AMD Instinct MI325X vs NVIDIA GeForce RTX 5090 Mobile Comparison

Head-to-Head Benchmarks

The AMD Instinct MI325X and NVIDIA GeForce RTX 5090 Mobile occupy entirely different segments of the GPU landscape, and the recorded data reflects that divergence clearly. The MI325X carries no benchmark entries in the database, resulting in an average benchmark score of zero and a percentile rank of 50 among all GPUs. The RTX 5090 Mobile, conversely, has a full suite of ten benchmark results, an average score of 45,152, and sits at the 84th percentile. This places the mobile NVIDIA part far ahead in terms of measured performance data, though the absence of MI325X results means the comparison relies on architectural and specification contrasts rather than direct head-to-head scores.

The RTX 5090 Mobile's benchmark suite shows its strongest showing in PassMark G3D with a score of 30,034, a figure that drives much of its average. Its OpenCL result of 201,834 and Vulkan result of 198,405 in Geekbench demonstrate substantial compute throughput for a mobile part. The 3DMark Steel Nomad DX12 test yields 5,871, while PassMark GPU Compute reaches 13,401. Lower-level DirectX tests show more modest numbers: DirectX 9 scores 324, DirectX 10 scores 183, DirectX 11 scores 269, and DirectX 12 scores 138. The PassMark G2D result of 1,057 rounds out the suite, indicating 2D performance that is typical for a notebook-class GPU.

Against its nearest rivals in the database, the RTX 5090 Mobile shows a tight competitive cluster. It trails the AMD Radeon Pro 5500 XT by 0.5 percent, with the rival averaging 45,384 points. It leads the NVIDIA GeForce RTX 4070 Ti by 0.8 percent, as that card averages 44,795 points. The Intel Arc A730M sits 1 percent ahead with 45,592 points, while the NVIDIA RTX 5880 Ada Generation holds a 1.8 percent lead with 45,972 points. These delta values indicate that the RTX 5090 Mobile performs within a narrow band of established desktop and mobile parts, with none of the four rivals separating itself by more than two percent.

The MI325X, by contrast, has no comparable benchmark entries, so its percentile of 50 reflects an unmeasured position rather than a competitive result. The data shows that any performance comparison between these two products must be inferred from their specifications, clock rates, and memory subsystems, not from direct benchmark outputs.

Where Each One Wins

The RTX 5090 Mobile wins decisively in any scenario where measured benchmark scores matter, simply because it has them and the MI325X does not. The database records ten separate tests for the NVIDIA part, covering DirectX 9 through 12, OpenCL, Vulkan, G2D, G3D, and GPU compute. Its average score of 45,152 and 84th percentile placement indicate that it outperforms the vast majority of GPUs in the database, with only a small cluster of rivals within two percent. For workloads that rely on established graphics APIs, DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 support give the RTX 5090 Mobile a clear functional advantage, as the MI325X lists N/A for all three APIs.

The MI325X wins in raw compute capacity and memory scale, areas where the RTX 5090 Mobile cannot compete. Its FP32 throughput of 81.72 TFLOPS more than doubles the RTX 5090 Mobile's 31.80 TFLOPS, and its FP16 figure matches at 81.72 TFLOPS (1:1) versus the NVIDIA part's 31.80 TFLOPS (1:1). Texture rate also favors AMD heavily: 2,553.6 GTexel/s versus 496.9 GTexel/s. The memory subsystem is even more lopsided, with 256 GB of HBM3e on an 8192-bit bus delivering 6.14 TB/s, compared to 24 GB of GDDR7 on a 256-bit bus at 896.0 GB/s. For workloads that fit within large memory pools, such as massive model inference or high-resolution scientific datasets, the MI325X provides capacity that the RTX 5090 Mobile cannot approach.

The RTX 5090 Mobile counters with pixel rate, recording 169.7 GPixel/s while the MI325X sits at 0 MPixel/s, reflecting the AMD part's lack of display outputs. The NVIDIA GPU also brings ray tracing hardware with 82 RT cores and 328 tensor cores, features absent from the MI325X data. The MI325X's 19,456 shading units dwarf the RTX 5090 Mobile's 10,496, and its 1,216 TMUs compare to 328, so in compute-heavy shader workloads the AMD part has a theoretical edge. However, the MI325X has 0 ROPs, meaning it cannot rasterize output pixels, reinforcing its role as a compute accelerator rather than a graphics card.

Architecture Differences

The MI325X uses the Aqua Vanjaram chip built on CDNA 3.0 architecture, fabricated by TSMC on a 5 nm process. It packs 153,000 million transistors onto a 1017 mm² die, yielding a transistor density of 150.4 million per square millimeter. The RTX 5090 Mobile uses the GB203 chip with Blackwell 2.0 architecture, also on TSMC's 5 nm node, but with 45,600 million transistors on a 378 mm² die, resulting in a density of 120.6 million per square millimeter. The MI325X's die is nearly three times larger in area and holds more than three times the transistor count, reflecting its accelerator-class design versus the mobile GPU's constraints.

