AMD Instinct MI300X vs NVIDIA GeForce RTX 5090 D Comparison

The Verdict

The AMD Instinct MI300X and NVIDIA GeForce RTX 5090 D target fundamentally different workloads, and the data confirms that. The MI300X is a compute-first accelerator with a colossal 192 GB of HBM3 memory and a 100th percentile ranking among all GPUs in the database. The RTX 5090 D is a consumer-oriented Blackwell card with a 92nd percentile ranking, built for rendering, ray tracing, and general graphics workloads. If the task is large-scale compute, especially with massive datasets that need to reside on the GPU, the MI300X is the clear choice. If the task involves real-time graphics, display output, or DirectX/Vulkan applications, the RTX 5090 D is the only one of the two that can handle it at all, as the MI300X has no display outputs and no DirectX, OpenGL, or Vulkan support. The RTX 5090 D also holds the edge in raw FP32 throughput (104.8 TFLOPS vs 81.72 TFLOPS), making it the better fit for workloads that leverage shader-heavy computation rather than memory capacity.

Architecture Differences

The two chips come from different architectural families and are built for different purposes. The MI300X uses AMD's CDNA 3.0 architecture on a chip called Aqua Vanjaram, fabricated on a 5 nm process at TSMC. It packs 153,000 million transistors on a 1017 mm² die, yielding a transistor density of 150.4M per mm². This is a massive, compute-optimized design with 19,456 shading units and 1,216 texture mapping units, but critically, it has zero ROPs, no ray tracing cores, and no tensor cores listed. Its pixel rate is 0 MPixel/s, confirming it is not designed for rasterization. The memory subsystem is the standout: 192 GB of HBM3 on a 8192-bit bus, delivering 5.32 TB/s of bandwidth. The MI300X is an OAM module with no power connectors (power is delivered through the module interface), and it has a 750 W TDP with a suggested PSU of 1150 W.

The RTX 5090 D uses NVIDIA's Blackwell 2.0 architecture on the GB202 chip, also built on a 5 nm TSMC process. It has 92,200 million transistors on a 750 mm² die, with a lower transistor density of 122.9M per mm². This card is built for graphics: it has 21,760 shading units, 680 TMUs, 176 ROPs, 170 ray tracing cores, and 680 tensor cores. Its pixel rate is 423.6 GPixel/s, and its texture rate is 1,636.8 GTexel/s. Memory is 32 GB of GDDR7 on a 512-bit bus, providing 1.79 TB/s of bandwidth. The RTX 5090 D is a dual-slot card with a 16-pin power connector, a 575 W TDP, and a suggested PSU of 950 W. It also supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, along with one HDMI 2.1b and three DisplayPort 2.1b outputs. The MI300X lists no API support and no display outputs, making it a pure compute accelerator.

Head-to-Head Benchmarks

The only direct benchmark comparison available in the database is Geekbench OpenCL, where the MI300X scores 317,994 and the RTX 5090 D scores 310,674. The MI300X wins by 2.4%. This is a narrow margin, but it is consistent with the MI300X's positioning as a compute-first part. For context, the MI300X's nearest rivals in the database are all NVIDIA data center parts: the B200 (345,482, MI300X is 8% behind), the H200 NVL (334,891, MI300X is 5% behind), the L40S (295,763, MI300X is 7.5% ahead), and the RTX 6000 Ada Generation (287,237, MI300X is 10.7% ahead). The RTX 5090 D's nearest rivals, by contrast, are mobile and older data center parts: the AMD Radeon RX 6850M XT (78,940, RTX 5090 D is 1.6% behind), the NVIDIA Tesla P100 PCIe 12 GB (79,396, RTX 5090 D is 2.1% behind), the Tesla P100 PCIe 16 GB (79,605, RTX 5090 D is 2.4% behind), and the AMD Radeon RX 6650M XT (76,904, RTX 5090 D is 1.1% ahead). This comparison highlights that the RTX 5090 D's average benchmark score of 77,712 is dragged down by its diverse benchmark suite, which includes DirectX and PassMark tests, while the MI300X has only the single OpenCL score.

The RTX 5090 D has more benchmark entries in the database, covering a wide range of tests. Its best scores include 376,915 in Geekbench Vulkan, 44,065 in PassMark G3D, and 28,396 in PassMark GPU Compute. It also has scores in PassMark DirectX 9, 10, 11, and 12, with the DirectX 9 score (434) being the highest among those. The MI300X has no such graphics benchmarks, reinforcing that it is not intended for those workloads. The wins tally is 1 for the MI300X and 0 for the RTX 5090 D in direct head-to-head tests, but that is solely because the database only has one shared test.

FAQ

Q: Which card wins in Geekbench OpenCL?

A: The AMD Instinct MI300X scores 317,994, which is 2.4% higher than the RTX 5090 D's 310,674. The MI300X wins that specific test.

Q: Can the RTX 5090 D be used for gaming?

A: The data indicates yes, it has full graphics support. It includes DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4, and has display outputs (1x HDMI 2.1b, 3x DisplayPort 2.1b). Its 176 ROPs and 170 ray tracing cores support rasterized and ray-traced rendering.

