AMD Instinct MI350P vs NVIDIA Jetson T5000 Comparison

The AMD Instinct MI350P and the NVIDIA Jetson T5000 are both server-oriented accelerator modules with no display outputs and no graphics API support, but they are engineered for different segments of the compute market. The MI350P is a dual-slot, 600 W PCIe 5.0 x16 card built on TSMC's 3 nm process, while the Jetson T5000 is a 120 W integrated graphics processor (IGP) on a 5 nm process with a PCIe 5.0 x8 interface. The recorded data shows that the AMD part is the larger, more power-hungry, and substantially faster device, while the NVIDIA part is a compact, low-power alternative with a different memory architecture and a published launch MSRP of 2,999 USD.

Head-to-Head Benchmarks

The benchmark database contains no direct head-to-head test results for these two parts, and both units have an average benchmark score of zero. Their percentile rank against all GPUs is identical at 50, which means the overall performance distribution does not separate them based on any measured workload data. Without recorded benchmark scores, the comparison must rely entirely on the hardware specifications and computed throughput figures.

The largest win for the AMD Instinct MI350P is in raw floating-point throughput. The MI350P delivers 36.04 TFLOPS in FP32 and the same 36.04 TFLOPS in FP16, with a 1:1 ratio. The Jetson T5000 delivers 8.064 TFLOPS in FP32 and 8.064 TFLOPS in FP16, also at a 1:1 ratio. The AMD part is exactly 4.47 times faster in both FP32 and FP16, a difference that translates to roughly 347% more compute throughput. This is the most decisive advantage in the specification sheet.

Texture throughput is another area where the AMD part dominates. The MI350P reaches 1,126.4 GTexel/s, while the Jetson T5000 reaches 126.0 GTexel/s. That is 8.94 times higher texture fill, which indicates a much larger allocation of texture mapping units and a higher clock-driven fill capability. The pixel rate tells the opposite story in terms of raw numbers, but the MI350P reports 0 MPixel/s because it has no ROPs, while the Jetson T5000 reports 50.40 GPixel/s with 32 ROPs. The AMD part has zero ROPs, so it cannot perform conventional rasterized pixel output, while the NVIDIA part has a functional pixel pipeline.

Memory bandwidth is also a clear win for AMD. The MI350P uses 144 GB of HBM3e across an 8192-bit bus, producing 8.19 TB/s of bandwidth. The Jetson T5000 uses 128 GB of LPDDR5X across a 256-bit bus, producing 273.2 GB/s. The AMD bandwidth is 29.98 times higher, a massive gap that matters for memory-bound compute workloads. The NVIDIA part does have a smaller memory capacity advantage on paper in terms of bus efficiency, but the absolute bandwidth difference is enormous.

Clock speeds favor the NVIDIA part. The Jetson T5000 has a base clock of 1386 MHz and a boost clock of 1575 MHz, while the MI350P has a base clock of 1000 MHz and a boost clock of 2200 MHz. The AMD boost clock is 39.7% higher than the NVIDIA boost clock, but the NVIDIA base clock is 38.6% higher than the AMD base clock. The effective memory clock also differs: the MI350P runs at 2000 MHz with 8 Gbps effective, while the Jetson T5000 runs at 1067 MHz with 8.5 Gbps effective. Even with a lower memory clock, the MI350P achieves far higher bandwidth due to its 8192-bit bus.

FAQ

Q: Which processor has more shading units?

A: The AMD Instinct MI350P has 8192 shading units, while the NVIDIA Jetson T5000 has 2560 shading units. The AMD part has 3.2 times more shading units, which aligns with its higher FP32 throughput of 36.04 TFLOPS versus 8.064 TFLOPS.

Q: How does the memory bandwidth compare between the two?

A: The AMD Instinct MI350P provides 8.19 TB/s of bandwidth from 144 GB of HBM3e on an 8192-bit bus. The NVIDIA Jetson T5000 provides 273.2 GB/s from 128 GB of LPDDR5X on a 256-bit bus. The AMD part has roughly 30 times the bandwidth.

Q: What is the power consumption difference?

A: The AMD Instinct MI350P has a TDP of 600 W and requires a 1000 W suggested PSU. The NVIDIA Jetson T5000 has a TDP of 120 W and a 300 W suggested PSU. The NVIDIA part uses 20% of the power of the AMD part.

Q: Do either of these parts support display outputs?

