AMD Instinct MI300 vs NVIDIA GeForce RTX 5090 D Comparison

Head-to-Head Benchmarks

The database contains recorded benchmark scores for the NVIDIA GeForce RTX 5090 D, while the AMD Instinct MI300 has no benchmark entries. This makes a direct comparison of measured results impossible, but the RTX 5090 D's data can still be interpreted against its nearest rivals.

The RTX 5090 D posts an average benchmark score of 77,712 across its recorded tests. Its nearest rival, the AMD Radeon RX 6650M XT, scores 76,904, placing the RTX 5090 D 1.1% ahead. The gap widens against the AMD Radeon RX 6850M XT, which scores 78,940, putting the RTX 5090 D 1.6% behind that part. Two NVIDIA Tesla P100 variants also appear in the rival set: the PCIe 12 GB version scores 79,396 (2.1% ahead of the RTX 5090 D), and the PCIe 16 GB version scores 79,605 (2.4% ahead).

The RTX 5090 D's individual benchmark results show a wide spread. In 3DMark Steel Nomad DX12, it records 14,326 points. Geekbench OpenCL returns 310,674, while Geekbench Vulkan reaches 376,915. Passmark results include a G3D score of 44,065, a GPU compute score of 28,396, and a G2D score of 1,487. The legacy DirectX tests are lower: DirectX 9 scores 434, DirectX 11 scores 371, DirectX 10 scores 231, and DirectX 12 scores 219.

The MI300's percentile placement sits at 50 among all GPUs in the database, with an average benchmark score of zero due to the absence of recorded tests. The RTX 5090 D, by contrast, holds the 92nd percentile, indicating that 92% of all GPUs in the database fall below its performance level. With no MI300 scores to compare, the head-to-head benchmark story is defined by absence on one side and a strong, well-documented showing on the other.

Where Each One Wins

For the MI300, every category is a win by default in the sense that it has no measured competition in the database, but that is not a useful analytical outcome. The MI300's design targets are clear from its specifications: 128 GB of HBM3 memory on an 8192-bit bus delivers 5.32 TB/s of bandwidth, a figure that dwarfs the RTX 5090 D's 1.79 TB/s from 32 GB of GDDR7 on a 512-bit bus. Any workload that depends on memory capacity or raw bandwidth, such as large model inference or data-intensive compute, should favor the MI300 on paper.

The RTX 5090 D wins in raw shading throughput. Its 21,760 shading units at a boost clock of 2407 MHz produce 104.8 TFLOPS of FP32, while the MI300's 14,080 shading units at 1700 MHz yield 47.87 TFLOPS. That is a 2.19x advantage in FP32 for the NVIDIA part. The RTX 5090 D also has functional graphics pipelines: 176 ROPs and a pixel rate of 423.6 GPixel/s, while the MI300 reports 0 ROPs and 0 MPixel/s. Texture rate favors NVIDIA as well, 1,636.8 GTexel/s versus 1,496.0 GTexel/s.

The RTX 5090 D is the only one of the two with display outputs, listing 1x HDMI 2.1b and 3x DisplayPort 2.1b. The MI300 has no outputs, confirming its role as an accelerator rather than a graphics card. API support follows the same split: the RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the MI300 lists N/A for all three.

Architecture Differences

The MI300 uses the Aqua Vanjaram chip built on CDNA 3.0 architecture, manufactured on a 5 nm process at TSMC. The RTX 5090 D uses the GB202 chip on Blackwell 2.0 architecture, also 5 nm at TSMC. Both are TSMC 5 nm parts, but the transistor counts diverge sharply.

The MI300 packs 153,000 million transistors on a 1017 mm² die, yielding a transistor density of 150.4 million per mm². The RTX 5090 D has 92,200 million transistors on a 750 mm² die, for a density of 122.9 million per mm². The MI300's die is 35.6% larger in area and holds 66% more transistors, but the RTX 5090 D achieves a higher boost clock: 2407 MHz versus 1700 MHz. Base clocks also differ, 2017 MHz for NVIDIA versus 1000 MHz for AMD.

Memory architecture is fundamentally different. The MI300 uses HBM3 with 128 GB capacity, an 8192-bit bus, and 5.32 TB/s bandwidth. The RTX 5090 D uses GDDR7 with 32 GB capacity, a 512-bit bus, and 1.79 TB/s bandwidth. The MI300's memory clock is listed as 1300 MHz with 5.2 Gbps effective, while the RTX 5090 D runs 1750 MHz with 28 Gbps effective.

The RTX 5090 D includes dedicated hardware that the MI300 lacks in the recorded data: 170 ray tracing cores and 680 tensor cores. The MI300 lists neither. Shading units, TMUs, and ROPs also differ: 14,080 shading units, 880 TMUs, and 0 ROPs for the MI300 versus 21,760 shading units, 680 TMUs, and 176 ROPs for the RTX 5090 D. The MI300 has more TMUs, but the RTX 5090 D has more shading units and the only ROP count.

Power delivery separates them further. The MI300 has a 600 W TDP with 2x 8-pin connectors and a suggested PSU of 1000 W. The RTX 5090 D has a 575 W TDP with a single 16-pin connector and a suggested PSU of 950 W. Physical dimensions: the MI300 measures 267 mm long and 111 mm tall, while the RTX 5090 D is longer at 304 mm, taller at 137 mm, and has a width of 48 mm with a dual-slot designation.

Release timing places the MI300 first, with a release date of 2023-01-03, while the RTX 5090 D arrived on 2025-01-29. The MI300's predecessor is listed as Radeon Instinct, and the RTX 5090 D's predecessor is GeForce 40, with a successor of GeForce 60. The RTX 5090 D carries a launch MSRP of 2,299 USD.