Clock behavior differs markedly. The MI325X runs at a 1000 MHz base and 2100 MHz boost, while the RTX 5090 Mobile operates at 990 MHz base and 1515 MHz boost. The AMD part's higher boost clock, combined with its massive shader array, explains its FP32 and FP16 throughput advantage. Memory clocks also diverge: the MI325X lists 1500 MHz with 6 Gbps effective, while the RTX 5090 Mobile runs at 1750 MHz with 28 Gbps effective. Despite the NVIDIA part's higher memory clock, its narrow 256-bit bus limits total bandwidth to 896.0 GB/s, far below the MI325X's 6.14 TB/s from the 8192-bit HBM3e interface.

Both products use PCIe 5.0 x16 for host connectivity, but their form factors and power profiles could not be more different. The MI325X is an OAM module with no power connectors, a 1000 W TDP, and a suggested PSU of 1400 W. The RTX 5090 Mobile is an IGP with a 95 W TDP and no suggested PSU listed. The MI325X has no display outputs, while the RTX 5090 Mobile's outputs are portable device dependent. The RTX 5090 Mobile supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4; the MI325X lists N/A for all three, confirming its server-oriented nature.

Release timing shows the MI325X launched on 2024-10-09, over five months before the RTX 5090 Mobile's 2025-03-26 date. The MI325X's predecessor is Radeon Instinct, while the RTX 5090 Mobile's predecessor is GeForce 40 Mobile. The production status for the MI325X is not recorded, whereas the RTX 5090 Mobile is marked active. Neither product has a launch MSRP in the database.

FAQ

Q: Which GPU has higher raw compute throughput?

A: The AMD Instinct MI325X delivers 81.72 TFLOPS in both FP32 and FP16 (1:1), compared to the NVIDIA GeForce RTX 5090 Mobile's 31.80 TFLOPS in both formats. This gives the MI325X more than double the compute throughput.

Q: How do the memory configurations compare?

A: The MI325X offers 256 GB of HBM3e on an 8192-bit bus with 6.14 TB/s bandwidth. The RTX 5090 Mobile has 24 GB of GDDR7 on a 256-bit bus with 896.0 GB/s bandwidth. The AMD part provides over ten times the capacity and nearly seven times the bandwidth.

Q: Which GPU supports modern graphics APIs?

A: Only the RTX 5090 Mobile supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI325X lists N/A for all three APIs, reflecting its compute-only design with no display outputs.

Q: What is the power consumption difference?

A: The MI325X has a 1000 W TDP with a suggested PSU of 1400 W, while the RTX 5090 Mobile has a 95 W TDP and no suggested PSU listed. The mobile part consumes a fraction of the power.

Q: Does the RTX 5090 Mobile have ray tracing and tensor cores?

A: Yes, the RTX 5090 Mobile includes 82 RT cores and 328 tensor cores. The MI325X data does not list RT cores or tensor cores, consistent with its CDNA 3.0 architecture that lacks these features.

Q: How does the RTX 5090 Mobile rank among all GPUs?

A: The RTX 5090 Mobile sits at the 84th percentile with an average benchmark score of 45,152. Its closest rivals are within 1.8 percent: the AMD Radeon Pro 5500 XT at 45,384 (-0.5 percent), the NVIDIA GeForce RTX 4070 Ti at 44,795 (+0.8 percent), the Intel Arc A730M at 45,592 (-1 percent), and the NVIDIA RTX 5880 Ada Generation at 45,972 (-1.8 percent).

Specification Differences

| Field | AMD Instinct MI325X | NVIDIA GeForce RTX 5090 Mobile |

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

| Chip | Aqua Vanjaram | GB203 |

| Architecture | CDNA 3.0 | Blackwell 2.0 |

| Generation | Instinct (MIx) | GeForce 50 Mobile |

| Transistors | 153,000 million | 45,600 million |

| Die Size | 1017 mm² | 378 mm² |

| Transistor Density | 150.4M / mm² | 120.6M / mm² |

| Base Clock | 1000 MHz | 990 MHz |

| Boost Clock | 2100 MHz | 1515 MHz |

| Memory Clock | 1500 MHz (6 Gbps effective) | 1750 MHz (28 Gbps effective) |

| Memory Size | 256 GB | 24 GB |

| Memory Type | HBM3e | GDDR7 |

| Memory Bus | 8192 bit | 256 bit |

| Memory Bandwidth | 6.14 TB/s | 896.0 GB/s |

| Shading Units | 19,456 | 10,496 |

| TMUs | 1,216 | 328 |

| ROPs | 0 | 112 |

| RT Cores | N/A | 82 |

| Tensor Cores | N/A | 328 |

| Pixel Rate | 0 MPixel/s | 169.7 GPixel/s |

| Texture Rate | 2,553.6 GTexel/s | 496.9 GTexel/s |

| FP32 | 81.72 TFLOPS | 31.80 TFLOPS |

| FP16 | 81.72 TFLOPS (1:1) | 31.80 TFLOPS (1:1) |

| TDP | 1000 W | 95 W |

| Slot Width | OAM Module | IGP |

| Suggested PSU | 1400 W | N/A |

| Display Outputs | No outputs | Portable Device Dependent |

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

| OpenGL | N/A | 4.6 |

| Vulkan | N/A | 1.4 |

| Production Status | N/A | Active |

| Release Date | 2024-10-09 | 2025-03-26 |

| Predecessor | Radeon Instinct | GeForce 40 Mobile |