AMD Instinct MI325X vs NVIDIA GeForce RTX 5090 D V2 Comparison

FAQ

Q: What is the primary architectural difference between the AMD Instinct MI325X and the NVIDIA GeForce RTX 5090 D V2?

A: The AMD Instinct MI325X uses the CDNA 3.0 architecture with the Aqua Vanjaram chip, while the NVIDIA GeForce RTX 5090 D V2 uses the Blackwell 2.0 architecture with the GB202 chip. Both are fabricated on a 5 nm process at TSMC.

Q: How do the memory configurations compare between the two cards?

A: The AMD Instinct MI325X has 256 GB of HBM3e memory on a 8192-bit bus with 6.14 TB/s bandwidth. The NVIDIA GeForce RTX 5090 D V2 has 24 GB of GDDR7 memory on a 384-bit bus with 1.34 TB/s bandwidth.

Q: Which card has a higher FP32 compute rating?

A: The NVIDIA GeForce RTX 5090 D V2 delivers 104.8 TFLOPS FP32, while the AMD Instinct MI325X delivers 81.72 TFLOPS FP32. The NVIDIA card is approximately 28% higher in this metric.

Q: What is the power requirement difference?

A: The AMD Instinct MI325X has a TDP of 1000 W and a suggested PSU of 1400 W. The NVIDIA GeForce RTX 5090 D V2 has a TDP of 575 W and a suggested PSU of 950 W.

Q: Does the AMD card support display outputs?

A: No, the AMD Instinct MI325X has no display outputs. The NVIDIA GeForce RTX 5090 D V2 includes 1x HDMI 2.1b and 3x DisplayPort 2.1b outputs.

Q: What is the release timing for each product?

A: The AMD Instinct MI325X was released on October 9, 2024. The NVIDIA GeForce RTX 5090 D V2 was released on August 14, 2025.

Where Each One Wins

The data reveals two entirely different usage profiles. The AMD Instinct MI325X targets compute-heavy environments where massive memory capacity and bandwidth matter more than rendering features. Its 256 GB HBM3e pool with 6.14 TB/s bandwidth positions it for large-scale data workloads, model training, and memory-bound scientific computing. The absence of display outputs and its OAM Module form factor confirm it is designed for server racks, not desktop workstations.

The NVIDIA GeForce RTX 5090 D V2 wins on raw shader throughput, pixel processing, and API support. It delivers 104.8 TFLOPS FP32 versus 81.72 TFLOPS for the AMD part, a 28% advantage. It also has 21760 shading units versus 19456, along with 176 ROPs where the AMD card lists 0. The NVIDIA card supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4; the AMD card lists N/A for all three APIs. This makes the RTX 5090 D V2 the functional choice for graphics rendering, gaming, and any workload requiring a display output.

The NVIDIA card also carries the only recorded benchmark result in the database. Its 3DMark Steel Nomad DX12 score of 16504 places it at the 59th percentile among all GPUs. The AMD Instinct MI325X has no benchmark entries and sits at the 50th percentile by default. For synthetic graphics benchmarks, the NVIDIA card is the only one with measurable data.

Architecture Differences

The AMD Instinct MI325X uses the CDNA 3.0 architecture, AMD's compute-optimized design lineage. It has no RT cores and no tensor cores listed, which means it does not expose dedicated hardware for ray tracing or tensor operations in the recorded specifications. The chip, Aqua Vanjaram, contains 153,000 million transistors on a 1017 mm² die, giving a transistor density of 150.4M per mm². The card has 19456 shading units and 1216 TMUs but zero ROPs, which is consistent with a compute accelerator that does not rasterize frames.

The NVIDIA GeForce RTX 5090 D V2 uses the Blackwell 2.0 architecture, a graphics-first design. It includes 170 RT cores and 680 tensor cores, providing dedicated hardware for ray tracing and AI acceleration. The GB202 chip contains 92,200 million transistors on a 750 mm² die, with a lower transistor density of 122.9M per mm². It has 21760 shading units, 680 TMUs, and 176 ROPs, enabling full rasterization and pixel output at 423.6 GPixel/s.

The memory architectures differ fundamentally. The AMD card uses HBM3e on an 8192-bit interface, which explains its 6.14 TB/s bandwidth. The NVIDIA card uses GDDR7 on a 384-bit interface, delivering 1.34 TB/s. The AMD card's memory clock is listed as 1500 MHz with 6 Gbps effective, while the NVIDIA card runs at 1750 MHz with 28 Gbps effective. The higher effective data rate per pin on GDDR7 compensates partially for the narrower bus, but the aggregate bandwidth still favors the AMD card by a factor of roughly 4.6.

