AMD Instinct MI300 vs NVIDIA GeForce RTX 5090 D V2 Comparison

FAQ

Q: What is the process node and die size of each GPU?

A: Both the AMD Instinct MI300 and the NVIDIA GeForce RTX 5090 D V2 are built on a 5 nm process at TSMC. The MI300 has a die size of 1017 mm², while the RTX 5090 D V2 has a die size of 750 mm².

Q: How much memory does each card have, and what type?

A: The AMD Instinct MI300 features 128 GB of HBM3 memory with an 8192-bit bus and 5.32 TB/s bandwidth. The NVIDIA GeForce RTX 5090 D V2 has 24 GB of GDDR7 memory on a 384-bit bus with 1.34 TB/s bandwidth.

Q: What is the peak FP32 performance for each GPU?

A: The AMD Instinct MI300 delivers 47.87 TFLOPS of FP32 compute. The NVIDIA GeForce RTX 5090 D V2 delivers 104.8 TFLOPS of FP32 compute, more than double the MI300.

Q: Which card has a higher boost clock?

A: The NVIDIA GeForce RTX 5090 D V2 boosts to 2407 MHz, compared to the AMD Instinct MI300's boost clock of 1700 MHz. The RTX 5090 D V2 also has a higher base clock at 2017 MHz versus 1000 MHz.

Q: What is the power requirement for each card?

A: The AMD Instinct MI300 has a TDP of 600 W and requires a 1000 W suggested PSU. The NVIDIA GeForce RTX 5090 D V2 has a TDP of 575 W and requires a 950 W suggested PSU.

Q: Does the RTX 5090 D V2 support modern graphics APIs?

A: Yes, the NVIDIA GeForce RTX 5090 D V2 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The AMD Instinct MI300 reports N/A for DirectX, OpenGL, and Vulkan, indicating it is not oriented toward conventional graphics API workloads.

Architecture Differences

The AMD Instinct MI300 uses the CDNA 3.0 architecture with the Aqua Vanjaram chip, designed for compute acceleration rather than graphics output. Its transistor count is 153,000 million, packed into a 1017 mm² die with a density of 150.4M transistors per mm². The MI300 has 14,080 shading units, 880 texture mapping units, and zero ROPs, which aligns with its lack of display outputs. Memory is 128 GB of HBM3 on an 8192-bit bus, producing 5.32 TB/s of bandwidth. The base clock is 1000 MHz and the boost clock is 1700 MHz. It uses two 8-pin power connectors and a 600 W TDP. The card measures 267 mm in length and 111 mm in height, with no display outputs.

The NVIDIA GeForce RTX 5090 D V2 uses the Blackwell 2.0 architecture with the GB202 chip. It has 92,200 million transistors on a 750 mm² die, yielding a density of 122.9M transistors per mm². The RTX 5090 D V2 has 21,760 shading units, 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores. Its memory subsystem is 24 GB of GDDR7 on a 384-bit bus, delivering 1.34 TB/s. The base clock is 2017 MHz and boost is 2407 MHz. This card includes 1x HDMI 2.1b and 3x DisplayPort 2.1b outputs, occupies a dual-slot form factor, and uses a single 16-pin power connector with a 575 W TDP. Its dimensions are 304 mm length, 137 mm height, and 48 mm width.

The architectural split is stark. The MI300 emphasizes memory capacity and bandwidth for compute-dense workloads, with 5.32 TB/s versus 1.34 TB/s, and 128 GB versus 24 GB. The RTX 5090 D V2 emphasizes shading throughput, clock speed, and graphics features, with 104.8 TFLOPS FP32 versus 47.87 TFLOPS, and a 2407 MHz boost versus 1700 MHz. The MI300 has no ROPs and no graphics API support, while the RTX 5090 D V2 is fully equipped for rasterization, ray tracing, and modern API compatibility. The MI300's 1017 mm² die is larger than the RTX 5090 D V2's 750 mm², but the RTX 5090 D V2 achieves higher clocks and FP32 throughput despite fewer transistors.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark results between the AMD Instinct MI300 and the NVIDIA GeForce RTX 5090 D V2. The MI300 has no recorded benchmark scores and a percentile versus all GPUs of 50. The RTX 5090 D V2 has one recorded benchmark score: 16,504 in 3DMark Steel Nomad DX12, placing it at the 59th percentile versus all GPUs.

Without direct comparisons, the nearest rivals for the RTX 5090 D V2 provide context. The NVIDIA T400 scores 16,508, a delta of 0 percent from the RTX 5090 D V2. The AMD Radeon PRO W7500 scores 16,415, a 0.5 percent delta. The NVIDIA RTX PRO 6000 Blackwell scores 16,408, a 0.6 percent delta. The AMD Radeon RX 5700 XT scores 16,361, a 0.9 percent delta. These rivals are all within a single percentage point of the RTX 5090 D V2's Steel Nomad result, indicating that this benchmark places the card in a tightly clustered performance band rather than at the top of the range.

The FP32 compute figures offer a clearer separation. The RTX 5090 D V2 delivers 104.8 TFLOPS, which is 2.19 times the MI300's 47.87 TFLOPS. Texture rate also favors the RTX 5090 D V2 at 1,636.8 GTexel/s versus 1,496.0 GTexel/s, a 9.4 percent advantage. The MI300 counters with memory bandwidth: 5.32 TB/s versus 1.34 TB/s, a 3.97 times advantage. The MI300 also has 128 GB of memory versus 24 GB, a 5.33 times capacity advantage.

The data indicates that the RTX 5090 D V2 wins on compute throughput and texture fill, while the MI300 wins overwhelmingly on memory capacity and bandwidth. Pixel rate is zero for the MI300, confirming it cannot rasterize, while the RTX 5090 D V2 achieves 423.6 GPixel/s.