A: No. Both the AMD Instinct MI350P and the NVIDIA Jetson T5000 have no display outputs. Both also report N/A for DirectX, OpenGL, and Vulkan APIs, indicating neither is intended for graphics rendering or display tasks.

Q: What is the physical size difference?

A: The AMD Instinct MI350P is a dual-slot card measuring 267 mm in length, 111 mm in height, and 40 mm in width. The NVIDIA Jetson T5000 is an IGP module measuring 87 mm in length, 100 mm in height, and 15 mm in width. The NVIDIA part is much shorter and thinner.

Q: Which part has tensor cores and ray tracing cores?

A: The NVIDIA Jetson T5000 has 96 tensor cores and 20 ray tracing cores. The AMD Instinct MI350P does not list tensor cores or ray tracing cores in the database, and it has zero ROPs.

Architecture Differences

The two parts come from different architectural families. The AMD Instinct MI350P uses the CDNA 4.0 architecture and belongs to the Instinct (MIx) generation, with the chip design named MI350 128CU. The NVIDIA Jetson T5000 uses the Blackwell architecture and belongs to the Server Blackwell (Bxx) generation, with the chip design named GB10B. These are not direct competitors in the same product tier; they are separate product lines aimed at different compute roles.

The manufacturing process differs. The MI350P is built on a 3 nm process at TSMC, while the Jetson T5000 is built on a 5 nm process at TSMC. The AMD chip has a die size of 1190 mm² and 73,000 million transistors, giving a transistor density of 61.3 million per mm². The Jetson T5000 has a die size of 391 mm² and its transistor count is listed as unknown. The AMD die is roughly 3.04 times larger in area and carries a known transistor count that is far higher.

Memory architecture is a fundamental split. The MI350P uses HBM3e memory with 144 GB capacity, an 8192-bit bus, and 8.19 TB/s bandwidth. The Jetson T5000 uses LPDDR5X memory with 128 GB capacity, a 256-bit bus, and 273.2 GB/s bandwidth. HBM3e is designed for high-bandwidth compute workloads, while LPDDR5X is a lower-power memory type. The bus width difference is extreme: 8192 bits versus 256 bits, a factor of 32.

The compute resources differ in type and count. The MI350P has 8192 shading units, 512 texture mapping units, and zero ROPs. It has no listed tensor cores or ray tracing cores. The Jetson T5000 has 2560 shading units, 80 texture mapping units, 32 ROPs, 20 ray tracing cores, and 96 tensor cores. The NVIDIA part includes dedicated hardware for ray tracing and tensor operations, which the AMD part does not list. The MI350P reports a texture rate of 1,126.4 GTexel/s and a pixel rate of 0 MPixel/s, while the Jetson T5000 reports 126.0 GTexel/s and 50.40 GPixel/s.

Power delivery also reflects the architectural difference. The MI350P has a TDP of 600 W, a dual-slot cooler footprint, and a single 16-pin power connector. The Jetson T5000 has a TDP of 120 W, an IGP slot width, and no power connectors. The suggested PSU ratings are 1000 W for the AMD part and 300 W for the NVIDIA part. The MI350P uses a PCIe 5.0 x16 interface, while the Jetson T5000 uses PCIe 5.0 x8.

Production status differs as well. The Jetson T5000 is marked as Active, with a predecessor in Server Hopper and a successor in Server Rubin. The MI350P has no production status listed, and its predecessor is listed as Radeon Instinct. The release dates are close: the Jetson T5000 was released on 2025-08-26, and the MI350P on 2026-05-06. The MI350P is the later product.

The Verdict

The data shows two distinct devices with almost no overlap in intended use. The AMD Instinct MI350P is a high-power, high-throughput accelerator with 36.04 TFLOPS in FP32 and FP16, 8.19 TB/s of memory bandwidth, and 144 GB of HBM3e. Its zero ROP count and lack of display outputs confirm it is not meant for rasterized graphics. The NVIDIA Jetson T5000 is a low-power IGP with 8.064 TFLOPS in FP32 and FP16, 273.2 GB/s of bandwidth, 128 GB of LPDDR5X, and a 120 W TDP. It includes tensor cores and ray tracing cores, making it a more versatile compute module within a much smaller power envelope.

For workloads that demand maximum FP32 or FP16 throughput, the MI350P is the clear choice. It delivers 4.47 times more compute than the Jetson T5000 and 29.98 times more memory bandwidth. For deployments where power draw, physical size, and integration ease matter more than raw throughput, the Jetson T5000 is the practical option. The MI350P requires a 1000 W PSU and a dual-slot PCIe slot, while the Jetson T5000 runs with a 300 W PSU and fits in an IGP form factor.

