NVIDIA GeForce RTX 5090 SE vs NVIDIA Rubin GPU Comparison

FAQ

Q: What are the core architectural identities of the NVIDIA GeForce RTX 5090 SE and the NVIDIA Rubin GPU?

A: The RTX 5090 SE is built on the Blackwell 2.0 architecture using the GB202 chip, fabricated on a 5 nm process at TSMC. The Rubin GPU uses the Rubin architecture with the GR100 chip, fabricated on a 3 nm process at TSMC.

Q: How do the memory subsystems compare between the two GPUs?

A: The RTX 5090 SE has 24 GB of GDDR7 memory on a 384-bit bus, delivering 1.34 TB/s of bandwidth. The Rubin GPU has 288 GB of HBM4 memory on a 16384-bit bus, delivering 22.1 TB/s of bandwidth, which is roughly 16.5 times the bandwidth of the RTX 5090 SE.

Q: Which GPU has higher FP32 compute throughput?

A: The Rubin GPU delivers 130.0 TFLOPS of FP32 performance, while the RTX 5090 SE delivers 66.94 TFLOPS. The Rubin GPU leads by approximately 1.94 times in raw FP32 throughput.

Q: What are the power requirements for each GPU?

A: The RTX 5090 SE has a TDP of 500 W with a suggested PSU of 900 W, using a single 16-pin power connector. The Rubin GPU has a TDP of 2300 W with a suggested PSU of 2700 W and uses an SXM Module slot width.

Q: Do both GPUs support the same API feature sets?

A: No. The RTX 5090 SE supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The Rubin GPU has no graphics API support recorded, with DirectX, OpenGL, and Vulkan all listed as N/A.

Q: What is the transistor density difference between the two chips?

A: The Rubin GPU has a transistor density of 230.8M per mm², while the RTX 5090 SE has a density of 122.9M per mm². The Rubin GPU's density is about 1.88 times higher, reflecting its smaller 3 nm process node.

Architecture Differences

The RTX 5090 SE and the Rubin GPU represent two distinct design philosophies within NVIDIA's product stack. The RTX 5090 SE is a consumer-oriented graphics card in the GeForce 50-series, based on the Blackwell 2.0 architecture with the GB202 chip. It uses a 5 nm TSMC process, houses 92,200 million transistors on a 750 mm² die, and achieves a transistor density of 122.9M per mm². The Rubin GPU is a server-class accelerator from the Server Rubin (Rxx) generation, using the GR100 chip on the Rubin architecture. It is fabricated on a 3 nm TSMC process, integrates 336,000 million transistors across a 1456 mm² die, and reaches a transistor density of 230.8M per mm².

The memory architecture diverges sharply. The RTX 5090 SE uses 24 GB of GDDR7 on a 384-bit bus, yielding 1.34 TB/s bandwidth. The Rubin GPU uses 288 GB of HBM4 on a 16384-bit bus, yielding 22.1 TB/s bandwidth. The Rubin GPU's memory capacity is 12 times larger, and its bus width is over 42 times wider. The clock speeds also differ: the RTX 5090 SE runs at a base of 1740 MHz and a boost of 2377 MHz, while the Rubin GPU runs at a base of 700 MHz and a boost of 2267 MHz. Despite a much lower base clock, the Rubin GPU achieves similar boost clocks.

Compute resources are heavily skewed toward the Rubin GPU. It has 28,672 shading units, 896 TMUs, and 896 tensor cores. The RTX 5090 SE has 14,080 shading units, 440 TMUs, and 440 tensor cores. The Rubin GPU doubles the shading units, TMUs, and tensor cores. The RTX 5090 SE has 160 ROPs, while the Rubin GPU has only 24 ROPs. This low ROP count, combined with a 54.41 GPixel/s pixel rate versus the RTX 5090 SE's 380.3 GPixel/s, indicates the Rubin GPU is not designed for traditional rasterization output.

The FP32 throughput is 66.94 TFLOPS on the RTX 5090 SE, while the Rubin GPU delivers 130.0 TFLOPS. FP16 performance differs in ratio: the RTX 5090 SE sustains 66.94 TFLOPS (1:1), while the Rubin GPU delivers 260.0 TFLOPS (2:1). Texture rate also favors the Rubin GPU at 2,031.2 GTexel/s versus 1,045.9 GTexel/s.

Interface and physical specifications diverge completely. The RTX 5090 SE uses PCIe 5.0 x16, is a dual-slot card measuring 267 mm in length, 111 mm in height, and 40 mm in width, and provides display outputs of 1x HDMI 2.1b and 3x DisplayPort 2.1b. The Rubin GPU uses PCIe 6.0 x16, is an SXM Module with no display outputs, and has no recorded dimensions. Power delivery also differs: the RTX 5090 SE draws 500 W via a single 16-pin connector, while the Rubin GPU draws 2300 W with no discrete power connector listed, relying on the SXM module's backplane.

Head-to-Head Benchmarks

The recorded benchmark suite contains no direct comparison scores, so the analysis relies on recorded specifications and computed rates. The most decisive advantage for the Rubin GPU is in memory bandwidth. The Rubin GPU's 22.1 TB/s is approximately 16.5 times the RTX 5090 SE's 1.34 TB/s. This is a fundamental gap that affects any memory-bound workload, particularly large model inference or training.

