AMD Radeon Instinct MI300 vs NVIDIA GeForce RTX 5090 SE Comparison

Head-to-Head Benchmarks

The database contains no recorded benchmark scores for either the AMD Radeon Instinct MI300 or the NVIDIA GeForce RTX 5090 SE. Both cards have an average benchmark score of 0 and hold the 50th percentile against all GPUs in the database, indicating that no direct performance measurements have been logged for these two accelerators. Without recorded data, no head-to-head wins can be assigned; the win count for each product is 0. The absence of scores means any comparison between these two must rely on architectural specifications rather than measured application performance.

The lack of benchmarks is notable because both products target entirely different workloads. The AMD Radeon Instinct MI300 is an accelerator with no display outputs, while the NVIDIA GeForce RTX 5090 SE includes standard display connectivity. The data shows that neither card has been subjected to the database's standard test suite, so any claims about relative speed in gaming, rendering, or compute tasks cannot be substantiated from the recorded information. What can be compared are the physical and architectural characteristics that determine theoretical performance ceilings.

Architecture Differences

The AMD Radeon Instinct MI300 uses the CDNA 3.0 architecture on the Aqua Vanjaram chip, manufactured by TSMC on a 5 nm process. The NVIDIA GeForce RTX 5090 SE uses the Blackwell 2.0 architecture on the GB202 chip, also manufactured by TSMC on a 5 nm process. Both are built on the same process node, but the transistor counts differ substantially: the MI300 packs 153,000 million transistors on a 1017 mm² die, while the RTX 5090 SE contains 92,200 million transistors on a 750 mm² die. The MI300 achieves a transistor density of 150.4 million per mm², compared to 122.9 million per mm² for the RTX 5090 SE.

Clock behavior differs significantly between the two. The MI300 has a base clock of 1000 MHz and a boost clock of 1700 MHz. The RTX 5090 SE operates at a base clock of 1740 MHz and boosts to 2377 MHz. Despite the MI300's lower clocks, its memory subsystem is vastly larger: 128 GB of HBM3 on an 8192-bit bus delivering 6.55 TB/s of bandwidth, versus 24 GB of GDDR7 on a 384-bit bus delivering 1.34 TB/s. The MI300's memory bandwidth is nearly five times that of the RTX 5090 SE, and its capacity is over five times greater.

Compute resources differ in configuration. Both cards have 14,080 shading units, but the MI300 has 880 texture mapping units and no ROPs, while the RTX 5090 SE has 440 TMUs and 160 ROPs. The MI300 reports a pixel rate of 0 MPixel/s, reflecting its lack of any rasterization output stage. The RTX 5090 SE delivers 380.3 GPixel/s. Texture rate favors the MI300 at 1,496.0 GTexel/s versus 1,045.9 GTexel/s for the RTX 5090 SE. The RTX 5090 SE includes 110 ray tracing cores and 440 tensor cores; the MI300 lists no ray tracing or tensor core counts.

Floating-point throughput shows divergent design priorities. The MI300 delivers 47.87 TFLOPS FP32 and 383.0 TFLOPS FP16 using an 8:1 ratio, indicating heavy investment in reduced-precision compute. The RTX 5090 SE provides 66.94 TFLOPS FP32 and 66.94 TFLOPS FP16 at a 1:1 ratio, showing balanced throughput across precision levels. The RTX 5090 SE is 40% ahead in FP32, while the MI300 is 472% ahead in FP16, confirming the MI300's orientation toward AI and scientific workloads that favor half-precision arithmetic.

Power requirements also differ. The MI300 has a TDP of 600 W with dual 8-pin power connectors and a suggested PSU of 1000 W. The RTX 5090 SE has a TDP of 500 W with a single 16-pin connector and a suggested PSU of 900 W. Physical dimensions are identical in length and height (267 mm and 111 mm), but the RTX 5090 SE has a defined width of 40 mm and is listed as dual-slot, while the MI300's width and slot width are unspecified. The RTX 5090 SE is in active production; the MI300's production status is not recorded.

FAQ

Q: Which card has more memory bandwidth?

A: The AMD Radeon Instinct MI300 delivers 6.55 TB/s from its 128 GB HBM3 memory on an 8192-bit bus. The NVIDIA GeForce RTX 5090 SE provides 1.34 TB/s from 24 GB of GDDR7 on a 384-bit bus. The MI300's bandwidth is approximately 4.9 times higher.

Q: How do the FP32 and FP16 throughputs compare?

A: In FP32, the RTX 5090 SE leads with 66.94 TFLOPS versus 47.87 TFLOPS for the MI300. In FP16, the MI300 dominates with 383.0 TFLOPS versus 66.94 TFLOPS for the RTX 5090 SE. The RTX 5090 SE maintains identical FP16 and FP32 rates (1:1), while the MI300's FP16 is 8 times its FP32 rate.

Q: Do both cards have the same number of shading units?

A: Yes, both have 14,080 shading units. However, the MI300 has 880 TMUs and no ROPs, while the RTX 5090 SE has 440 TMUs and 160 ROPs. The RTX 5090 SE also includes 110 ray tracing cores and 440 tensor cores, which the MI300 does not list.

