AMD Instinct MI350X vs NVIDIA GeForce RTX 5090 D Comparison

Head-to-Head Benchmarks

The recorded data shows a complete asymmetry between these two accelerators. The AMD Instinct MI350X has no benchmark entries in the database, resulting in an average benchmark score of zero and a percentile ranking of 50 among all GPUs. The NVIDIA GeForce RTX 5090 D, by contrast, holds a substantial benchmark portfolio with an average score of 77,712 and a percentile ranking of 92. This means the RTX 5090 D sits well above the MI350X in the database’s overall performance distribution, a gap that reflects the absence of measured results rather than necessarily a raw compute deficit.

Looking at the RTX 5090 D’s individual scores, the strongest showing comes from Geekbench Vulkan at 376,915, followed by Geekbench OpenCL at 310,674. These are the highest recorded figures for this card and indicate strong general-purpose compute throughput under those specific workloads. The Passmark suite tells a more varied story: Passmark G3D scores 44,065, Passmark GPU Compute scores 28,396, and Passmark G2D scores 1,487. The DirectX-specific tests are notably lower, with DirectX 9 at 434, DirectX 11 at 371, DirectX 10 at 231, and DirectX 12 at 219. The delta between the G3D aggregate and the individual DirectX tests suggests that the card’s overall 3D mark is driven by factors beyond those particular legacy API paths.

In terms of nearest rivals for the RTX 5090 D, the database lists the AMD Radeon RX 6650M XT with an average score of 76,904 and a delta of 1.1%, meaning the RTX 5090 D trails that mobile card by a hair over one percent. The AMD Radeon RX 6850M XT scores 78,940, putting the RTX 5090 D 1.6% behind. Two NVIDIA Tesla P100 variants bracket the RTX 5090 D as well: the PCIe 12 GB version scores 79,396 (2.1% ahead of the RTX 5090 D) and the PCIe 16 GB version scores 79,605 (2.4% ahead). These are remarkably close margins, indicating that in the database’s aggregate scoring, the RTX 5090 D lands in a tight cluster with several other accelerators rather than dominating them.

Because the MI350X has no recorded benchmarks, there are no direct head-to-head test results to compare. The wins count stands at zero for the AMD part and zero for the NVIDIA part, which is consistent with an empty head-to-head benchmark array. The practical implication is that any performance comparison must rely on architectural and specification-level analysis rather than measured test outcomes.

Architecture Differences

The two chips diverge fundamentally in design philosophy. The AMD Instinct MI350X uses the MI350 256CU die built on CDNA 4.0 architecture, manufactured at TSMC on a 3 nm process. The NVIDIA GeForce RTX 5090 D uses the GB202 die on Blackwell 2.0 architecture, also TSMC but on a 5 nm process. The transistor counts reflect the process gap: the MI350X packs 185,000 million transistors into a 2380 mm² die, yielding a transistor density of 77.7 million per square millimeter. The RTX 5090 D contains 92,200 million transistors on a 750 mm² die, achieving a higher density of 122.9 million per square millimeter despite the larger process node.

Memory architecture is another stark differentiator. The MI350X carries 288 GB of HBM3e across an 8192-bit bus, producing 8.19 TB/s of bandwidth. The RTX 5090 D uses 32 GB of GDDR7 on a 512-bit bus, delivering 1.79 TB/s. The MI350X’s memory bandwidth is roughly 4.6 times that of the RTX 5090 D, a massive advantage for data movement, while the RTX 5090 D’s memory capacity is far smaller, reflecting its consumer-oriented design versus the MI350X’s server-level HBM configuration.

Compute resources tell a different story. The RTX 5090 D has 21,760 shading units, 680 texture mapping units, and 176 raster operations pipelines, plus 170 ray tracing cores and 680 tensor cores. The MI350X has 16,384 shading units and 1,024 TMUs, but reports zero ROPs and no ray tracing or tensor core counts. The MI350X also lists a pixel rate of 0 MPixel/s, consistent with a compute accelerator that has no display rasterization path. Its texture rate is 2,252.8 GTexel/s, which exceeds the RTX 5090 D’s 1,636.8 GTexel/s. The RTX 5090 D counters with a pixel rate of 423.6 GPixel/s.

