AMD Instinct MI308X vs NVIDIA GeForce RTX 5090 Comparison

Head-to-Head Benchmarks

The recorded database contains no head-to-head benchmark entries for the AMD Instinct MI308X and NVIDIA GeForce RTX 5090. The MI308X has no benchmark scores listed, while the RTX 5090 has a full suite of 10 recorded tests. This asymmetry defines the entire comparison. The RTX 5090's average benchmark score is 79,842, placing it in the 92nd percentile of all GPUs in the database. The MI308X holds a 50th percentile ranking with an average score of zero, meaning the database contains no measured performance data for it.

The RTX 5090's nearest rival in the database is the NVIDIA Tesla P100 PCIe 16 GB, which scores 79,605, a delta of 0.3 percent. The Tesla P100 PCIe 12 GB trails by 0.6 percent with a score of 79,396. The AMD Radeon RX 6850M XT sits 1.1 percent behind at 78,940, while the AMD Radeon Pro Vega 64X leads the RTX 5090 by 1.4 percent with a score of 80,959. These deltas are narrow, indicating the RTX 5090 operates within a tightly clustered performance band among its nearest rivals, despite its high percentile placement.

The RTX 5090's individual test results show its strongest output in Geekbench Vulkan at 376,728, followed by OpenCL at 334,370. Passmark G3D produces a score of 39,650, while GPU Compute reaches 26,756. The DirectX-oriented Passmark tests are far lower: DirectX 9 scores 395, DirectX 11 scores 341, DirectX 10 scores 226, and DirectX 12 scores 185. The 2D graphics test records 1,413. Because the MI308X has no benchmark entries, no direct score comparisons can be made between the two cards. Any claim of one card beating the other in a specific test would require data that does not exist in the database.

Where Each One Wins

The RTX 5090 wins in every category where measurements exist. It has 10 recorded benchmarks across DirectX, OpenCL, Vulkan, 2D, 3D, and compute workloads. Its 92nd percentile ranking versus all GPUs confirms that its recorded scores place it near the top of the database's entire population. The MI308X has zero wins, zero recorded benchmarks, and a 50th percentile ranking that reflects its lack of measured performance rather than any demonstrated capability.

The RTX 5090's architecture includes 170 ray tracing cores and 680 tensor cores, both absent from the MI308X's listed specifications. The RTX 5090 supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The MI308X lists no supported graphics APIs, with DirectX, OpenGL, and Vulkan all marked as not applicable. This means the MI308X cannot participate in the standard consumer graphics benchmark suite at all. The database records a pixel rate of 0 MPixel/s for the MI308X, while the RTX 5090 delivers 423.6 GPixel/s. The MI308X also has zero ROPs, compared to 176 on the RTX 5090. These structural differences explain why the MI308X has no benchmark presence: it is not designed for the rasterized graphics workloads that the database's tests measure.

The RTX 5090 also leads in raw shading throughput. Its FP32 output is 104.8 TFLOPS, and its FP16 output is 104.8 TFLOPS at a 1:1 ratio. The MI308X provides 81.72 TFLOPS in both FP32 and FP16. The texture rate favors the MI308X at 2,553.6 GTexel/s versus 1,636.8 GTexel/s for the RTX 5090, but the MI308X's zero pixel rate and zero ROP count mean that texture throughput cannot be delivered to a display pipeline. The RTX 5090 wins in every workload category that the database can actually measure.

Architecture Differences

The two chips target entirely different roles. The MI308X uses the Aqua Vanjaram chip on CDNA 3.0 architecture, built on a 5 nm process at TSMC. It packs 153,000 million transistors on a 1017 mm² die, producing a transistor density of 150.4 million per mm². The RTX 5090 uses the GB202 chip on Blackwell 2.0 architecture, also built on a 5 nm process at TSMC, with 92,200 million transistors on a 750 mm² die and a density of 122.9 million per mm². The MI308X has 66 percent more transistors and a 36 percent larger die, but its architecture prioritizes compute throughput rather than graphics output.

Memory configurations diverge sharply. The MI308X carries 192 GB of HBM3 on an 8192-bit bus, delivering 5.32 TB/s of bandwidth. The RTX 5090 carries 32 GB of GDDR7 on a 512-bit bus, delivering 1.79 TB/s. The MI308X's memory capacity is six times larger, and its bandwidth is nearly three times higher. The memory clock also differs: the MI308X runs at 1300 MHz with 5.2 Gbps effective, while the RTX 5090 runs at 1750 MHz with 28 Gbps effective. The GDDR7 on the RTX 5090 achieves much higher per-pin data rates, but the MI308X's massive HBM3 stack wins on total bandwidth.

