AMD Instinct MI325X vs NVIDIA GeForce RTX 5090 D Comparison

The Verdict

The database shows two accelerators built for entirely different workloads. The AMD Instinct MI325X is a compute-focused accelerator with no display outputs, no graphics API support, and a massive 256 GB HBM3e memory pool. The NVIDIA GeForce RTX 5090 D is a consumer graphics card with full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support, plus display outputs. The recorded benchmark data covers only the RTX 5090 D, which holds a 92nd percentile ranking across all GPUs with an average benchmark score of 77,712. The MI325X has no recorded benchmark scores and sits at the 50th percentile with an average score of zero, so direct performance comparisons in this database are limited to architectural and specification analysis rather than measured head-to-head results.

The RTX 5090 D delivers higher FP32 throughput at 104.8 TFLOPS compared to 81.72 TFLOPS for the MI325X, and it is the only one of the two with rasterization hardware, featuring 176 ROPs and a pixel rate of 423.6 GPixel/s. The MI325X instead prioritizes memory capacity and bandwidth, offering 256 GB of HBM3e across an 8192-bit bus for 6.14 TB/s of bandwidth, which is roughly 3.4 times the bandwidth of the RTX 5090 D's 1.79 TB/s over a 512-bit GDDR7 interface. Buyers needing graphics output, ray tracing, or consumer software support should select the RTX 5090 D. Buyers needing maximum memory capacity for large datasets should select the MI325X.

Architecture Differences

The MI325X uses the Aqua Vanjaram chip built on CDNA 3.0 architecture, fabricated on a 5 nm process at TSMC. The RTX 5090 D uses the GB202 chip built on Blackwell 2.0 architecture, also on a 5 nm TSMC process. The MI325X packs 153,000 million transistors into a 1017 mm² die, yielding a transistor density of 150.4 million per mm². The RTX 5090 D contains 92,200 million transistors on a 750 mm² die, for a density of 122.9 million per mm². The MI325X has the larger die and the higher transistor count, while the RTX 5090 D achieves higher clock speeds: its base clock is 2017 MHz and boost clock is 2407 MHz, versus 1000 MHz base and 2100 MHz boost for the MI325X.

The MI325X has 19,456 shading units and 1,216 texture mapping units, but zero ROPs, zero ray tracing cores, and zero tensor cores. Its pixel rate is listed as 0 MPixel/s. The RTX 5090 D has 21,760 shading units, 680 TMUs, 176 ROPs, 170 ray tracing cores, and 680 tensor cores. The MI325X texture rate is 2,553.6 GTexel/s, which exceeds the RTX 5090 D's 1,636.8 GTexel/s despite the NVIDIA card having more shading units. The RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4; the MI325X lists N/A for all three graphics APIs. The MI325X has no display outputs, while the RTX 5090 D provides 1x HDMI 2.1b and 3x DisplayPort 2.1b.

Head-to-Head Benchmarks

The database contains no shared head-to-head benchmark entries between the two products. The MI325X has an empty benchmarks array, an average benchmark score of zero, and no nearest rivals. The RTX 5090 D has ten recorded benchmark scores. Its 3DMark Steel Nomad DX12 score is 14,326. In Geekbench, it scores 310,674 in OpenCL and 376,915 in Vulkan. Passmark results include 231 in DirectX 10, 371 in DirectX 11, 219 in DirectX 12, 434 in DirectX 9, 1,487 in G2D, 44,065 in G3D, and 28,396 in GPU compute.

The nearest rivals for the RTX 5090 D in the database are AMD Radeon RX 6650M XT with an average score of 76,904 and a delta of +1.1%, AMD Radeon RX 6850M XT at 78,940 with a delta of -1.6%, NVIDIA Tesla P100 PCIe 12 GB at 79,396 with a delta of -2.1%, and NVIDIA Tesla P100 PCIe 16 GB at 79,605 with a delta of -2.4%. These deltas indicate the RTX 5090 D's average benchmark score of 77,712 sits within roughly 2.4% of these four rivals, meaning its measured performance is closely clustered with those products despite its high percentile ranking. The RTX 5090 D's 92nd percentile placement reflects its position above the majority of all GPUs in the database, but its nearest rivals show that the immediate competitive field is tight.

Specification Differences

The two products differ across nearly every specification field. The MI325X uses HBM3e memory with 256 GB capacity, an 8192-bit bus, and 6.14 TB/s bandwidth. The RTX 5090 D uses GDDR7 memory with 32 GB capacity, a 512-bit bus, and 1.79 TB/s bandwidth. Memory clock differs substantially: the MI325X runs at 1500 MHz with 6 Gbps effective, while the RTX 5090 D runs at 1750 MHz with 28 Gbps effective. The MI325X has more TMUs at 1,216 versus 680, but the RTX 5090 D has more shading units at 21,760 versus 19,456. The MI325X has zero ROPs; the RTX 5090 D has 176. The RTX 5090 D has 170 ray tracing cores and 680 tensor cores; the MI325X has none of either.