Q: Can the MI300X output video to a monitor?

A: No. The database lists no display outputs for the MI300X, and its pixel rate is 0 MPixel/s. It is a compute-only OAM module.

Q: How much memory does each card have?

A: The MI300X has 192 GB of HBM3 on a 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.

Q: Which card has higher FP32 compute?

A: The RTX 5090 D has 104.8 TFLOPS FP32, while the MI300X has 81.72 TFLOPS. The RTX 5090 D is ahead by roughly 28% in raw shader throughput.

Q: What are the power requirements?

A: The MI300X has a 750 W TDP and a suggested PSU of 1150 W. The RTX 5090 D has a 575 W TDP and a suggested PSU of 950 W. The MI300X uses an OAM module with no power connectors, while the RTX 5090 D uses a 16-pin connector.

Where Each One Wins

The AMD Instinct MI300X wins in memory capacity and bandwidth by a massive margin. With 192 GB of HBM3 versus 32 GB of GDDR7, it can hold datasets six times larger on the GPU itself, avoiding PCIe transfers. Its 5.32 TB/s bandwidth is nearly three times the RTX 5090 D's 1.79 TB/s, which is critical for large matrix operations, inference batches, and scientific computing. The MI300X also wins in the only direct benchmark comparison, Geekbench OpenCL, by 2.4%. Its 100th percentile ranking means it sits at the top of the database's performance distribution, while the RTX 5090 D sits at the 92nd percentile. For workloads that fit entirely in memory and rely on sustained compute throughput, the MI300X is the stronger part, despite its lower FP32 number. The MI300X also has a higher texture rate (2,553.6 GTexel/s vs 1,636.8 GTexel/s), which suggests it excels at texture-heavy compute tasks even without traditional ROPs.

The NVIDIA GeForce RTX 5090 D wins in raw shader compute, graphics features, and practical system integration. Its 104.8 TFLOPS FP32 is higher than the MI300X's 81.72 TFLOPS, giving it an edge in workloads that are shader-bound rather than memory-bound. It has 21,760 shading units, more than the MI300X's 19,456, and it includes 170 ray tracing cores and 680 tensor cores, which the MI300X lacks entirely. The RTX 5090 D also has a huge lead in pixel rate (423.6 GPixel/s vs 0), making it the only choice for any rasterized output. It supports modern graphics APIs (DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4) and has display outputs, so it can be used in a standard desktop PC. Its dual-slot form factor and 16-pin power connector are conventional, while the MI300X requires an OAM slot with no standard power connectors. The RTX 5090 D is also a production-status card with a successor listed (GeForce 60), whereas the MI300X has no successor listed.

Specification Differences

The two cards differ in nearly every specification category. The MI300X uses AMD's CDNA 3.0 architecture with 153,000 million transistors on a 1017 mm² die, while the RTX 5090 D uses Blackwell 2.0 with 92,200 million transistors on a 750 mm² die. Transistor density is 150.4M per mm² for the MI300X and 122.9M per mm² for the RTX 5090 D. Clock speeds differ significantly: the MI300X has a base clock of 1000 MHz and boost of 2100 MHz, while the RTX 5090 D starts at 2017 MHz and boosts to 2407 MHz. Memory speed also differs: the MI300X runs at 1300 MHz (5.2 Gbps effective) with HBM3, while the RTX 5090 D runs at 1750 MHz (28 Gbps effective) with GDDR7.

Compute resources diverge sharply. The MI300X has 19,456 shading units and 1,216 TMUs, but 0 ROPs. The RTX 5090 D has 21,760 shading units, 680 TMUs, and 176 ROPs. The MI300X has no ray tracing or tensor cores, while the RTX 5090 D has 170 RT cores and 680 tensor cores. Pixel rates are 0 MPixel/s for the MI300X and 423.6 GPixel/s for the RTX 5090 D. Texture rates are 2,553.6 GTexel/s for the MI300X and 1,636.8 GTexel/s for the RTX 5090 D. FP32 and FP16 are both 81.72 TFLOPS for the MI300X and both 104.8 TFLOPS for the RTX 5090 D.

Power and physical specs differ as well. The MI300X has a 750 W TDP with a suggested PSU of 1150 W, while the RTX 5090 D has a 575 W TDP with a suggested PSU of 950 W. The MI300X is an OAM module with no power connectors and no dimensions listed, while the RTX 5090 D is a dual-slot card measuring 304 mm by 137 mm by 48 mm with a 16-pin connector. The MI300X has no display outputs; the RTX 5090 D has 1x HDMI 2.1b and 3x DisplayPort 2.1b. API support is absent for the MI300X, while the RTX 5090 D supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The MI300X has no launch MSRP, while the RTX 5090 D has a launch MSRP of 2,299 USD. Release dates differ by over a year: the MI300X was released on 2023-12-05, and the RTX 5090 D on 2025-01-29. The MI300X's predecessor is Radeon Instinct, while the RTX 5090 D's predecessor is GeForce 40 and its successor is GeForce 60.