AMD Instinct MI308X vs NVIDIA Jetson T5000 Comparison

Head-to-Head Benchmarks

The recorded data shows no head-to-head benchmark entries for this pairing, and neither part has accumulated any wins in the database. Both the AMD Instinct MI308X and the NVIDIA Jetson T5000 hold a percentile rank of 50 against all GPUs, with an average benchmark score of 0. This absence of measured performance data means the comparison must rest entirely on the architectural and specification fields recorded for each part.

The AMD Instinct MI308X delivers 81.72 TFLOPS of FP32 and FP16 (1:1) throughput, while the NVIDIA Jetson T5000 reaches 8.064 TFLOPS in both precisions. That places the AMD part at roughly 10.1 times the raw compute throughput of the NVIDIA part, a direct ratio derived from the recorded figures. In texture rate, the MI308X posts 2,553.6 GTexel/s against the Jetson T5000's 126.0 GTexel/s, a factor of about 20.3. Pixel rate tells a different story: the MI308X records 0 MPixel/s, while the Jetson T5000 achieves 50.40 GPixel/s. The AMD accelerator has no ROPs, which explains the zero pixel rate, while the NVIDIA part has 32 ROPs and produces a measurable pixel throughput.

Memory bandwidth separates the two decisively. The MI308X uses HBM3 across an 8192-bit bus to reach 5.32 TB/s, whereas the Jetson T5000 relies on LPDDR5X across a 256-bit bus for 273.2 GB/s. The AMD part exceeds the NVIDIA part's bandwidth by a factor of roughly 19.5. Memory capacity also favors AMD: 192 GB versus 128 GB. Clock behavior differs as well. The Jetson T5000 has a base clock of 1386 MHz and a boost of 1575 MHz, while the MI308X runs at 1000 MHz base and boosts to 2100 MHz. The higher boost clock on the AMD part contributes to its compute advantage, though the NVIDIA part maintains a higher base clock.

FAQ

Q: Which part has higher FP32 compute?

A: The AMD Instinct MI308X records 81.72 TFLOPS of FP32 performance, which is about 10.1 times the 8.064 TFLOPS posted by the NVIDIA Jetson T5000.

Q: How do the memory subsystems compare?

A: The MI308X uses 192 GB of HBM3 on a 8192-bit bus with 5.32 TB/s bandwidth. The Jetson T5000 uses 128 GB of LPDDR5X on a 256-bit bus with 273.2 GB/s bandwidth. The AMD part's bandwidth is roughly 19.5 times higher.

Q: What is the pixel fillrate capability of each?

A: The MI308X records 0 MPixel/s because it has no ROPs. The Jetson T5000, with 32 ROPs, achieves 50.40 GPixel/s.

Q: What process nodes are used?

A: Both parts are fabricated by TSMC on a 5 nm process node.

Q: What are the power requirements?

A: The MI308X has a TDP of 750 W with a suggested PSU of 1150 W. The Jetson T5000 has a TDP of 120 W with a suggested PSU of 300 W.

Q: What is the release timeline?

A: The MI308X was released on 2023-12-05. The Jetson T5000 was released later, on 2025-08-26, and is currently marked as Active in production status.

Architecture Differences

The two accelerators derive from different architectural lineages. The AMD Instinct MI308X uses the Aqua Vanjaram chip built on CDNA 3.0 architecture, part of the Instinct (MIx) generation. The NVIDIA Jetson T5000 uses the GB10B chip built on Blackwell architecture, part of the Server Blackwell (Bxx) generation. Both are fabricated by TSMC on a 5 nm process, but the physical implementations diverge sharply. The MI308X die measures 1017 mm² with 153,000 million transistors, yielding a transistor density of 150.4M per mm². The Jetson T5000 die measures 391 mm² with transistor count listed as unknown.

Compute resources differ in scale and composition. The MI308X carries 19,456 shading units and 1,216 TMUs, with no ROPs, no RT cores, and no tensor cores recorded. The Jetson T5000 carries 2,560 shading units, 80 TMUs, 32 ROPs, 20 RT cores, and 96 tensor cores. The AMD part's lack of RT and tensor cores reflects its compute-focused CDNA design, while the NVIDIA part includes dedicated ray tracing and tensor hardware despite its lower overall throughput.

Memory architecture reflects different design priorities. The MI308X uses HBM3 stacked memory with an 8192-bit interface, while the Jetson T5000 uses LPDDR5X with a 256-bit interface. The AMD part's memory clock is 1300 MHz (5.2 Gbps effective), while the NVIDIA part runs at 1067 MHz (8.5 Gbps effective). The higher effective data rate on the NVIDIA memory cannot compensate for the vastly wider bus on the AMD side.

Physical form factors also diverge. The MI308X is an OAM Module with no power connectors and no display outputs. The Jetson T5000 is an IGP (integrated graphics processor) with no power connectors and no display outputs, measuring 87 mm by 100 mm by 15 mm. Neither part exposes DirectX, OpenGL, or Vulkan APIs, as both are compute-oriented accelerators rather than graphics cards.