The Verdict

The recorded data points to two different purpose-built devices rather than direct competitors. The AMD Instinct MI300 is a compute accelerator with no display outputs, no ROPs, and no graphics API support. It delivers 47.87 TFLOPS FP32, 128 GB of HBM3 memory, and 5.32 TB/s bandwidth. The NVIDIA GeForce RTX 5090 D V2 is a graphics card with 104.8 TFLOPS FP32, 24 GB of GDDR7, 176 ROPs, 170 RT cores, and full modern API support.

For any workload requiring rasterization, ray tracing, or graphics output, the RTX 5090 D V2 is the only option between the two, as the MI300 has zero pixel rate and no outputs. For compute workloads that rely on massive memory capacity and bandwidth, the MI300's 128 GB and 5.32 TB/s are unmatched by the RTX 5090 D V2's 24 GB and 1.34 TB/s.

The RTX 5090 D V2 has a recorded benchmark score and a percentile of 59, indicating it sits above the median of all GPUs in the database. The MI300 has no benchmark score and a percentile of 50, which is the median position by default. The FP32 gap is substantial: the RTX 5090 D V2 has 2.19 times the FP32 throughput of the MI300. The MI300 has 3.97 times the memory bandwidth of the RTX 5090 D V2.

The verdict depends on the workload. The RTX 5090 D V2 is the data-backed choice for graphics and high-throughput compute. The MI300 is the data-backed choice for memory-bound compute with very large datasets. Neither card dominates the other across all metrics; they dominate different domains.

Specification Differences

| Specification | AMD Instinct MI300 | NVIDIA GeForce RTX 5090 D V2 |

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

| Architecture | CDNA 3.0 | Blackwell 2.0 |

| Chip | Aqua Vanjaram | GB202 |

| 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 Size | 128 GB | 24 GB |

| Memory Type | HBM3 | GDDR7 |

| Memory Bus Width | 8192 bit | 384 bit |

| Memory Bandwidth | 5.32 TB/s | 1.34 TB/s |

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

| TMUs | 880 | 680 |

| ROPs | 0 | 176 |

| RT Cores | N/A | 170 |

| Tensor Cores | N/A | 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 |

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

| Suggested PSU | 1000 W | 950 W |

| 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 |

| Dimensions (L x H x W) | 267 mm x 111 mm | 304 mm x 137 mm x 48 mm |

| Slot Width | N/A | Dual-slot |

| Release Date | 2023-01-03 | 2025-08-14 |

| Predecessor | Radeon Instinct | GeForce 40 |

| Successor | N/A | GeForce 60 |

| Production Status | N/A | Active |

| Launch MSRP | N/A | 2,299 USD |

Where Each One Wins

The AMD Instinct MI300 wins on memory capacity, memory bandwidth, and transistor count. It has 128 GB versus 24 GB, 5.32 TB/s versus 1.34 TB/s, and 153,000 million transistors versus 92,200 million. The MI300 also has a larger die at 1017 mm² versus 750 mm² and higher transistor density at 150.4M per mm² versus 122.9M per mm². Its 8192-bit memory bus is far wider than the RTX 5090 D V2's 384-bit bus. The MI300's 880 TMUs exceed the RTX 5090 D V2's 680 TMUs, though its texture rate is lower at 1,496.0 GTexel/s versus 1,636.8 GTexel/s. These wins point to workloads where data movement and dataset size dominate: large-scale matrix operations, high-bandwidth streaming, and memory-resident models.

The NVIDIA GeForce RTX 5090 D V2 wins on shading units, clock speeds, FP32 and FP16 throughput, pixel rate, texture rate, ROPs, RT cores, tensor cores, graphics API support, and display outputs. It has 21,760 shading units versus 14,080, a boost clock of 2407 MHz versus 1700 MHz, and a base clock of 2017 MHz versus 1000 MHz. Its FP32 output of 104.8 TFLOPS is 2.19 times the MI300's 47.87 TFLOPS. The RTX 5090 D V2 delivers 423.6 GPixel/s versus 0 MPixel/s, confirming it is the only card capable of rasterization. It also has 170 RT cores and 680 tensor cores, features entirely absent from the MI300. The RTX 5090 D V2 supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, while the MI300 returns N/A for all three. It includes HDMI 2.1b and DisplayPort 2.1b outputs, whereas the MI300 has none.

The RTX 5090 D V2 also wins on physical dimensions and power efficiency per unit of compute. Its TDP is 575 W versus 600 W, and its suggested PSU is 950 W versus 1000 W, despite delivering more than double the FP32 throughput. The RTX 5090 D V2 occupies a dual-slot form factor, while the MI300's slot width is not recorded. The RTX 5090 D V2's production status is active, and its release date is 2025-08-14, later than the MI300's 2023-01-03 release. The RTX 5090 D V2 has a successor listed as GeForce 60, while the MI300 has no successor recorded.

The database records one benchmark for the RTX 5090 D V2, with a score of 16,504 in 3DMark Steel Nomad DX12 and a percentile of 59. The MI300 has no benchmark scores and a percentile of 50. The RTX 5090 D V2's nearest rivals, including the NVIDIA T400, AMD Radeon PRO W7500, NVIDIA RTX PRO 6000 Blackwell, and AMD Radeon RX 5700 XT, all score within 0.9 percent, placing the RTX 5090 D V2 in a competitive mid-range cluster for that test. The MI300 has no nearest rivals recorded.

Use-case separation is clear from the data. The MI300 serves compute-heavy environments where memory capacity and bandwidth are the limiting factors. The RTX 5090 D V2 serves graphics workloads, ray tracing, AI inference with tensor cores, and any application requiring display output. Each card wins where its architecture is designed to operate.