AMD Instinct MI308X vs NVIDIA GeForce RTX 5090 D Comparison

The Verdict

The AMD Instinct MI308X and NVIDIA GeForce RTX 5090 D serve fundamentally different purposes, and the data confirms that neither is a substitute for the other. The MI308X is an accelerator with no display outputs, no graphics API support, and a 192 GB HBM3 memory pool, making it a compute-oriented product aimed at memory-capacity-bound workloads. The RTX 5090 D is a consumer graphics card with full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support, plus display outputs and a 32 GB GDDR7 frame buffer. The recorded benchmark data only covers the RTX 5090 D, which posts an average score of 77712 and sits at the 92nd percentile among all GPUs. The MI308X has no recorded benchmarks and a 50th percentile ranking, so direct performance comparisons rest entirely on architectural and specification differences rather than measured scores. From the data, the RTX 5090 D is the choice for any workload requiring graphics output, real-time rendering, or API-level compatibility. The MI308X is the choice for deployments where 192 GB of memory and 5.32 TB/s of bandwidth matter more than rasterization or ray tracing.

Architecture Differences

The MI308X uses AMD's CDNA 3.0 architecture on the Aqua Vanjaram chip, built on a TSMC 5 nm process. It integrates 153,000 million transistors on a 1017 mm² die, yielding a transistor density of 150.4M per mm². The RTX 5090 D uses NVIDIA's Blackwell 2.0 architecture on the GB202 chip, also on TSMC 5 nm, but with 92,200 million transistors on a 750 mm² die, for a density of 122.9M per mm². The MI308X packs roughly 66% more transistors into a 36% larger die, which reflects its focus on memory controllers and compute units rather than graphics fixed-function hardware.

The MI308X has 19,456 shading units and 1,216 texture mapping units, but zero ROPs and no dedicated ray tracing or tensor core counts listed. Its pixel rate is recorded as 0 MPixel/s, and its texture rate is 2,553.6 GTexel/s. The RTX 5090 D has 21,760 shading units, 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores. Its pixel rate is 423.6 GPixel/s and texture rate is 1,636.8 GTexel/s. The MI308X posts higher raw texture throughput but cannot produce any pixels, confirming it is not a rendering device. The RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the MI308X reports N/A for all graphics APIs.

Clock speeds differ substantially. The MI308X runs at a 1000 MHz base and 2100 MHz boost, with memory at 1300 MHz (5.2 Gbps effective). The RTX 5090 D runs at 2017 MHz base and 2407 MHz boost, with memory at 1750 MHz (28 Gbps effective). Despite the lower clocks, the MI308X achieves 5.32 TB/s bandwidth through an 8192-bit bus with HBM3, versus 1.79 TB/s over a 512-bit bus with GDDR7 on the RTX 5090 D. The MI308X memory bus is 16 times wider, and its bandwidth is nearly 3 times higher. The MI308X has no power connectors because it is an OAM module, while the RTX 5090 D is dual-slot with a single 16-pin connector and a 950 W suggested PSU.

Where Each One Wins

The RTX 5090 D wins in every scenario that involves rendering. Its 176 ROPs and 423.6 GPixel/s pixel rate enable actual rasterization, which the MI308X cannot perform at all. The RTX 5090 D also has 170 RT cores for real-time ray tracing, a capability absent from the MI308X's specification list. Its 104.8 TFLOPS FP32 and FP16 (1:1) throughput exceeds the MI308X's 81.72 TFLOPS in both precisions. For graphics API compatibility, the RTX 5090 D supports the full modern stack, while the MI308X reports N/A for DirectX, OpenGL, and Vulkan. The RTX 5090 D also has display outputs (1x HDMI 2.1b and 3x DisplayPort 2.1b), making it usable as a standard GPU.

The MI308X wins in memory capacity and bandwidth. Its 192 GB of HBM3 dwarfs the RTX 5090 D's 32 GB of GDDR7. Its 8192-bit memory bus and 5.32 TB/s bandwidth are unmatched by the RTX 5090 D's 512-bit bus and 1.79 TB/s. The MI308X also has a higher texture rate at 2,553.6 GTexel/s versus 1,636.8 GTexel/s, though this matters only for compute-style texture operations, not rendering. The MI308X carries a 750 W TDP versus 575 W for the RTX 5090 D, indicating higher power draw but also a larger memory subsystem to feed.

In the recorded benchmark data, the RTX 5090 D scores 14326 in 3DMark Steel Nomad DX12, 310674 in Geekbench OpenCL, 376915 in Geekbench Vulkan, and 44065 in Passmark G3D. Its Passmark GPU Compute score is 28396. These are the only measured performance figures available in the database, and the MI308X has none, so the RTX 5090 D wins every recorded benchmark by default.

FAQ

Q: Which card has more memory?

A: The AMD Instinct MI308X has 192 GB of HBM3, while the NVIDIA GeForce RTX 5090 D has 32 GB of GDDR7.

Q: Can the MI308X output video?

A: No. The MI308X has no display outputs and reports N/A for DirectX, OpenGL, and Vulkan APIs. The RTX 5090 D has 1x HDMI 2.1b and 3x DisplayPort 2.1b outputs and supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4.

Q: Which GPU has higher FP32 compute?