Specification Differences

The two parts differ across nearly every recorded specification field. Clock speeds: the MI308X runs at 1000 MHz base and 2100 MHz boost, while the Jetson T5000 runs at 1386 MHz base and 1575 MHz boost. Memory: the MI308X has 192 GB of HBM3 with a 8192-bit bus and 5.32 TB/s bandwidth, while the Jetson T5000 has 128 GB of LPDDR5X with a 256-bit bus and 273.2 GB/s bandwidth. Memory clocks: 1300 MHz (5.2 Gbps effective) for AMD versus 1067 MHz (8.5 Gbps effective) for NVIDIA.

Compute units: the MI308X has 19,456 shading units, 1,216 TMUs, and 0 ROPs, while the Jetson T5000 has 2,560 shading units, 80 TMUs, and 32 ROPs. The NVIDIA part also has 20 RT cores and 96 tensor cores, neither of which is present on the AMD part. Throughput rates: the MI308X posts 2,553.6 GTexel/s and 0 MPixel/s, while the Jetson T5000 posts 126.0 GTexel/s and 50.40 GPixel/s. FP32 and FP16 both stand at 81.72 TFLOPS for AMD and 8.064 TFLOPS for NVIDIA, with both listed as 1:1 ratios.

Power and physical specs: TDP is 750 W for the MI308X versus 120 W for the Jetson T5000. Suggested PSU is 1150 W versus 300 W. The MI308X uses an OAM Module slot width, while the Jetson T5000 uses IGP. Bus interface: PCIe 5.0 x16 for AMD versus PCIe 5.0 x8 for NVIDIA. The Jetson T5000 has recorded dimensions of 87 mm length, 100 mm height, and 15 mm width, while the MI308X dimensions are not listed.

Die and production details: the MI308X die is 1017 mm² with 153,000 million transistors and a density of 150.4M per mm². The Jetson T5000 die is 391 mm² with unknown transistor count and no density figure. Release dates differ by over a year and a half. The MI308X successor field is null, while the Jetson T5000 lists its predecessor as Server Hopper and its successor as Server Rubin. The Jetson T5000 carries a launch MSRP of 2,999 USD, while the MI308X has no launch MSRP recorded.

The Verdict

The data describes two accelerators with fundamentally different roles. The AMD Instinct MI308X is a high-throughput compute engine built for massive parallel workloads. Its 81.72 TFLOPS FP32 figure, 5.32 TB/s memory bandwidth, and 192 GB capacity place it in a performance class far above the Jetson T5000. The NVIDIA Jetson T5000, by contrast, operates at lower absolute performance but includes features the AMD part lacks entirely: 20 RT cores, 96 tensor cores, and 32 ROPs.

The MI308X's 10.1 times higher FP32 throughput and 19.5 times higher memory bandwidth indicate it is intended for workloads that saturate compute and memory resources, such as large-scale inference or scientific simulation. The Jetson T5000's smaller footprint, 120 W TDP, and integrated form factor suggest deployment in edge or embedded contexts where power and space constraints dominate. Neither part records any benchmark scores in the database, so performance rankings rely on specification analysis rather than measured results.

The Jetson T5000's active production status and known successor (Server Rubin) indicate an ongoing product cycle, while the MI308X's production status is not recorded. The NVIDIA part's launch MSRP of 2,999 USD provides a reference point, though the database does not offer pricing for the AMD part. The selection between these two depends entirely on workload characteristics: raw throughput versus integrated feature set and power efficiency.

Where Each One Wins

The AMD Instinct MI308X wins in every raw compute and memory metric recorded. It delivers 81.72 TFLOPS of FP32 and FP16, which is 10.1 times the NVIDIA part's 8.064 TFLOPS. Its texture rate of 2,553.6 GTexel/s exceeds the Jetson T5000's 126.0 GTexel/s by a factor of 20.3. Memory bandwidth of 5.32 TB/s versus 273.2 GB/s gives the AMD part a 19.5 times advantage, and its 192 GB capacity exceeds the NVIDIA part's 128 GB by 50%. The MI308X also has a wider bus interface (PCIe 5.0 x16 versus x8) and a higher boost clock (2100 MHz versus 1575 MHz). For workloads that demand maximum throughput per operation, the MI308X is the clear choice from the recorded specifications.

The NVIDIA Jetson T5000 wins in areas the AMD part cannot address. It has 32 ROPs and delivers 50.40 GPixel/s, while the MI308X records 0 MPixel/s with no ROPs. It includes 20 RT cores and 96 tensor cores, features entirely absent from the MI308X. Its base clock of 1386 MHz is higher than the AMD part's 1000 MHz, which may benefit latency-sensitive tasks. The Jetson T5000's 120 W TDP is substantially lower than the MI308X's 750 W, and its suggested PSU of 300 W versus 1150 W reflects a much smaller power envelope. Its 87 mm by 100 mm by 15 mm dimensions and IGP form factor make it suitable for compact installations. The Jetson T5000 also has a recorded release date of 2025-08-26 and active production status, whereas the MI308X's production status is not recorded.

The database shows no head-to-head benchmark wins for either part. The MI308X targets high-performance compute environments where its massive memory bandwidth and shading throughput can be fully utilized. The Jetson T5000 targets embedded or server-edge scenarios where its integrated tensor and RT cores, pixel output, and low power draw provide capabilities the MI308X does not offer. The choice between them follows from which metric matters more: the AMD part for sheer compute density, the NVIDIA part for feature breadth and efficiency.