NVIDIA Jetson AGX Orin 64 GB

Benchmark Performance

The NVIDIA Jetson AGX Orin 64 GB is a specialized compute module built on the Ampere architecture with a GA10B chip, fabricated on Samsung's 8 nm process with a 200 mm² die. Its benchmark standing places it at the 50th percentile among all GPUs tracked in the database, indicating a mid-pack performance profile that is neither entry-level nor high-end in the broader GPU landscape.

The raw compute specifications reveal a device optimized for parallel workloads rather than raw rasterization. The FP32 throughput is rated at 5.325 TFLOPS, while FP16 performance doubles to 10.65 TFLOPS via the 2:1 ratio. This asymmetry is a clear signal that the hardware is engineered for AI inference and mixed-precision compute tasks, where FP16 acceleration matters more than traditional single-precision gaming loads. The 2048 shading units, 64 texture mapping units, and 32 raster output units form the classical render pipeline, but the pixel rate of 41.60 GPixel/s and texture rate of 83.20 GTexel/s are modest figures that reflect the power-constrained design envelope of a 60 W integrated graphics processor.

Memory configuration is where this module distinguishes itself. The 64 GB LPDDR5 pool operates across a 256-bit bus, delivering 204.8 GB/s of bandwidth. The memory clock runs at 800 MHz, translating to 6.4 Gbps effective. This is a considerable capacity advantage over conventional discrete GPUs, which rarely exceed 24 GB even at the enthusiast tier. The bandwidth, however, is not exceptional; it is adequate for the compute density but does not rival high-end GDDR6X or HBM implementations. Benchmark results indicate that workloads requiring large model residency in memory will benefit disproportionately from this configuration, whereas latency-sensitive or bandwidth-bound tasks may see less uplift.

The production status is listed as end-of-life, which positions this part as a legacy offering. Its launch MSRP is 1,599 USD. With an average benchmark score of zero, the database does not contain standardized performance metrics from typical GPU benchmarks, so the analysis must rely on architectural specifications and the percentile ranking. The lack of nearest rivals in the data set further complicates direct percentage comparisons, but the 50th percentile ranking implies a balanced position—capable in compute-heavy scenarios but not competitive with contemporary desktop gaming GPUs in pure frame-rate terms.

Who Should Consider It

The Jetson AGX Orin 64 GB is not a consumer gaming card. Its design targets embedded and edge computing environments where the combination of a 60 W TDP, integrated form factor (IGP slot width), and PCIe 4.0 x4 interface makes it suitable for compact, power-sensitive deployments. The 100 mm length and 87 mm height dimensions underscore its small physical footprint, enabling integration into robotics, autonomous machines, and portable devices where space is at a premium.

For resolution and settings recommendations, the data suggests that traditional gaming at high refresh rates is not the intended use case. The FP32 throughput of 5.325 TFLOPS places it in the range of older mid-range desktop GPUs, which would struggle with modern titles at 1440p or 4K with high detail settings. Benchmark results indicate that users should not expect playable frame rates in demanding AAA games beyond 1080p with reduced settings. The 64 GB memory capacity is the standout feature, so the target audience is developers running large language models, computer vision pipelines, or multi-application containerized workloads that require substantial memory residency without a discrete GPU's power draw.

The 204.8 GB/s bandwidth is a limiting factor for memory-intensive rendering, but for AI inference where model weights are loaded once and repeatedly queried, the capacity far outweighs the bandwidth constraint. System integrators building edge servers, medical imaging devices, or industrial inspection systems will find the balance of compute, memory, and power consumption more relevant than raw gaming scores. The end-of-life status suggests that new designs should evaluate alternatives, but existing deployments can continue to leverage the 64 GB footprint for years of service.

