AMD Instinct MI308X vs NVIDIA GeForce RTX 5090 D V2 Comparison

Where Each One Wins

The AMD Instinct MI308X and the NVIDIA GeForce RTX 5090 D V2 occupy entirely different corners of the hardware spectrum, and the data makes that split explicit. The MI308X is an Instinct-series accelerator built around the CDNA 3.0 architecture with an Aqua Vanjaram chip, while the RTX 5090 D V2 is a GeForce 50-series consumer graphics card on the Blackwell 2.0 architecture. The benchmark database records zero head-to-head wins for either product, which means there are no direct comparative measurements available. Instead, the split is defined by their physical design, memory subsystem, and API support.

The MI308X wins on raw scale. It carries 192 GB of HBM3 memory across an 8192-bit bus, delivering 5.32 TB/s of bandwidth. The RTX 5090 D V2 counters with 24 GB of GDDR7 on a 384-bit bus, reaching 1.34 TB/s. In memory capacity, the MI308X is 8 times larger, and in bandwidth, it is nearly 4 times higher. The MI308X also has a larger die at 1017 mm² versus 750 mm², and it packs 153,000 million transistors compared to 92,200 million. The texture rate favors AMD as well: 2,553.6 GTexel/s versus 1,636.8 GTexel/s.

The RTX 5090 D V2 wins on rendering features and real-world graphics capability. It has 176 ROPs, 170 RT cores, and 680 tensor cores, whereas the MI308X lists 0 ROPs, no RT cores, and no tensor cores. The pixel rate tells the same story: 423.6 GPixel/s for NVIDIA, 0 MPixel/s for AMD. The RTX 5090 D V2 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the MI308X has no API support listed. The NVIDIA card also boosts higher at 2407 MHz versus 2100 MHz, and it has a higher base clock at 2017 MHz versus 1000 MHz. In FP32 compute, the RTX 5090 D V2 reaches 104.8 TFLOPS, ahead of the MI308X's 81.72 TFLOPS.

The use-case split is therefore straightforward: the MI308X is a memory-capacity and bandwidth-focused accelerator with no display outputs, while the RTX 5090 D V2 is a fully featured graphics card with HDMI 2.1b and DisplayPort 2.1b outputs, designed for rendering workloads that require rasterization, ray tracing, and tensor operations.

FAQ

Q: Which card has more memory bandwidth?

A: The AMD Instinct MI308X has 5.32 TB/s of bandwidth from HBM3 memory on an 8192-bit bus. The NVIDIA GeForce RTX 5090 D V2 has 1.34 TB/s from GDDR7 on a 384-bit bus.

Q: Does the MI308X support ray tracing?

A: No. The database lists no RT cores for the MI308X, while the RTX 5090 D V2 has 170 RT cores.

Q: What is the FP32 compute difference?

A: The RTX 5090 D V2 delivers 104.8 TFLOPS of FP32, which is approximately 28% higher than the MI308X's 81.72 TFLOPS.

Q: Which card has display outputs?

A: Only the RTX 5090 D V2 has display outputs: 1x HDMI 2.1b and 3x DisplayPort 2.1b. The MI308X lists no outputs.

Q: How do their power requirements compare?

A: The MI308X has a 750 W TDP with a suggested PSU of 1150 W. The RTX 5090 D V2 has a 575 W TDP with a suggested PSU of 950 W.

Q: What is the release date difference?

A: The MI308X was released on 2023-12-05, while the RTX 5090 D V2 was released on 2025-08-14, roughly 20 months later.

Head-to-Head Benchmarks

The benchmark database contains no direct head-to-head measurements between the AMD Instinct MI308X and the NVIDIA GeForce RTX 5090 D V2. The wins count is 0 for both products. The MI308X has no benchmark entries at all, and its average benchmark score is 0. The RTX 5090 D V2 has a single recorded benchmark result: 16,504 points in 3dmark_3dmark_steel_nomad_dx12. That score places it at the 59th percentile among all GPUs in the database.

The nearest rivals for the RTX 5090 D V2 provide context. The NVIDIA T400 scores 16,508, a delta of 0 percent. The AMD Radeon PRO W7500 scores 16,415, putting the RTX 5090 D V2 0.5 percent ahead. The NVIDIA RTX PRO 6000 Blackwell scores 16,408, a 0.6 percent gap. The AMD Radeon RX 5700 XT scores 16,361, and the RTX 5090 D V2 leads it by 0.9 percent. These deltas are all within one percentage point, meaning the RTX 5090 D V2's Steel Nomad result is effectively tied with a set of very different cards, from a low-end workstation GPU to a 2019-era gaming card.

The absence of MI308X benchmark data means the comparison is purely architectural. The biggest wins each way are in specification categories, not measured performance. The MI308X holds a 4x bandwidth advantage: 5.32 TB/s versus 1.34 TB/s. The RTX 5090 D V2 holds a 28% FP32 compute advantage: 104.8 TFLOPS versus 81.72 TFLOPS. The MI308X has 8x the memory capacity: 192 GB versus 24 GB. The RTX 5090 D V2 has a 170 RT core count advantage versus none, and 680 tensor cores versus none. The pixel rate is 423.6 GPixel/s for NVIDIA versus 0 MPixel/s for AMD.

Specification Differences

The two products differ in nearly every measurable specification. The MI308X uses a 5 nm process at TSMC, the same node as the RTX 5090 D V2. Transistor counts diverge heavily: 153,000 million for AMD versus 92,200 million for NVIDIA. Die size is 1017 mm² for the MI308X and 750 mm² for the RTX 5090 D V2. Transistor density is 150.4 million per mm² for AMD and 122.9 million per mm² for NVIDIA.

