NVIDIA's Jetson AGX Xavier Carmel Performance vs. Low-Power x86 Processors
Written by Michael Larabel in Processors on 9 February 2019. Page 1 of 6. 23 Comments

Back in our NVIDIA Jetson AGX Xavier benchmarks from December, besides looking at the incredible Carmel+Volta GPU compute potential for machine learning and other edge computing scenarios, we also looked at the ARMv8 Carmel CPU core performance against various other ARM SoCs on different single board computers. But how do these eight NVIDIA Carmel CPU cores compare to x86_64 low-power processors? Here are some of those benchmarks for those curious about the NVIDIA CPU potential.

As a refresher, the NVIDIA Tegra Xavier SoC features eight ARMv8.2 64-bit CPU cores with 4MB of L3 cache and 2MB of L2 cache per two CPU cores (duplex). The "Carmel" cores are one of the many improvements with Xavier over its predecessors.

While the 512-core Volta GPU with tensor cores and the NVDLA accelerators are what make the Jetson AGX Xavier development board really powerful and opening it up to a realm of use-cases, in this article we are just benchmarking the CPU performance. See our aforelinked benchmarks for more well-rounded tests on the AGX Xavier. Due to these low-power x86_64 CPUs not supporting CUDA and the same interfaces, GPU testing wasn't done, as well as because the AGX Xavier does not support OpenCL in order to test the same GPU APIs across all test candidates.

The power consumption / performance-per-Watt was also out of the scope of this round of testing due to differences in other hardware components between the Jetson AGX Xavier and the other low-power/mini PCs being tested and due to the WattsUp Pro reportedly potentially suffering from inaccuracies at very low wattages. Anyhow, this round of testing features a CPU performance look at:

Jetson AGX Xavier - This latest NVIDIA Jetson board 1ith its eight ARMv8.2 cores, 16GB of LPDDR4 memory, and 32GB of eMMC storage. The standard Linux 4 Tegra file-system setup with Ubuntu was used for testing, as was Ubuntu Linux on all of the devices under test today.

Jetson TX2 - The previous-generation with its two Denver2 CPU cores plus a quad-core Cortex-A57, 8GB of LPDDR4 memory, and 32GB of eMMC storage.

ODROID-XU4 - The ODROID-XU4 was included as a capable ARM SBC that packs a fair amount of performance potential. The XU4 features a Samsung Exynos 5422 with Cortex-A15 2GHz and Cortex-A7 octa-core CPU design, 2GB of LPDDR3 memory, and 16GB of storage.

Core i5 7200U - An Intel Core i5 7200U within an MSI Cubi nettop and 8GB of memory. The Core i5 7200U Kabylake is a dual-core processor plus Hyper Threading, 2.5GHz base frequency, and 3.1GHz turbo.

Core i7 4600U - An Intel Core i7 4600U within a CompuLab Intense-PC2 with 8GB of RAM. This old Haswell U-series CPU is dual-core plus HT and has a 2.1GHz base frequency with 3.3GHz turbo and 15 Watt TDP.

Celeron J3455 - An Intel Celeron J3455 within a CompuLab Fitlet2 with 8GB of RAM. The Celeron J3455 is a quad-core processor without Hyper Threading, 1.5GHz base frequency, 2.3GHz burst frequency, and 10 Watt TDP.

Core i5 7500U - An Intel Core i5 7500U Kabylake within a CompuLab Intense-PC3 sporting 16GB of RAM. The 7500U has two cores plus HT, 2.7GHz base frequency, 3.5GHz turbo frequency, and 15 Watt TDP.

These x86 low-power CPU comparison points were based on the devices I had locally for testing that were within a reasonable range to compare against the Jetson AGX Xavier's Carmel CPU cores. All of these Ubuntu Linux AArch64/x86_64 benchmarks were carried out using the Phoronix Test Suite. These weekend benchmarks are primarily being done for reference purposes as opposed to a formal comparison due to the lack of the perf-per-Watt testing this round, etc.

Related Articles
Trending Linux News