In FP32 compute, the Rubin GPU scores 130.0 TFLOPS versus 66.94 TFLOPS for the RTX 5090 SE, a 1.94x lead. In FP16, the Rubin GPU's 260.0 TFLOPS (2:1) is 3.88 times the RTX 5090 SE's 66.94 TFLOPS (1:1). The texture rate shows a smaller but still clear advantage: 2,031.2 GTexel/s for Rubin versus 1,045.9 GTexel/s for the RTX 5090 SE, a 1.94x difference.

The RTX 5090 SE wins in pixel throughput. Its 380.3 GPixel/s is 6.99 times the Rubin GPU's 54.41 GPixel/s. This is a consequence of the RTX 5090 SE's 160 ROPs versus 24 ROPs on the Rubin GPU. The RTX 5090 SE also has a higher base clock: 1740 MHz versus 700 MHz, a 2.49x advantage. Boost clocks are closer, with the RTX 5090 SE at 2377 MHz versus 2267 MHz, a 1.05x lead.

The RTX 5090 SE supports graphics APIs fully, including DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The Rubin GPU lists N/A for all graphics APIs, confirming its server-oriented role without display or rasterization focus. The RTX 5090 SE's dual-slot form factor and display outputs make it suitable for workstation and consumer use, while the Rubin GPU's SXM Module form factor and lack of outputs target data center deployment.

Neither GPU has recorded benchmark scores or nearest rivals in the database. Both share the same percentile rank (50th) and average benchmark score (0), indicating that no performance measurements have been logged. The wins count is 0 for each, meaning no head-to-head comparison entries exist yet. The analysis therefore draws from the listed architectural and specification data only.

The Verdict

The data indicates that the NVIDIA Rubin GPU is the superior compute device in raw throughput metrics. It delivers 130.0 TFLOPS FP32, 260.0 TFLOPS FP16, and 22.1 TB/s memory bandwidth, all far ahead of the RTX 5090 SE. It also has double the shading units (28,672 versus 14,080), double the TMUs (896 versus 440), and double the tensor cores (896 versus 440). Memory capacity is 288 GB versus 24 GB, a 12x advantage. These figures align with a server accelerator intended for large-scale parallel workloads, neural network training, and high-bandwidth data movement.

The RTX 5090 SE is the appropriate choice for traditional GPU tasks that require rasterization and display output. It has 160 ROPs versus 24, a pixel rate of 380.3 GPixel/s versus 54.41 GPixel/s, and full support for DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. Its dual-slot form factor, 267 mm length, and display outputs (1x HDMI 2.1b, 3x DisplayPort 2.1b) make it a standard graphics card. Its power draw of 500 W with a 900 W suggested PSU is within the range of high-end consumer hardware, whereas the Rubin GPU's 2300 W TDP and 2700 W suggested PSU place it firmly in data center infrastructure.

The process node difference matters. The Rubin GPU uses 3 nm versus 5 nm, enabling a transistor density of 230.8M per mm² versus 122.9M per mm². This allows 336,000 million transistors on a 1456 mm² die, versus 92,200 million on 750 mm². The smaller node gives Rubin a denser, more power-efficient transistor layout per area, though total power consumption is much higher due to the massive scale.

Neither GPU has active benchmark scores in the database, so there is no direct measured performance comparison. The recorded specifications favor the Rubin GPU for compute and memory-bound tasks, and the RTX 5090 SE for graphics rendering and client-side use. The choice depends on workload: the Rubin GPU for server or HPC deployments, the RTX 5090 SE for desktop graphics, content creation, or any application requiring display output and graphics API compatibility.

Specification Differences

| Specification | NVIDIA GeForce RTX 5090 SE | NVIDIA Rubin GPU |

| --- | --- | --- |

| Architecture | Blackwell 2.0 | Rubin |

| Chip | GB202 | GR100 |

| Process node | 5 nm | 3 nm |

| Transistors | 92,200 million | 336,000 million |

| Die size | 750 mm² | 1456 mm² |

| Transistor density | 122.9M / mm² | 230.8M / mm² |

| Base clock | 1740 MHz | 700 MHz |

| Boost clock | 2377 MHz | 2267 MHz |

| Memory clock | 1750 MHz 28 Gbps effective | 2695 MHz 10.8 Gbps effective |

| Memory size | 24 GB | 288 GB |

| Memory type | GDDR7 | HBM4 |

| Memory bus width | 384 bit | 16384 bit |

| Memory bandwidth | 1.34 TB/s | 22.1 TB/s |

| Shading units | 14080 | 28672 |

| TMUs | 440 | 896 |

| ROPs | 160 | 24 |

| Tensor cores | 440 | 896 |

| Pixel rate | 380.3 GPixel/s | 54.41 GPixel/s |

| Texture rate | 1,045.9 GTexel/s | 2,031.2 GTexel/s |

| FP32 | 66.94 TFLOPS | 130.0 TFLOPS |

| FP16 | 66.94 TFLOPS (1:1) | 260.0 TFLOPS (2:1) |

| TDP | 500 W | 2300 W |

| Slot width | Dual-slot | SXM Module |

| Suggested PSU | 900 W | 2700 W |

| Bus interface | PCIe 5.0 x16 | PCIe 6.0 x16 |

| Display outputs | 1x HDMI 2.1b3x DisplayPort 2.1b | No outputs |

| DirectX | 12 Ultimate (12_2) | N/A |

| OpenGL | 4.6 | N/A |

| Vulkan | 1.4 | N/A |

| Length | 267 mm 10.5 inches | Not recorded |

| Height | 111 mm 4.4 inches | Not recorded |

| Width | 40 mm 1.6 inches | Not recorded |

| Power connector | 1x 16-pin | Not recorded |

| Launch MSRP | 1,499 USD | Not recorded |