Clock speeds favor the NVIDIA part. The RTX 5090 D runs a 2017 MHz base and 2407 MHz boost, while the MI350X runs a 1000 MHz base and 2200 MHz boost. The floating-point throughput figures are notable: the RTX 5090 D delivers 104.8 TFLOPS for both FP32 and FP16 (at a 1:1 ratio), while the MI350X delivers 72.09 TFLOPS for both FP32 and FP16. Despite having fewer shading units and lower clocks, the MI350X’s FP32 figure is still substantial, though roughly 31% below the RTX 5090 D in this metric.

Power and physical design differ as well. The MI350X has a TDP of 1000 W and requires a suggested PSU of 1400 W. It comes as an OAM Module with no power connectors and no display outputs. The RTX 5090 D has a TDP of 575 W with a suggested PSU of 950 W, uses a dual-slot form factor with a single 16-pin connector, and provides 1x HDMI 2.1b plus 3x DisplayPort 2.1b outputs. The MI350X measures 102 mm in length and 165 mm in width, while the RTX 5090 D is 304 mm long, 137 mm high, and 48 mm wide.

API support is another clear divide. The MI350X lists DirectX, OpenGL, and Vulkan all as N/A, confirming its role as a non-graphics compute device. The RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Both use a PCIe 5.0 x16 bus interface.

The Verdict

The data supports a clear split based on workload type. For any application requiring graphics output, rasterization, ray tracing, or standard consumer API support, the RTX 5090 D is the only functional choice. It has display outputs, a full DirectX 12 Ultimate feature set, OpenGL 4.6, Vulkan 1.4, 176 ROPs, and a pixel rate of 423.6 GPixel/s. The MI350X has none of these capabilities, with zero ROPs, no display outputs, and N/A API support.

For compute-heavy, memory-bound workloads, the MI350X offers structural advantages that the RTX 5090 D cannot match. The 288 GB HBM3e pool with 8.19 TB/s bandwidth is more than eight times the memory capacity and over four times the bandwidth of the RTX 5090 D. The MI350X also has a higher texture rate at 2,252.8 GTexel/s. However, the RTX 5090 D holds the FP32 and FP16 throughput advantage at 104.8 TFLOPS versus 72.09 TFLOPS.

The RTX 5090 D is the only part with any recorded benchmark data, and its nearest-rival deltas are all within 2.4%, meaning the database places it in a competitive mid-pack for its class. The MI350X has no such data, so its percentile ranking of 50 is an artifact of missing measurements rather than a meaningful performance statement. The release dates differ: the RTX 5090 D launched on 2025-01-29, while the MI350X followed on 2025-06-11. The RTX 5090 D lists a launch MSRP of 2,299 USD, while the MI350X has no listed launch MSRP.

The verdict, strictly from the data, is that these are different tools. The RTX 5090 D is a complete graphics and compute solution with measured performance and consumer connectivity. The MI350X is a high-capacity compute accelerator with no graphics path and no measured benchmarks, aimed at workloads where 288 GB of HBM3e and 8.19 TB/s of bandwidth matter more than FP32 peak throughput.

FAQ

Q: Which card has higher FP32 peak throughput?

A: The NVIDIA GeForce RTX 5090 D delivers 104.8 TFLOPS FP32, while the AMD Instinct MI350X delivers 72.09 TFLOPS FP32.

Q: What memory configuration does each card use?

A: The MI350X uses 288 GB of HBM3e on an 8192-bit bus with 8.19 TB/s bandwidth. The RTX 5090 D uses 32 GB of GDDR7 on a 512-bit bus with 1.79 TB/s bandwidth.

Q: Does the MI350X support any graphics APIs?

A: No. The MI350X lists DirectX, OpenGL, and Vulkan all as N/A, and it has no display outputs. The RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

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

A: The RTX 5090 D sits within 2.4% of its nearest rivals. It trails the AMD Radeon RX 6650M XT by 1.1%, the AMD Radeon RX 6850M XT by 1.6%, the NVIDIA Tesla P100 PCIe 12 GB by 2.1%, and the NVIDIA Tesla P100 PCIe 16 GB by 2.4%.

Q: What are the TDP requirements for each card?

A: The MI350X has a TDP of 1000 W with a suggested PSU of 1400 W. The RTX 5090 D has a TDP of 575 W with a suggested PSU of 950 W.

Q: Which card has more texture mapping units?