Specification Differences

The two cards differ in nearly every measurable specification. The AMD Instinct MI325X has a base clock of 1000 MHz and a boost clock of 2100 MHz. The NVIDIA GeForce RTX 5090 D V2 has a base clock of 2017 MHz and a boost clock of 2407 MHz, making it significantly faster at both points.

Memory capacity favors AMD heavily: 256 GB versus 24 GB. Memory type differs as HBM3e versus GDDR7. Bus width is 8192 bit versus 384 bit. Bandwidth is 6.14 TB/s versus 1.34 TB/s.

Shading units favor NVIDIA at 21760 versus 19456. TMUs favor AMD at 1216 versus 680. ROPs favor NVIDIA at 176 versus 0. RT cores exist only on NVIDIA at 170. Tensor cores exist only on NVIDIA at 680.

Pixel rate is 0 MPixel/s for AMD and 423.6 GPixel/s for NVIDIA. Texture rate is 2,553.6 GTexel/s for AMD and 1,636.8 GTexel/s for NVIDIA. FP32 and FP16 are both 81.72 TFLOPS for AMD and 104.8 TFLOPS for NVIDIA.

Power draw differs substantially: 1000 W TDP for AMD versus 575 W for NVIDIA. The AMD card uses an OAM Module slot width with no power connectors listed, while NVIDIA uses a dual-slot design with a single 16-pin connector. The suggested PSU is 1400 W for AMD and 950 W for NVIDIA.

The AMD card has no display outputs; NVIDIA has 1x HDMI 2.1b and 3x DisplayPort 2.1b. API support is N/A across the board for AMD, while NVIDIA supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The NVIDIA card measures 304 mm in length, 137 mm in height, and 48 mm in width.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark comparisons between these two cards. The only recorded test result belongs to the NVIDIA GeForce RTX 5090 D V2: a 3DMark Steel Nomad DX12 score of 16504. That result places the NVIDIA card at the 59th percentile among all GPUs, with an average benchmark score of 16504.

The nearest rivals for the NVIDIA card in the database show how tightly clustered this score is. The NVIDIA T400 scores 16508, a delta of 0%. The AMD Radeon PRO W7500 scores 16415, a delta of 0.5%. The NVIDIA RTX PRO 6000 Blackwell scores 16408, a delta of 0.6%. The AMD Radeon RX 5700 XT scores 16361, a delta of 0.9%. This means the RTX 5090 D V2 performs within 1% of these four cards in the Steel Nomad test, despite the massive architectural differences between them.

The AMD Instinct MI325X has no benchmark scores in the database. Its percentile of 50 is a neutral default, not a measured performance result. The wins count is 0 for both cards in head-to-head comparisons, reflecting the absence of paired test data.

Without direct comparisons, the recorded specifications provide the only basis for analysis. The NVIDIA card holds a 28% FP32 advantage, a 170-to-0 advantage in RT cores, and a 680-to-0 advantage in tensor cores. The AMD card holds a 4.6-to-1 advantage in memory bandwidth and a 10.7-to-1 advantage in memory capacity.

The Verdict

The data describes two accelerators built for different purposes. The AMD Instinct MI325X is a server-oriented compute module with no display output, no consumer API support, and no benchmark entries. Its strengths are memory capacity and bandwidth: 256 GB of HBM3e at 6.14 TB/s. That configuration suits workloads where the dataset exceeds what a 24 GB frame buffer can hold. The 1000 W TDP and OAM Module form factor indicate a rack-mount deployment with dedicated cooling and power infrastructure.

The NVIDIA GeForce RTX 5090 D V2 is a conventional graphics card with a dual-slot design, display outputs, and full API support. Its measured 3DMark Steel Nomad score of 16504 places it in the 59th percentile, and its nearest rivals all fall within 0.9% of that result. The card delivers 104.8 TFLOPS FP32, 423.6 GPixel/s pixel rate, and includes 170 RT cores and 680 tensor cores for ray tracing and AI workloads.

A builder choosing between these cards should look at the workload first. If the task requires rasterization, ray tracing, display output, or consumer software compatibility, the NVIDIA card is the only option with recorded support for those features. If the task requires processing datasets larger than 24 GB with maximum memory bandwidth, the AMD card offers 10.7 times the capacity and 4.6 times the bandwidth. The NVIDIA card's launch MSRP is 2,299 USD. The AMD card has no launch MSRP recorded. The two cards do not compete in the same segment; they serve adjacent but distinct use cases.