FAQ

Q: Which GPU has more FP32 compute throughput?

A: The RTX 5090 D delivers 104.8 TFLOPS of FP32, while the MI300 delivers 47.87 TFLOPS. The NVIDIA part is 2.19x higher.

Q: Which GPU has more memory bandwidth?

A: The MI300 has 5.32 TB/s of bandwidth from 128 GB of HBM3 on an 8192-bit bus. The RTX 5090 D has 1.79 TB/s from 32 GB of GDDR7 on a 512-bit bus.

Q: Does the MI300 support DirectX or Vulkan?

A: No. The MI300 lists N/A for DirectX, OpenGL, and Vulkan. The RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: Can the MI300 output video to a display?

A: No. The MI300 has no display outputs. The RTX 5090 D has 1x HDMI 2.1b and 3x DisplayPort 2.1b.

Q: How does the RTX 5090 D compare to its nearest rivals in the database?

A: The RTX 5090 D's average benchmark score of 77,712 places it 1.1% above the AMD Radeon RX 6650M XT (76,904), 1.6% below the AMD Radeon RX 6850M XT (78,940), 2.1% below the NVIDIA Tesla P100 PCIe 12 GB (79,396), and 2.4% below the NVIDIA Tesla P100 PCIe 16 GB (79,605).

Q: What are the transistor counts for each chip?

A: The MI300 has 153,000 million transistors on a 1017 mm² die. The RTX 5090 D has 92,200 million transistors on a 750 mm² die.

The Verdict

The data does not support a direct performance verdict because the MI300 has no recorded benchmarks. What the specifications show is a clear division of purpose. The MI300 is a compute accelerator built around massive memory capacity and bandwidth: 128 GB of HBM3 and 5.32 TB/s are its defining traits. It has no display outputs, no graphics API support, and no ROPs. It is not a graphics card in any conventional sense.

The RTX 5090 D is a complete graphics solution. It holds the 92nd percentile in the database, supports DirectX 12 Ultimate, Vulkan 1.4, and OpenGL 4.6, has display outputs, and includes ray tracing and tensor cores. Its FP32 output of 104.8 TFLOPS is more than double the MI300's 47.87 TFLOPS. Its memory subsystem, while smaller at 32 GB, still provides 1.79 TB/s of bandwidth, which is substantial for a consumer part.

For workloads that fit in 32 GB and need graphics, rendering, or general compute, the RTX 5090 D is the only viable option between these two. For workloads that need more than 32 GB of memory or maximum memory bandwidth, the MI300's 128 GB and 5.32 TB/s are unmatched by the RTX 5090 D, but the MI300 cannot display anything and lacks all graphics APIs. The choice depends entirely on whether the workload requires graphics output and API support, in which case the RTX 5090 D wins, or requires extreme memory capacity and bandwidth with no display need, in which case the MI300's specifications are superior. The RTX 5090 D is also the newer part, released on 2025-01-29 versus the MI300's 2023-01-03, and it has an active production status while the MI300's production status is not recorded.

Specification Differences

| Field | AMD Instinct MI300 | NVIDIA GeForce RTX 5090 D |

|---|---|---|

| Chip | Aqua Vanjaram | GB202 |

| Architecture | CDNA 3.0 | Blackwell 2.0 |

| Process Node | 5 nm | 5 nm |

| Transistors | 153,000 million | 92,200 million |

| Die Size | 1017 mm² | 750 mm² |

| Transistor Density | 150.4M / mm² | 122.9M / mm² |

| Base Clock | 1000 MHz | 2017 MHz |

| Boost Clock | 1700 MHz | 2407 MHz |

| Memory Clock | 1300 MHz, 5.2 Gbps effective | 1750 MHz, 28 Gbps effective |

| Memory Size | 128 GB | 32 GB |

| Memory Type | HBM3 | GDDR7 |

| Memory Bus Width | 8192 bit | 512 bit |

| Memory Bandwidth | 5.32 TB/s | 1.79 TB/s |

| Shading Units | 14,080 | 21,760 |

| TMUs | 880 | 680 |

| ROPs | 0 | 176 |

| RT Cores | None listed | 170 |

| Tensor Cores | None listed | 680 |

| Pixel Rate | 0 MPixel/s | 423.6 GPixel/s |

| Texture Rate | 1,496.0 GTexel/s | 1,636.8 GTexel/s |

| FP32 | 47.87 TFLOPS | 104.8 TFLOPS |

| FP16 | 47.87 TFLOPS (1:1) | 104.8 TFLOPS (1:1) |

| TDP | 600 W | 575 W |

| Slot Width | Not listed | Dual-slot |

| Power Connectors | 2x 8-pin | 1x 16-pin |

| Suggested PSU | 1000 W | 950 W |

| Bus Interface | PCIe 5.0 x16 | PCIe 5.0 x16 |

| Display Outputs | No outputs | 1x HDMI 2.1b, 3x DisplayPort 2.1b |

| DirectX | N/A | 12 Ultimate (12_2) |

| OpenGL | N/A | 4.6 |

| Vulkan | N/A | 1.4 |

| Length | 267 mm, 10.5 inches | 304 mm, 12 inches |

| Height | 111 mm, 4.4 inches | 137 mm, 5.4 inches |

| Width | Not listed | 48 mm, 1.9 inches |

| Production Status | Not listed | Active |

| Release Date | 2023-01-03 | 2025-01-29 |

| Predecessor | Radeon Instinct | GeForce 40 |

| Successor | Not listed | GeForce 60 |

| Launch MSRP | Not listed | 2,299 USD |

| Percentile vs All GPUs | 50 | 92 |

| Average Benchmark Score | 0 | 77,712 |