The compute unit counts tell a mixed story. The MI308X has 19,456 shading units and 1,216 texture mapping units, but zero ROPs. The RTX 5090 has 21,760 shading units, 680 TMUs, and 176 ROPs. The MI308X has 11 percent fewer shaders but 79 percent more TMUs. The RTX 5090 compensates with 170 ray tracing cores and 680 tensor cores, neither of which appears in the MI308X's specification list. The MI308X also has no pixel rate, confirming it cannot rasterize frames. Its API support is marked N/A across DirectX, OpenGL, and Vulkan, while the RTX 5090 supports all three with current versions.

Clock speeds favor the RTX 5090. Its base clock is 2017 MHz and boost is 2407 MHz, versus 1000 MHz base and 2100 MHz boost for the MI308X. The RTX 5090's boost clock is 15 percent higher, and its base clock is more than double. Power draw goes the other direction: the MI308X has a 750 W TDP and requires a 1150 W suggested PSU, while the RTX 5090 has a 575 W TDP and a 950 W suggested PSU. The MI308X uses an OAM Module slot width with no power connectors, while the RTX 5090 is a dual-slot card with a single 16-pin connector. The RTX 5090 includes display outputs (one HDMI 2.1b and three DisplayPort 2.1b), while the MI308X has no outputs. The RTX 5090 measures 304 mm by 137 mm by 40 mm; the MI308X lists no physical dimensions.

Release timing also differs. The MI308X launched on December 5, 2023, as part of the Instinct (MIx) generation, succeeding the Radeon Instinct line. The RTX 5090 launched on January 29, 2025, in the GeForce 50 generation, succeeding the GeForce 40 series and followed by the GeForce 60. The RTX 5090's production status is active; the MI308X lists no production status. The RTX 5090 has a launch MSRP of 1,999 USD, while the MI308X has no listed launch price.

The Verdict

The data supports a clear split: the RTX 5090 is the only card of the two with any measured benchmark results. It scores in the 92nd percentile of all GPUs, with an average benchmark score of 79,842. It delivers 104.8 TFLOPS FP32, supports DirectX 12 Ultimate, Vulkan 1.4, and OpenGL 4.6, includes ray tracing and tensor cores, and outputs to displays through HDMI and DisplayPort connections. It is a finished consumer product with active production status, a dual-slot cooler, and a 16-pin power connector.

The MI308X is a compute accelerator with no graphics pipeline. Its 192 GB of HBM3 memory and 5.32 TB/s bandwidth exceed the RTX 5090's 32 GB and 1.79 TB/s, but it has no ROPs, no pixel rate, no display outputs, and no graphics API support. Its 81.72 TFLOPS FP32 is lower than the RTX 5090's 104.8 TFLOPS. Its texture rate of 2,553.6 GTexel/s is higher, but that throughput has no rasterization backend to feed. Its 750 W TDP exceeds the RTX 5090's 575 W, and it requires a 1150 W PSU versus 950 W.

A user seeking a graphics card for benchmarkable workloads should select the RTX 5090, as it is the only option with recorded scores. A user seeking a high-capacity compute accelerator for memory-bound workloads would look at the MI308X, based on its 192 GB HBM3 pool and 5.32 TB/s bandwidth. The database contains no performance evidence for the MI308X, so its compute advantage cannot be quantified. The RTX 5090's 92nd percentile placement and average score of 79,842 stand as the only verified performance data in this comparison.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The RTX 5090 has an average benchmark score of 79,842. The MI308X has an average benchmark score of 0, with no recorded benchmarks.

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

A: The RTX 5090 scores 0.3 percent above the NVIDIA Tesla P100 PCIe 16 GB, 0.6 percent above the Tesla P100 PCIe 12 GB, and 1.1 percent above the AMD Radeon RX 6850M XT. The AMD Radeon Pro Vega 64X scores 1.4 percent higher.

Q: Which card has more memory?

A: The MI308X has 192 GB of HBM3 memory on an 8192-bit bus with 5.32 TB/s bandwidth. The RTX 5090 has 32 GB of GDDR7 on a 512-bit bus with 1.79 TB/s bandwidth.

Q: Does the MI308X support graphics APIs?

A: No. The MI308X lists DirectX, OpenGL, and Vulkan as N/A. The RTX 5090 supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4.

Q: What is the FP32 performance of each card?

A: The RTX 5090 delivers 104.8 TFLOPS FP32. The MI308X delivers 81.72 TFLOPS FP32.

Q: What are the power requirements?

A: The MI308X has a 750 W TDP with a suggested 1150 W PSU and no power connectors. The RTX 5090 has a 575 W TDP with a suggested 950 W PSU and one 16-pin connector.