Q: What are the power requirements for each card?

A: The MI300 has a TDP of 600 W, requires dual 8-pin power connectors, and suggests a 1000 W PSU. The RTX 5090 SE has a TDP of 500 W, uses a single 16-pin connector, and suggests a 900 W PSU.

Q: Which card supports display outputs?

A: The RTX 5090 SE includes 1x HDMI 2.1b and 3x DisplayPort 2.1b outputs. The MI300 has no display outputs, indicating it is designed for compute or server environments without direct display connectivity.

Q: What API support does each card offer?

A: The RTX 5090 SE supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI300 lists no API support in the database, consistent with its lack of a graphics-focused feature set.

Specification Differences

| Specification | AMD Radeon Instinct MI300 | NVIDIA GeForce RTX 5090 SE |

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

| Architecture | CDNA 3.0 | Blackwell 2.0 |

| Chip | Aqua Vanjaram | GB202 |

| 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 | 1740 MHz |

| Boost Clock | 1700 MHz | 2377 MHz |

| Memory Size | 128 GB | 24 GB |

| Memory Type | HBM3 | GDDR7 |

| Memory Bus Width | 8192 bit | 384 bit |

| Memory Bandwidth | 6.55 TB/s | 1.34 TB/s |

| Shading Units | 14080 | 14080 |

| TMUs | 880 | 440 |

| ROPs | 0 | 160 |

| Ray Tracing Cores | Not listed | 110 |

| Tensor Cores | Not listed | 440 |

| Pixel Rate | 0 MPixel/s | 380.3 GPixel/s |

| Texture Rate | 1,496.0 GTexel/s | 1,045.9 GTexel/s |

| FP32 | 47.87 TFLOPS | 66.94 TFLOPS |

| FP16 | 383.0 TFLOPS (8:1) | 66.94 TFLOPS (1:1) |

| TDP | 600 W | 500 W |

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

| Suggested PSU | 1000 W | 900 W |

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

| DirectX | Not listed | 12 Ultimate (12_2) |

| OpenGL | Not listed | 4.6 |

| Vulkan | Not listed | 1.4 |

| Width | Not listed | 40 mm |

| Release Date | 2023-01-03 | 2025-12-31 |

| Production Status | Not listed | Active |

Where Each One Wins

The AMD Radeon Instinct MI300 wins in memory capacity and bandwidth. Its 128 GB of HBM3 with 6.55 TB/s bandwidth is suited for large datasets that cannot fit in the RTX 5090 SE's 24 GB allocation. The MI300's FP16 throughput of 383.0 TFLOPS is more than five times the RTX 5090 SE's 66.94 TFLOPS, indicating an advantage in half-precision compute tasks common in AI training and inference. Its higher texture rate of 1,496.0 GTexel/s also suggests strength in texture-heavy compute workloads.

The NVIDIA GeForce RTX 5090 SE wins in FP32 throughput with 66.94 TFLOPS versus 47.87 TFLOPS, a 40% advantage that benefits conventional single-precision compute. Its 160 ROPs and 380.3 GPixel/s pixel rate enable rasterization, which the MI300 cannot perform at all. The RTX 5090 SE includes ray tracing cores and tensor cores, providing hardware acceleration for graphics features absent from the MI300. Its higher boost clock of 2377 MHz versus 1700 MHz contributes to its FP32 lead. The RTX 5090 SE also consumes less power at 500 W versus 600 W and includes display outputs for direct monitor connectivity.

The MI300's release date of 2023-01-03 places it earlier in the market, while the RTX 5090 SE's release date of 2025-12-31 is newer. The RTX 5090 SE is in active production, while the MI300's production status is not recorded. Both cards share the same bus interface, PCIe 5.0 x16, and identical physical length and height.

The Verdict

The data indicates two accelerators with opposite design philosophies. The AMD Radeon Instinct MI300 prioritizes memory capacity, memory bandwidth, and reduced-precision compute, making it appropriate for large-scale AI and scientific workloads where FP16 throughput and massive datasets matter more than rasterization or single-precision speed. The absence of display outputs and ROPs confirms this is a compute-only device.

The NVIDIA GeForce RTX 5090 SE balances graphics and compute capabilities. Its 66.94 TFLOPS FP32 output, 380.3 GPixel/s pixel rate, 110 ray tracing cores, and 440 tensor cores, combined with standard display outputs, position it as a general-purpose GPU for gaming, content creation, and mixed workloads. Its 24 GB GDDR7 memory and 1.34 TB/s bandwidth represent a smaller memory footprint but still substantial throughput.

For buyers selecting between these two, the decision rests on the target application. The MI300 is the choice for memory-bound and half-precision compute tasks, where its 128 GB capacity and 383.0 TFLOPS FP16 output are unmatched by the RTX 5090 SE. The RTX 5090 SE is the choice for graphics work, single-precision compute, or any environment requiring display output, where its 160 ROPs, 66.94 TFLOPS FP32, and 500 W power draw provide a more conventional and efficient profile. The RTX 5090 SE also carries a launch MSRP of 1,499 USD, while the MI300 has no recorded launch MSRP. Neither card has benchmark scores in the database, so these conclusions derive from architectural specifications rather than measured performance.