The absence of benchmark scores means the database cannot confirm real-world performance beyond the specification-derived values. The percentile rank of 50 for both parts does not indicate superiority for either. The recorded data indicates that the MI350P dominates in raw compute and bandwidth, while the Jetson T5000 offers a lower power draw, a smaller footprint, and dedicated tensor and ray tracing hardware. The choice depends on whether the workload prioritizes peak throughput or power efficiency and physical integration.

Specification Differences

The two parts differ in nearly every major specification category. The MI350P uses a 3 nm process, while the Jetson T5000 uses a 5 nm process. The MI350P die is 1190 mm² with 73,000 million transistors and a density of 61.3 million per mm²; the Jetson T5000 die is 391 mm² with an unknown transistor count and no density figure. The MI350P has a base clock of 1000 MHz and a boost clock of 2200 MHz; the Jetson T5000 has a base clock of 1386 MHz and a boost clock of 1575 MHz. The MI350P memory clock is 2000 MHz with 8 Gbps effective, while the Jetson T5000 memory clock is 1067 MHz with 8.5 Gbps effective.

Memory capacity is 144 GB for the MI350P and 128 GB for the Jetson T5000. Memory type is HBM3e versus LPDDR5X. Bus width is 8192 bit versus 256 bit. Bandwidth is 8.19 TB/s versus 273.2 GB/s. Shading units are 8192 versus 2560. TMUs are 512 versus 80. ROPs are 0 versus 32. The Jetson T5000 has 20 ray tracing cores and 96 tensor cores, while the MI350P has none listed. Pixel rate is 0 MPixel/s versus 50.40 GPixel/s. Texture rate is 1,126.4 GTexel/s versus 126.0 GTexel/s. FP32 is 36.04 TFLOPS versus 8.064 TFLOPS. FP16 is also 36.04 TFLOPS versus 8.064 TFLOPS, both at 1:1.

TDP is 600 W versus 120 W. Slot width is dual-slot versus IGP. Power connectors are one 16-pin versus none. Suggested PSU is 1000 W versus 300 W. Bus interface is PCIe 5.0 x16 versus PCIe 5.0 x8. Physical dimensions are 267 mm by 111 mm by 40 mm versus 87 mm by 100 mm by 15 mm. Production status is not listed for the MI350P, while the Jetson T5000 is Active. Release dates are 2026-05-06 versus 2025-08-26. The Jetson T5000 has a launch MSRP of 2,999 USD; the MI350P has no launch MSRP listed.

Where Each One Wins

The AMD Instinct MI350P wins in raw compute throughput. It has 4.47 times the FP32 and FP16 performance of the Jetson T5000. It also wins decisively in memory bandwidth, with 8.19 TB/s versus 273.2 GB/s. Texture rate is 8.94 times higher. Shading unit count is 3.2 times higher. The larger 144 GB HBM3e pool and the 8192-bit bus make it the better choice for large, bandwidth-hungry compute tasks. The 600 W TDP and 1000 W PSU requirement are the cost of that performance, but the data shows the MI350P is the faster accelerator in every compute-focused metric.

The NVIDIA Jetson T5000 wins in power efficiency and physical integration. It uses 120 W versus 600 W, and its suggested PSU is 300 W versus 1000 W. It is an IGP with no power connectors, measuring 87 mm by 100 mm by 15 mm, versus the MI350P's dual-slot 267 mm card. The Jetson T5000 also has features the MI350P lacks: 96 tensor cores, 20 ray tracing cores, and 32 ROPs. It delivers a functional pixel rate of 50.40 GPixel/s, while the MI350P reports 0 MPixel/s. For any workload that needs tensor operations, ray tracing, or conventional pixel processing, the Jetson T5000 is the only one of the two that supports it.

The MI350P wins for peak FP16 and FP32 workloads where maximum throughput and memory bandwidth are essential. The Jetson T5000 wins for compact, low-power deployments that still need tensor acceleration and a full pixel pipeline. The Jetson T5000 also has an active production status and a successor listed in Server Rubin, while the MI350P has no production status. The data does not show a single winner across all categories, but it does show a clear split: the MI350P is the high-end compute part, and the Jetson T5000 is the efficient, feature-rich module.