is a full-tilt consumer PC processor being designed for notebooks and mini PC desktops. NVIDIA has heavily collaborated with Microsoft in its development, so it will exclusively power notebooks and desktops from major PC OEMs with Windows 11 AI PC pre-installed, along with full Copilot+ native acceleration readiness. NVIDIA is also heavily engaged with ISVs and game developers to build new Arm-native versions of popular applications that fully utilize the architecture-level advancements of RTX Spark, for everything else, Microsoft has a robust OS-level translation layer that lets you use x86-64 applications on Windows for Arm. NVIDIA is giving this processor a powerful iGPU based on the GeForce RTX "Blackwell" architecture, so it can also power 1440p-class AAA gaming, including full DirectX 12 Ultimate features such as ray tracing, path tracing; along with NVIDIA's own DLSS 4.5 and future DLSS 5.
The company is building the GB10 Superchip on the TSMC 3 nm EUV foundry node. The CPU complex is based on NVIDIA's "Grace" microarchitecture, and has been co-developed with MediaTek. It features 20 cores that are a mix of performance and efficiency cores.
The iGPU is based on the "Blackwell" graphics architecture, and features 48 streaming multiprocessors (SM) worth 6,144 CUDA cores, and up to 1 petaFLOP of FP4 AI performance. This features a similar specs-sheet to the desktop GeForce RTX 5070 graphics card. The chip features an LPDDR5X memory interface with bandwidth of 300 GB/s. It also features a 600 GB/s NVLink that's configured to put out five PCI-Express Gen 5 lanes. The chip supports up to 128 GB of unified memory, so it's capable of running up to 200B parameter AI models locally. There is also a low-power NPU on the silicon, so the chip meets Microsoft Copilot+ requirements. The chip gets the full NVIDIA software stack, including CUDA, Tensor RT, NVFP4, DLSS 4.5, RTX ray tracing, Reflex, and G-SYNC.
NVIDIA will announce several processor models based on the silicon, differentiating them with CPU core and iGPU SM counts. PC OEMs will offer these chips with memory options ranging between 16 GB and 128 GB.
NVIDIA has a reference notebook model specification that every OEM has followed to launch their first wave of notebooks powered by the RTX Spark. This specification calls for 14-inch or 16-inch sizes with 16:10 aspect-ratio displays with color-accurate tandem OLED panels that meet NVIDIA G-SYNC standards; HD cameras, provide all-day battery life, matte-glass touchpads, precision-machined aluminium chassis, and contemporary high-bandwidth connectivity such as USB4. Unlike the DGX Spark, it lacks ConnectX networking, instead featuring more contemporary PC networking interfaces such as Wi-Fi 7 and GbE up to 10 Gbps.
NVIDIA is engaging with nearly every company in the AI ecosystem to develop applications optimized for RTX Spark on Windows 11. The company is also engaging with all creator productivity application vendors to not just make their applications native to the Windows on Arm machine architecture, but also optimized for RTX Spark, along with new agentic AI user interfaces. An example of this is new editions of Adobe Premiere and Photoshop for RTX Spark.
The company is also working very closely with nearly every game developer to port and optimize their games to Windows on Arm with RTX Spark awareness, and not just through Prism. This includes availability of the full DirectX 12 Ultimate API, the entire NVIDIA RTX and DLSS software stacks.
Today, leading PC OEMs that include ASUS, Dell, HP, Lenovo, MSI, and Microsoft Surface, announced their first laptops that implement not just RTX Spark, but also NVIDIA's product design standards detailed earlier. The OEMs have pulled out their most premium brands, such as ASUS ProArt, Dell XPS, HP Omnibook, MSI Prestige, Lenovo Yoga, and Microsoft Surface Ultra. These notebooks will be available in the market from Fall 2026.
NVIDIA is also announcing RTX Spark desktops. These are essentially mini PCs, not much larger than a DGX Spark. OEMs building these desktops include Acer, ASUS, Dell, GIGABYTE, HP, MSI, and Lenovo.
This question was the elephant in the room during NVIDIA's press briefing, and the simple answer to that is that the DGX Spark is an AI development platform, driven by Linux. The target audience of DGX Spark has everything it needs on Linux, including the ConnectX high-bandwidth networking interface. On the other hand, RTX Spark is a proper consumer Windows 11 AI PC chip for not just AI development and inferencing, but also rich consumer entertainment and gaming experiences. It comes with PC-relevant connectivity and interface. The ace up its sleeve is that it's a Windows PC platform, with 99% of NVIDIA's ISV engagement effort directed Windows on Arm optimized ports of popular software and games.
While NVIDIA did not detail the processor models and differentiation within the product lineup in the press briefing, from what we gather, NVIDIA does intend to bring RTX Spark to the mainstream, with memory sizes starting at 16 GB, and lots of room to carve out mid-range SKUs. At the very least, you can expect a killer laptop from one of the OEMs that takes the fight to Apple MacBook Neo and MacBook Air.
The company also did not put out any performance numbers, but if the DGX Spark is anything to go by, the RTX Spark is competitive with Apple's M5 and M5 Pro chips, outclassing them in AI acceleration performance. NVIDIA's efforts to heavily optimize popular Windows PC apps and games for the RTX Spark is bound to make the platform extremely competitive with the MacBook.
The closest x86 competitor to the RTX Spark would be the AMD Ryzen AI Max 400, which combines up to 16 "Zen 5" AMD64 CPU cores (32 threads) with an RDNA 3.5 iGPU that has 40 compute units, but its RDNA 3.5 graphics architecture might get outclassed by the 48 "Blackwell" SMs on the RTX Spark. Intel has no real competitor to this chip, other than, perhaps, the Core Ultra Series 3 "Panther Lake" with 4P+8E+4LPE and 12 Xe3 cores.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。