A: The RTX 5090 D has 104.8 TFLOPS FP32, while the MI308X has 81.72 TFLOPS FP32. Both also deliver FP16 at a 1:1 ratio with the same TFLOPS values.

Q: What is the memory bandwidth difference?

A: The MI308X achieves 5.32 TB/s over an 8192-bit bus, while the RTX 5090 D achieves 1.79 TB/s over a 512-bit bus.

Q: Which card has more shading units?

A: The RTX 5090 D has 21,760 shading units, while the MI308X has 19,456.

Q: What are the power requirements?

A: The MI308X has a 750 W TDP and a 1150 W suggested PSU, with no power connectors because it is an OAM module. The RTX 5090 D has a 575 W TDP, a 950 W suggested PSU, and uses a single 16-pin connector.

Head-to-Head Benchmarks

The database records no direct head-to-head benchmark matches between these two products. However, the RTX 5090 D has a full set of measured scores, while the MI308X has none. The RTX 5090 D achieves an average benchmark score of 77712, placing it at the 92nd percentile of all GPUs. Its nearest rivals in the database are the AMD Radeon RX 6650M XT with an average score of 76904 and a delta of 1.1% (meaning the RTX 5090 D is 1.1% ahead), the AMD Radeon RX 6850M XT at 78940 with a delta of -1.6% (1.6% behind), the NVIDIA Tesla P100 PCIe 12 GB at 79396 with a delta of -2.1%, and the NVIDIA Tesla P100 PCIe 16 GB at 79605 with a delta of -2.4%. The RTX 5090 D leads the 6650M XT by a narrow margin but trails the 6850M XT and both Tesla P100 variants by small percentages.

In specific tests, the RTX 5090 D scores 14326 in 3DMark Steel Nomad DX12, which measures modern DirectX 12 rendering performance. Its Geekbench OpenCL score of 310674 and Vulkan score of 376915 indicate strong compute and cross-API performance. The Passmark suite shows a G3D score of 44065, a GPU Compute score of 28396, and individual DirectX tests: 231 for DX10, 371 for DX11, 219 for DX12, and 434 for DX9. The G2D score is 1487. These numbers confirm that the RTX 5090 D is a capable rendering and compute device across multiple generations of DirectX.

The MI308X has no benchmark entries, no average score, and no rival comparisons. Its percentile ranking of 50 is a placeholder rather than a measured result. The absence of data means the MI308X cannot be placed in the same performance hierarchy. The only quantitative comparison available is architectural: the MI308X has higher memory capacity, bandwidth, and texture rate, while the RTX 5090 D has higher pixel rate, FP32 throughput, and clock speeds.

Specification Differences

The two cards differ across nearly every measurable specification. The MI308X uses a 5 nm TSMC process with 153,000 million transistors on a 1017 mm² die, while the RTX 5090 D uses the same 5 nm TSMC process but with 92,200 million transistors on a 750 mm² die. Transistor density is 150.4M per mm² for the MI308X and 122.9M per mm² for the RTX 5090 D.

Clock speeds: the MI308X runs at 1000 MHz base and 2100 MHz boost, with memory at 1300 MHz (5.2 Gbps effective). The RTX 5090 D runs at 2017 MHz base and 2407 MHz boost, with memory at 1750 MHz (28 Gbps effective). The RTX 5090 D has a 1017 MHz higher base clock and a 307 MHz higher boost clock.

Memory: the MI308X has 192 GB of HBM3 on an 8192-bit bus with 5.32 TB/s bandwidth. The RTX 5090 D has 32 GB of GDDR7 on a 512-bit bus with 1.79 TB/s bandwidth. The MI308X has 6 times the memory capacity and 2.97 times the bandwidth.

Compute units: the MI308X has 19,456 shading units and 1,216 TMUs, with 0 ROPs. The RTX 5090 D has 21,760 shading units, 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores. The MI308X has 2304 fewer shading units but 536 more TMUs. Pixel rate is 0 MPixel/s for the MI308X versus 423.6 GPixel/s for the RTX 5090 D. Texture rate is 2,553.6 GTexel/s for the MI308X versus 1,636.8 GTexel/s for the RTX 5090 D.

FP32 and FP16: the MI308X delivers 81.72 TFLOPS in both, while the RTX 5090 D delivers 104.8 TFLOPS in both. The RTX 5090 D is 28% higher in raw floating-point throughput.

Power and physical: the MI308X has a 750 W TDP and a 1150 W suggested PSU, with no power connectors as an OAM module. The RTX 5090 D has a 575 W TDP, a 950 W suggested PSU, and a single 16-pin connector. The RTX 5090 D is a dual-slot card measuring 304 mm by 137 mm by 48 mm. The MI308X lists no dimensions.

Interfaces and outputs: both use PCIe 5.0 x16. The MI308X has no display outputs, while the RTX 5090 D has 1x HDMI 2.1b and 3x DisplayPort 2.1b. The MI308X reports N/A for DirectX, OpenGL, and Vulkan. The RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Release and status: the MI308X released on 2023-12-05 with no production status listed, while the RTX 5090 D released on 2025-01-29 and is marked Active. The MI308X lists its predecessor as Radeon Instinct, and the RTX 5090 D lists GeForce 40 as predecessor and GeForce 60 as successor.