AMD Instinct MI350P vs NVIDIA GeForce RTX 5090 D V2 Comparison

FAQ

Q: What are the architectural generations of the AMD Instinct MI350P and NVIDIA GeForce RTX 5090 D V2?

A: The AMD Instinct MI350P uses the CDNA 4.0 architecture with the MI350 128CU chip. The NVIDIA GeForce RTX 5090 D V2 uses the Blackwell 2.0 architecture with the GB202 chip.

Q: How do the memory configurations differ between these two cards?

A: The AMD Instinct MI350P has 144 GB of HBM3e memory on an 8192-bit bus, delivering 8.19 TB/s bandwidth. The NVIDIA GeForce RTX 5090 D V2 has 24 GB of GDDR7 memory on a 384-bit bus, delivering 1.34 TB/s bandwidth.

Q: Which card has higher FP32 compute performance?

A: The NVIDIA GeForce RTX 5090 D V2 delivers 104.8 TFLOPS FP32, which is substantially higher than the AMD Instinct MI350P's 36.04 TFLOPS. Both cards achieve their FP16 performance at a 1:1 ratio with FP32.

Q: What are the physical dimensions and power requirements of each card?

A: The AMD Instinct MI350P measures 267 mm in length, 111 mm in height, and 40 mm in width, with a 600 W TDP and a suggested 1000 W PSU. The NVIDIA GeForce RTX 5090 D V2 measures 304 mm by 137 mm by 48 mm, with a 575 W TDP and a suggested 950 W PSU.

Q: Does the AMD Instinct MI350P support display outputs?

A: No, the AMD Instinct MI350P 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 timeline for these products?

A: The NVIDIA GeForce RTX 5090 D V2 was released on 2025-08-14 and is listed as Active in production. The AMD Instinct MI350P has a release date of 2026-05-06.

Where Each One Wins

The recorded data splits these two cards into distinct usage domains. The AMD Instinct MI350P is built for memory-bound acceleration workloads. Its 144 GB HBM3e frame buffer at 8.19 TB/s bandwidth dwarfs the NVIDIA card's 24 GB GDDR7 at 1.34 TB/s. For large model inference, massive dataset processing, or any workload where capacity and memory bandwidth dominate, the MI350P holds the clear advantage. The 8192-bit memory bus is six times wider than the GeForce card's 384-bit bus, and the bandwidth advantage exceeds 6x.

The NVIDIA GeForce RTX 5090 D V2 wins on raw compute throughput and graphics features. Its 104.8 TFLOPS FP32 nearly triples the MI350P's 36.04 TFLOPS. The 21760 shading units, 680 TMUs, and 176 ROPs give it the texture rate of 1,636.8 GTexel/s and pixel rate of 423.6 GPixel/s, while the MI350P records 0 ROPs and 0 MPixel/s pixel rate. The NVIDIA card also integrates 170 RT cores and 680 tensor cores, features absent from the AMD accelerator.

For rendering, rasterization, ray tracing, or any DirectX 12 Ultimate workload, the GeForce RTX 5090 D V2 is the only viable option of the two. The MI350P reports N/A for DirectX, OpenGL, and Vulkan APIs, making it unsuitable for conventional graphics pipelines. The NVIDIA card supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

The benchmark data shows only the NVIDIA card has any recorded results. Its 3DMark Steel Nomad DX12 score of 16504 places it at the 59th percentile among all GPUs in the database. The MI350P sits at the 50th percentile with an average benchmark score of 0, indicating no comparable graphics benchmark results exist for it.

Architecture Differences

The AMD Instinct MI350P is built on CDNA 4.0, AMD's compute-optimized architecture. The chip is fabricated on a 3 nm process at TSMC, containing 73,000 million transistors on a 1190 mm² die. This yields a transistor density of 61.3M per mm². The design forgoes traditional graphics fixed-function units, as evidenced by the 0 ROPs and 0 MPixel/s pixel rate. It has 8192 shading units and 512 TMUs. The 144 GB HBM3e memory on an 8192-bit interface is the defining architectural trait, optimized for bandwidth and capacity rather than latency-sensitive graphics work.

The NVIDIA GeForce RTX 5090 D V2 uses Blackwell 2.0, a fully featured graphics and compute architecture. The GB202 chip is built on a 5 nm TSMC process with 92,200 million transistors on a 750 mm² die, achieving a transistor density of 122.9M per mm². This density advantage reflects a more compact design with higher integration. The architecture includes 21760 shading units, 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores. Memory is 24 GB of GDDR7 on a 384-bit bus.

The process node difference matters: the MI350P uses 3 nm versus 5 nm for the GeForce. Despite the smaller node, the AMD chip has a lower transistor density because of the massive HBM3e memory subsystem and the large 1190 mm² die. The NVIDIA chip packs more transistors into a smaller 750 mm² area.