Power draw differs by a wide margin. The MI325X has a TDP of 1000 W with a suggested PSU of 1400 W, uses an OAM Module slot width, and has no power connectors listed. The RTX 5090 D has a 575 W TDP with a suggested PSU of 950 W, is dual-slot, and uses a single 16-pin connector. Both use PCIe 5.0 x16. The RTX 5090 D measures 304 mm in length, 137 mm in height, and 48 mm in width. The MI325X has no listed dimensions. The MI325X was released on 2024-10-09 and has no production status listed; the RTX 5090 D was released on 2025-01-29 and is marked as Active production. The RTX 5090 D has a predecessor of GeForce 40 and a successor of GeForce 60; the MI325X predecessor is Radeon Instinct and it has no successor listed. The RTX 5090 D launch MSRP is 2,299 USD.

FAQ

Q: Which card has more memory bandwidth?

A: The AMD Instinct MI325X provides 6.14 TB/s of bandwidth from 256 GB of HBM3e on an 8192-bit bus, compared to 1.79 TB/s from 32 GB of GDDR7 on a 512-bit bus for the NVIDIA GeForce RTX 5090 D.

Q: Does the MI325X support graphics APIs?

A: No. The MI325X lists DirectX, OpenGL, and Vulkan as N/A and has no display outputs. The RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, and provides 1x HDMI 2.1b and 3x DisplayPort 2.1b.

Q: Which product has the higher FP32 compute throughput?

A: The NVIDIA GeForce RTX 5090 D delivers 104.8 TFLOPS FP32, while the AMD Instinct MI325X delivers 81.72 TFLOPS FP32. Both list FP16 at a 1:1 ratio with their FP32 figures.

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

A: Its average benchmark score is 77,712. It sits 1.1% above the AMD Radeon RX 6650M XT (76,904), 1.6% below the AMD Radeon RX 6850M XT (78,940), 2.1% below the NVIDIA Tesla P100 PCIe 12 GB (79,396), and 2.4% below the NVIDIA Tesla P100 PCIe 16 GB (79,605).

Q: What are the transistor counts of each chip?

A: The MI325X's Aqua Vanjaram chip contains 153,000 million transistors on a 1017 mm² die. The RTX 5090 D's GB202 chip contains 92,200 million transistors on a 750 mm² die.

Q: Which product requires more power?

A: The MI325X has a 1000 W TDP and a suggested PSU of 1400 W. The RTX 5090 D has a 575 W TDP and a suggested PSU of 950 W.

Where Each One Wins

The RTX 5090 D wins in graphics-centric workloads. It is the only product with ROPs, ray tracing cores, tensor cores, and a non-zero pixel rate of 423.6 GPixel/s. It supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, and it has display outputs. Its 104.8 TFLOPS FP32 exceeds the MI325X by roughly 28%. Its base clock of 2017 MHz is more than double the MI325X's 1000 MHz base, and its boost clock of 2407 MHz is higher as well. It has more shading units at 21,760 versus 19,456, and its 176 ROPs enable rasterization that the MI325X simply cannot perform. The RTX 5090 D is also the only one with recorded benchmark scores, holding a 92nd percentile ranking across all GPUs.

The MI325X wins in memory capacity and bandwidth. Its 256 GB HBM3e pool is eight times the capacity of the RTX 5090 D's 32 GB. Its 6.14 TB/s bandwidth is roughly 3.4 times the RTX 5090 D's 1.79 TB/s. The MI325X also has a wider memory bus at 8192 bits versus 512 bits. Its texture rate of 2,553.6 GTexel/s is higher than the RTX 5090 D's 1,636.8 GTexel/s, and it has more TMUs at 1,216 versus 680. The MI325X uses an OAM Module form factor with no power connectors, indicating a server-oriented design, and it carries more transistors at 153,000 million compared to 92,200 million, on a larger 1017 mm² die versus 750 mm².

The data indicates a clear split. The RTX 5090 D is the choice for graphics, ray tracing, and consumer software environments. The MI325X is the choice for memory-bound compute tasks where 256 GB of HBM3e and 6.14 TB/s of bandwidth matter more than rasterization or graphics API support. The RTX 5090 D holds the only recorded benchmark results in this database, so its measured performance is verifiable, while the MI325X's capabilities are documented only through its specifications.