A: The MI350X has 1,024 TMUs and a texture rate of 2,252.8 GTexel/s. The RTX 5090 D has 680 TMUs and a texture rate of 1,636.8 GTexel/s.

Where Each One Wins

The RTX 5090 D wins in every scenario involving graphics rendering or display output. It provides 1x HDMI 2.1b and 3x DisplayPort 2.1b connections, supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, and has 176 ROPs with a pixel rate of 423.6 GPixel/s. Its 170 ray tracing cores and 680 tensor cores give it dedicated acceleration paths for ray-traced workloads and tensor operations, neither of which the MI350X lists. The RTX 5090 D also wins on raw FP32 and FP16 throughput at 104.8 TFLOPS for both, compared to 72.09 TFLOPS for the MI350X. It also has a higher base clock (2017 MHz versus 1000 MHz) and boost clock (2407 MHz versus 2200 MHz), and a smaller physical footprint at 304 mm by 137 mm by 48 mm versus the MI350X’s 102 mm by 165 mm OAM module.

The MI350X wins in memory capacity and bandwidth by a wide margin. Its 288 GB HBM3e pool dwarfs the RTX 5090 D’s 32 GB GDDR7, and its 8.19 TB/s bandwidth is more than four times the RTX 5090 D’s 1.79 TB/s. The MI350X also has a wider memory bus at 8192 bit versus 512 bit, and a higher texture rate at 2,252.8 GTexel/s versus 1,636.8 GTexel/s. The MI350X uses a larger die at 2380 mm² versus 750 mm² and has more total transistors at 185,000 million versus 92,200 million. It also carries more TMUs at 1,024 versus 680.

In terms of database standing, the RTX 5090 D is the only card with measured results. Its percentile rank of 92 reflects actual benchmark scores, while the MI350X’s percentile of 50 is a placeholder for an unmeasured device. The RTX 5090 D also has an active production status, whereas the MI350X has no production status listed. The RTX 5090 D’s nearest rivals are all within 2.4% of its average score, placing it in a tightly competitive cluster. The MI350X has no nearest rivals listed.

Specification Differences

The two cards differ across nearly every specification field. The process node is 3 nm for the MI350X versus 5 nm for the RTX 5090 D, both at TSMC. Transistor count is 185,000 million versus 92,200 million, and die size is 2380 mm² versus 750 mm². Transistor density is 77.7M per mm² for the MI350X and 122.9M per mm² for the RTX 5090 D.

Clock speeds: base clock is 1000 MHz for the MI350X and 2017 MHz for the RTX 5090 D; boost clock is 2200 MHz versus 2407 MHz; memory clock is 2000 MHz (8 Gbps effective) versus 1750 MHz (28 Gbps effective).

Memory: 288 GB HBM3e versus 32 GB GDDR7; 8192-bit bus versus 512-bit; 8.19 TB/s versus 1.79 TB/s.

Compute units: 16,384 shading units versus 21,760; 1,024 TMUs versus 680; 0 ROPs versus 176; no RT cores listed versus 170; no tensor cores listed versus 680.

Rates: pixel rate is 0 MPixel/s versus 423.6 GPixel/s; texture rate is 2,252.8 GTexel/s versus 1,636.8 GTexel/s; FP32 is 72.09 TFLOPS versus 104.8 TFLOPS; FP16 is 72.09 TFLOPS (1:1) versus 104.8 TFLOPS (1:1).

Power: TDP is 1000 W versus 575 W; suggested PSU is 1400 W versus 950 W; slot width is OAM Module versus dual-slot; power connectors are none versus 1x 16-pin.

Connectivity and output: both use PCIe 5.0 x16; the MI350X has no display outputs, while the RTX 5090 D has 1x HDMI 2.1b and 3x DisplayPort 2.1b. APIs are N/A for the MI350X and DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 for the RTX 5090 D.

Physical dimensions: the MI350X is 102 mm long and 165 mm wide; the RTX 5090 D is 304 mm long, 137 mm high, and 48 mm wide.

Release and status: the MI350X released on 2025-06-11 with no production status; the RTX 5090 D released on 2025-01-29 and is marked Active. The MI350X’s predecessor is Radeon Instinct; the RTX 5090 D’s predecessor is GeForce 40 and its successor is GeForce 60. The MI350X has no launch MSRP; the RTX 5090 D lists a launch MSRP of 2,299 USD.