The MI350P's predecessor is Radeon Instinct, while the GeForce RTX 5090 D V2 succeeds the GeForce 40 series and has a successor listed as GeForce 60. The NVIDIA card is part of the GeForce 50-series generation; the AMD card belongs to the Instinct (MIx) generation.

Specification Differences

The two cards differ across nearly every specification field. The MI350P has a base clock of 1000 MHz and boost clock of 2200 MHz, while the GeForce RTX 5090 D V2 runs at 2017 MHz base and 2407 MHz boost. Memory clocks also diverge: the AMD card operates at 2000 MHz with 8 Gbps effective, the NVIDIA card at 1750 MHz with 28 Gbps effective.

Memory size is the largest gap. The MI350P offers 144 GB of HBM3e, the GeForce has 24 GB of GDDR7. Bus width is 8192 bit versus 384 bit. Bandwidth is 8.19 TB/s versus 1.34 TB/s.

Shading units are 8192 on the AMD card versus 21760 on the NVIDIA card. TMUs are 512 versus 680. ROPs are 0 versus 176. The NVIDIA card includes 170 RT cores and 680 tensor cores; the AMD card lists neither. Pixel rate is 0 MPixel/s versus 423.6 GPixel/s. Texture rate is 1,126.4 GTexel/s versus 1,636.8 GTexel/s. FP32 compute is 36.04 TFLOPS versus 104.8 TFLOPS, with both at 1:1 FP16.

Power draw is close: 600 W TDP for the AMD card, 575 W for the NVIDIA card. Both use a single 16-pin power connector. Suggested PSU is 1000 W for the AMD card and 950 W for the NVIDIA card. Both are dual-slot designs with PCIe 5.0 x16 interfaces.

Dimensions differ notably. The MI350P is 267 mm long, 111 mm tall, and 40 mm wide. The GeForce RTX 5090 D V2 is 304 mm long, 137 mm tall, and 48 mm wide. The NVIDIA card is larger in every dimension.

Display outputs are absent on the MI350P. The GeForce card has 1x HDMI 2.1b and 3x DisplayPort 2.1b. API support is N/A for the AMD card; the NVIDIA card supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The GeForce card has a launch MSRP of 2,299 USD.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark entries between the AMD Instinct MI350P and NVIDIA GeForce RTX 5090 D V2. The MI350P has no recorded benchmark scores and zero wins in the comparison. The GeForce RTX 5090 D V2 has one benchmark result and zero wins in this pairing.

The only available benchmark data belongs to the NVIDIA card. Its 3DMark Steel Nomad DX12 score is 16504. That result places it at the 59th percentile among all GPUs, with an average benchmark score of 16504. The nearest rivals in the database show how tightly grouped this performance tier 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 clustering indicates the GeForce RTX 5090 D V2's benchmark performance sits within 1% of several other cards in the database. The 16504 score is effectively at parity with the T400, which leads by just 4 points. The margin over the RX 5700 XT is 143 points, roughly 0.9%.

For the MI350P, the absence of benchmark scores means the data cannot confirm any graphics performance. The card's 50th percentile standing with an average score of 0 reflects this lack of recorded results. The architectural specifications suggest its strengths lie outside conventional graphics benchmarks.

The FP32 compute comparison is the clearest numerical gap. The GeForce RTX 5090 D V2 delivers 104.8 TFLOPS, which is 2.9 times the MI350P's 36.04 TFLOPS. Texture rate favors the NVIDIA card by 1,636.8 GTexel/s versus 1,126.4 GTexel/s, a 45% advantage. The memory bandwidth comparison reverses the outcome: the MI350P's 8.19 TB/s is 6.1 times the GeForce's 1.34 TB/s.

The transistor counts show the NVIDIA chip packs 92,200 million transistors versus 73,000 million for the AMD chip, a 26% advantage. Die size favors the AMD card at 1190 mm² versus 750 mm², which combined with the different process nodes explains the density gap of 61.3M per mm² versus 122.9M per mm².

Clock speeds favor the NVIDIA card on both base and boost. The 2017 MHz base clock is roughly double the MI350P's 1000 MHz. Boost clocks are 2407 MHz versus 2200 MHz. The shading unit count of 21760 versus 8192 gives the GeForce a 2.7x advantage in parallel execution width, though the AMD card's memory system is designed for different workload patterns.

The data indicates two specialized tools rather than direct competitors. The MI350P targets memory-capacity-bound acceleration with its 144 GB HBM3e and 8.19 TB/s bandwidth. The GeForce RTX 5090 D V2 targets graphics and compute throughput with its 104.8 TFLOPS, 170 RT cores, and full API support. Neither card's recorded specifications suggest it can substitute for the other's primary role.