Ray Tracing and Feature Set

The Jetson AGX Orin 64 GB includes 64 tensor cores, which are the dedicated units for AI acceleration and are fundamental to the Ampere architecture's deep learning capabilities. The tensor core count is modest relative to desktop Ampere parts, but it enables on-device neural network inference without cloud offloading. The architecture supports DirectX 12 Ultimate with feature level 12_2, which includes hardware-accelerated ray tracing and variable rate shading in principle, but the absence of dedicated RT core count in the specification sheet suggests that ray tracing performance is not a primary design focus.

The API support is comprehensive for an embedded module: OpenGL 4.6 and Vulkan 1.4 are both listed, providing broad compatibility with modern graphics and compute frameworks. The DirectX 12 Ultimate designation means the hardware is architecturally capable of DXR workloads, but the compute-first design and modest pixel throughput will limit practical ray tracing to simple scenes or non-real-time applications. The Vulkan 1.4 support is notable for cross-platform development, particularly in Linux-based embedded environments where Vulkan is the preferred low-level graphics and compute API.

Display outputs are listed as "Portable Device Dependent," which means the module does not dictate a fixed set of video connectors. This flexibility is typical for embedded boards where the carrier board determines the actual I/O. The tensor cores are the more significant feature for most adopters, as they enable FP16 accelerated inference at 10.65 TFLOPS, which is the same throughput as the FP32 rate for tensor operations. This positions the module as a capable edge AI processor rather than a graphics powerhouse.

FAQ

Q: What is the memory configuration of the NVIDIA Jetson AGX Orin 64 GB?

A: It features 64 GB of LPDDR5 memory on a 256-bit bus, with a bandwidth of 204.8 GB/s and memory running at 800 MHz (6.4 Gbps effective).

Q: Does the Jetson AGX Orin 64 GB support hardware ray tracing?

A: The architecture is compatible with DirectX 12 Ultimate (feature level 12_2), which includes ray tracing support, but the spec sheet does not list a dedicated RT core count, indicating ray tracing is not the primary strength.

Q: What is the power consumption of this module?

A: The TDP is rated at 60 W, and it uses an integrated graphics processor (IGP) slot width, making it suitable for power-constrained embedded systems.

Q: What is the FP16 compute performance?

A: FP16 performance is rated at 10.65 TFLOPS, which is exactly double the FP32 rate of 5.325 TFLOPS, enabling efficient mixed-precision AI workloads.

Q: Is this product still in production?

A: No, the production status is listed as end-of-life, though it remains a viable option for existing designs and legacy deployments.

Q: What is the physical size of the Jetson AGX Orin 64 GB?

A: The module measures 100 mm in length (3.9 inches) and 87 mm in height (3.4 inches), with a compact form factor for embedded integration.

How It Compares

The database does not list any nearest rivals for the Jetson AGX Orin 64 GB, which makes direct percentage comparisons impossible from the available data. This absence is itself informative: the module occupies a niche that does not overlap with typical consumer or workstation GPUs tracked in the benchmark database. Its 50th percentile ranking relative to all GPUs suggests it sits at the midpoint of the performance distribution, but the lack of peer data means that position is contextual rather than competitive.

Without rival scores or deltaPct values, the comparison must rely on qualitative assessment of the specification sheet. The 64 GB memory capacity is a significant differentiator from any desktop GPU in the database, which typically max out at 24 GB. The 60 W TDP is substantially lower than discrete GPUs, which often exceed 200 W. The integrated form factor and PCIe 4.0 x4 interface further separate it from add-in cards that use x16 slots. In pure compute terms, the 5.325 TFLOPS FP32 is comparable to entry-level desktop Ampere parts, but the memory capacity and power envelope make it a distinct tool for edge AI rather than a direct competitor to gaming or workstation products. The end-of-life status also means it is not positioned against newer generations; it is a final-generation Tegra product whose role is being filled by successors in the Orin family. For users evaluating this module, the comparison should be against system-level requirements—power budget, memory footprint, and physical space—rather than against discrete GPU benchmarks.