NVIDIA 64-bit Tegra K1 steps away from out-of-order execution, does 7 instructions per cycle

Some new details have been revealed about the 64-bit version of NVIDIA's Tegra K1 "Denver" SoC. This new dual-core chip is more than double the size of the Cortex-A15 that powers the 32-bit version of the Tegra K1 and one of the things that's really remarkable about it is that NVIDIA chose to step away from the out-of-order execution (OoOE) engine that is used by all modern high-end ARM and x86 processors.

In an OoOE design, the processor itself decides which code should be executed at any given cycle which makes these designs faster than in-order counterparts. The drawback however is that OoOE designs are larger and use more power consumption. NVIDIA did things a little different by crafting "Project Denver", an in-order architecture that uses a dynamic optimization program that runs on one of the two CPUs. This program calculates and optimizes the most efficient way to execute code and stores the data inside a 128MB memory buffer. Often used pieces of code are stored so there's no need to repeat this process over and over.

The execution engine of the 64-bit Tegra K1 is very wide, it's capable of processing up to seven instructions per cycle! NVIDIA says the 64-bit Tegra K1 will be available in frequencies of up to 2.5GHz. Overall this is shaping up to be one of the more interesting ARM designs in a long while but we'll have to wait for the first third-party benchmarks to see if NVIDIA can deliver what it promises. According to the company itself, its Tegra K1 can deliver performance similar to a low-end Intel Haswell processor. Unfortunately, NVIDIA doesn't mention any specifics about power draw.

Speaking about energy efficiency, NVIDIA also mentions that Denver will have new low latency power-state transitions and extensive power-gating and dynamic voltage and clock scaling based on workloads.

The 64-bit Tegra K1 is anticipated to show up in mobile devices later this year.





Sources: HotHardware and NVIDIA