NVIDIA Tegra SOC Chip Image

There is a lot of virtual ink spilled over the upcoming NVIDIA CPU architecture which company calls "Project Denver". Not a lot is known though, on when the design is expected to debut.

There is a lot of virtual ink spilled over the upcoming NVIDIA CPU architecture which company calls "Project Denver". Not a lot is known though, when the design is expected to debut.

According to Jen-Hsun's talk at the third quarter financial results conference call, NVIDIA is working on Project Denver to "supplement, add to ARM's capabilities by extending the ARM's architecture to segments in a marketplace that they are not themselves focused on".

It is no secret that NVIDIA sees Project Denver as a processor not oriented towards tablets and netbooks, but rather desktops and high-performance computing. According to information available, the first product based on the 64-bit Denver architecture will be Stark, i.e. Iron Man – Tegra SOC set for introduction in late 2014, with real availability in 2015.

Prior to the debut of NVIDIA's own 64-bit architecture, the company will launch Kal-El and Kal-El+ (28nm die-shrink), both based on 32-bit ARM Cortex-A9 architecture, followed by Wayne (Cortex-A15) and Logan, all before Stark comes alive.

NVIDIA is developing Project Denver for better part of the past decade. The company acquired Colorado-based Stexar in 2006, with a plan to develop a binary compatible architecture with the x86 instruction set, with or without Intel's permission/license. The original architecture called for a similar approach used by Transmeta with a translation layer being the core part of the architecture. However, luckily for NVIDIA, then tiny ARM started its ascent into the contemporary computing with the new generation of smartphones. As such, NVIDIA's R&D shifted focus from being binary compatible with x86 into an extended ARM architecture. Needless to say, this practically required starting from scratch.

Given that NVIDIA takes four years to develop a new GPU and that typically, CPU requires five years of development, the first silicon coming out of Stexar will only come after eight years of development. One of terms we heard about Project Denver from a highly ranked executive in a competing firm was "Lost in Rockies".

Yet, NVIDIA cannot take any chances and needs a proprietary, ARM-compatible architecture to extract X86 out of their GPGPU servers. Project Denver is expected to bridge the performance gap between Tegras oriented towards smartphones and other mobile devices with the desktop and server performance requirements. According to Jen-Hsun, ARM currently has no plans to do that in mid-to-long term. 

The project shares similar fate to Intel's own Haswell, which changed its description and architecture for about four times, first being on the map as a NetBu(r)st architecture, only to launch half a decade later as a completely new architecture.

In any case, Stexar team has full support of NVIDIA management, and Stark should arrive on the market only after the market accepts 64-bit ARM as a mainstream computing architecture.