Clock speeds favor NVIDIA. The MI308X has a 1000 MHz base clock and a 2100 MHz boost clock. The RTX 5090 D V2 has a 2017 MHz base clock and a 2407 MHz boost clock, which is 101.7% higher at base and 14.6% higher at boost. Memory clocks also differ: the MI308X runs at 1300 MHz with 5.2 Gbps effective, while the RTX 5090 D V2 runs at 1750 MHz with 28 Gbps effective.

Shading units favor NVIDIA: 21,760 versus 19,456. TMUs favor AMD: 1,216 versus 680. ROPs favor NVIDIA: 176 versus 0. The MI308X has no RT cores or tensor cores listed, while the RTX 5090 D V2 has 170 and 680, respectively. Pixel rate is 423.6 GPixel/s for NVIDIA and 0 MPixel/s for AMD. Texture rate is 2,553.6 GTexel/s for AMD and 1,636.8 GTexel/s for NVIDIA.

Power and physical design differ sharply. The MI308X has a 750 W TDP and uses an OAM Module slot width with no power connectors. The RTX 5090 D V2 has a 575 W TDP, a dual-slot design, and a single 16-pin power connector. Suggested PSU is 1150 W for AMD and 950 W for NVIDIA. The MI308X has no display outputs; the RTX 5090 D V2 has 1x HDMI 2.1b and 3x DisplayPort 2.1b. The NVIDIA card has physical dimensions of 304 mm length, 137 mm height, and 48 mm width, while the MI308X lists no dimensions. The MI308X has no API support; the RTX 5090 D V2 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Production status is active for the RTX 5090 D V2, while the MI308X has no status listed.

Architecture Differences

The MI308X uses the CDNA 3.0 architecture on the Aqua Vanjaram chip, belonging to the Instinct (MIx) generation. The RTX 5090 D V2 uses the Blackwell 2.0 architecture on the GB202 chip, belonging to the GeForce 50 generation. CDNA is AMD's compute-focused architecture, which is reflected in the absence of ROPs, RT cores, tensor cores, and display outputs. Blackwell 2.0 is a graphics-oriented architecture, evidenced by the full rasterization pipeline, ray tracing hardware, and tensor cores.

Memory architecture is a fundamental divide. The MI308X uses HBM3 with an 8192-bit bus, which is why it can reach 5.32 TB/s with 192 GB of capacity. The RTX 5090 D V2 uses GDDR7 on a 384-bit bus, achieving 1.34 TB/s with 24 GB. The HBM3 configuration is designed for massive datasets that need to stay resident on the accelerator, while the GDDR7 configuration is suited to consumer rendering workloads with lower memory pressure.

The transistor density difference is notable: AMD achieves 150.4 million transistors per mm² versus NVIDIA's 122.9 million per mm², despite both using the same 5 nm TSMC process. This suggests the MI308X packs more logic per area, likely due to the memory controllers and compute units required for the 8192-bit bus. The die size difference of 267 mm² favors AMD's larger physical footprint, which accommodates the additional 60,800 million transistors.

The release timeline shows a major generational gap. The MI308X launched on 2023-12-05, while the RTX 5090 D V2 launched on 2025-08-14. The NVIDIA card's predecessor is GeForce 40 and its successor is GeForce 60, while the MI308X's predecessor is Radeon Instinct and its successor is not listed. The MI308X has a 50th percentile rank among all GPUs in the database, while the RTX 5090 D V2 sits at the 59th percentile. The RTX 5090 D V2 has a launch MSRP of 2,299 USD, while the MI308X has no launch MSRP recorded.

The Verdict

The data indicates two different purchasing populations. The AMD Instinct MI308X is for workloads that demand enormous memory capacity and bandwidth. Its 192 GB of HBM3 at 5.32 TB/s dwarfs the RTX 5090 D V2's 24 GB at 1.34 TB/s. The OAM Module form factor, 750 W TDP, and absence of display outputs confirm it is a server or accelerator card, not a desktop graphics solution. The lack of any API support, ROPs, RT cores, or tensor cores means it is not designed for traditional rendering, ray tracing, or AI inference via tensor operations. Its 81.72 TFLOPS of FP32 compute is still substantial, but it is 28% behind the NVIDIA card.

The NVIDIA GeForce RTX 5090 D V2 is for graphics-intensive workloads. It has 176 ROPs, 170 RT cores, 680 tensor cores, and full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support. Its 104.8 TFLOPS of FP32 compute is the highest in this comparison. The 16,504 Steel Nomad score at the 59th percentile, with nearest rivals within 0.9 percent, shows it performs in line with cards like the NVIDIA T400 and AMD Radeon PRO W7500 in that specific DX12 test, though those rivals have drastically different specifications. The RTX 5090 D V2 also has display outputs, making it usable as a consumer GPU, and its dual-slot, 304 mm design fits standard desktop cases.

The verdict from the recorded data is that the MI308X is the choice for memory-bound compute tasks with massive datasets, while the RTX 5090 D V2 is the choice for rendering, ray tracing, and applications that need a full graphics pipeline. The RTX 5090 D V2 has a higher FP32 peak and a measured benchmark score, but the MI308X's memory subsystem is in a different class entirely. There is no overlap in their intended use cases based on the specifications available. The database shows no direct benchmark comparison, so any performance ranking between them remains dependent on the workload, but the architectural evidence is clear: the MI308X is a bandwidth specialist, and the RTX 5090 D V2 is a graphics generalist.