NVIDIA GeForce RTX 5090 vs NVIDIA Rubin GPU Comparison

NVIDIA GeForce RTX 5090 vs NVIDIA Rubin GPU: the database shows two very different NVIDIA parts. The RTX 5090 is a retail graphics card with a complete benchmark record. The Rubin GPU is a server compute module with no recorded benchmark scores. Direct head-to-head results are absent, so the comparison relies on architecture, specifications, and the measured performance of the RTX 5090 alone.

Head-to-Head Benchmarks

The database contains no head-to-head benchmark entries for these two products. The RTX 5090 has ten recorded benchmark scores, while the Rubin GPU has none. This absence is itself informative. The RTX 5090’s average benchmark score across all tests is 79,842, placing it in the 92nd percentile of all GPUs in the database. Its nearest rivals in the database are the NVIDIA Tesla P100 PCIe 16 GB with an average score of 79,605 (0.3% behind), the Tesla P100 PCIe 12 GB at 79,396 (0.6% behind), the AMD Radeon RX 6850M XT at 78,940 (1.1% behind), and the AMD Radeon Pro Vega 64X at 80,959 (1.4% ahead). These deltas are small, indicating the RTX 5090 sits in a dense performance cluster at the top of the database.

Looking at individual tests, the RTX 5090 delivers its strongest absolute result in Geekbench Vulkan with a score of 376,728, followed by Geekbench OpenCL at 334,370. In 3DMark Steel Nomad DX12, it posts 18,355. PassMark G3D gives 39,650, and PassMark GPU Compute yields 26,756. The lower-level PassMark DirectX tests show 395 for DX9, 341 for DX11, 226 for DX10, and 185 for DX12. PassMark G2D records 1,413. These numbers establish a performance profile for the RTX 5090, but without Rubin scores, no direct percentage comparison can be made.

The Rubin GPU’s benchmark section is empty, and its average benchmark score is 0. It holds the 50th percentile in the database, a default position that reflects no measured data rather than a performance level. The database records zero wins for both sides in head-to-head comparisons. Any quantitative statement about Rubin’s performance relative to the RTX 5090 is not supported by measured results. The only recourse is to examine what the hardware specifications imply.

The Verdict

The data indicates a clear split in intended use. The RTX 5090 is a finished consumer product with extensive benchmark coverage. Its 92nd percentile rank and average score of 79,842 show it as a top-tier graphics card in the database. The Rubin GPU, with no benchmarks and a 50th percentile placeholder, cannot be ranked on measured performance. The recorded production status for both is Active, but the Rubin’s release date is set at the end of 2025, while the RTX 5090’s release date is in early 2025.

From the data, the RTX 5090 is the only one of the two that can be assessed through actual test scores. It belongs to the GeForce 50-series and the Blackwell 2.0 architecture, targeting the desktop graphics segment. The Rubin GPU belongs to the Server Rubin generation, architecture Rubin, with no display outputs and a server module form factor. The database shows the Rubin as a compute-oriented device, not a graphics card intended for benchmark-style testing.

The RTX 5090’s nearest rivals are older Tesla and AMD parts, with deltas under 1.5%. This suggests the RTX 5090’s measured performance is well established. The Rubin GPU has no rivals listed, reinforcing its status as a part without comparable measured data. For any user seeking a graphics card with verified benchmark results, the RTX 5090 is the only option supported by the database. For a server compute module, the Rubin’s specifications indicate a different purpose, but its performance remains unquantified.

Architecture Differences

The RTX 5090 uses the GB202 chip based on Blackwell 2.0 architecture, fabricated on a 5 nm process at TSMC. The Rubin GPU uses the GR100 chip based on Rubin architecture, fabricated on a 3 nm process, also at TSMC. The process node difference is significant: 5 nm versus 3 nm. Transistor counts differ dramatically. The RTX 5090 has 92,200 million transistors on a 750 mm² die, giving a transistor density of 122.9 million per square millimeter. The Rubin GPU has 336,000 million transistors on a 1,456 mm² die, with a density of 230.8 million per square millimeter. The Rubin’s die is nearly double the size and holds more than three times the transistor count.

Ray tracing cores are present on the RTX 5090 with 170 units. The Rubin GPU lists no ray tracing cores in the database, with a null value. Tensor cores appear on both: 680 on the RTX 5090 and 896 on the Rubin. Shading units differ, with 21,760 on the RTX 5090 and 28,672 on the Rubin. Texture mapping units number 680 on the RTX 5090 and 896 on the Rubin. Raster output units show a stark contrast: 176 on the RTX 5090 versus only 24 on the Rubin. This suggests the Rubin is not designed for traditional rasterization workloads.

The memory subsystem reflects different design goals. The RTX 5090 uses GDDR7 memory with 32 GB capacity, a 512-bit bus, and 1.79 TB/s bandwidth. The Rubin uses HBM4 memory with 288 GB capacity, a 16,384-bit bus, and 22.1 TB/s bandwidth. The Rubin’s memory bandwidth is over twelve times higher, and its capacity is nine times larger. The memory clock differs: the RTX 5090 runs at 1750 MHz (28 Gbps effective), while the Rubin runs at 2695 MHz (10.8 Gbps effective). The effective rate is lower on the Rubin because HBM4 uses a different signaling scheme, but the massive bus width compensates.

Compute rates show the Rubin’s server orientation. The RTX 5090 delivers 104.8 TFLOPS for both FP32 and FP16, a 1:1 ratio. The Rubin delivers 130.0 TFLOPS FP32 and 260.0 TFLOPS FP16, a 2:1 ratio. The FP16 advantage on the Rubin indicates an emphasis on mixed-precision compute. Pixel rates are nearly opposite: the RTX 5090 reaches 423.6 GPixel/s, while the Rubin manages only 54.41 GPixel/s. Texture rates are closer, with 1,636.8 GTexel/s on the RTX 5090 and 2,031.2 GTexel/s on the Rubin.

Specification Differences

The two products differ in nearly every measured specification. Process node: 5 nm for the RTX 5090, 3 nm for the Rubin. Transistors: 92,200 million versus 336,000 million. Die size: 750 mm² versus 1,456 mm². Transistor density: 122.9M per mm² versus 230.8M per mm². Base clock: 2017 MHz versus 700 MHz. Boost clock: 2407 MHz versus 2267 MHz. Memory clock: 1750 MHz versus 2695 MHz. Memory size: 32 GB versus 288 GB. Memory type: GDDR7 versus HBM4. Bus width: 512 bit versus 16,384 bit. Bandwidth: 1.79 TB/s versus 22.1 TB/s.

Shading units: 21,760 versus 28,672. TMUs: 680 versus 896. ROPs: 176 versus 24. Tensor cores: 680 versus 896. Pixel rate: 423.6 GPixel/s versus 54.41 GPixel/s. Texture rate: 1,636.8 GTexel/s versus 2,031.2 GTexel/s. FP32: 104.8 TFLOPS versus 130.0 TFLOPS. FP16: 104.8 TFLOPS versus 260.0 TFLOPS. TDP: 575 W versus 2300 W. Slot width: Dual-slot versus SXM Module. Power connector: 1x 16-pin versus none listed. Suggested PSU: 950 W versus 2700 W. Bus interface: PCIe 5.0 x16 versus PCIe 6.0 x16. Display outputs: 1x HDMI 2.1b and 3x DisplayPort 2.1b versus no outputs. API support: DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4 on the RTX 5090, versus N/A for all three on the Rubin.

Dimensions are recorded only for the RTX 5090: 304 mm length, 137 mm height, 40 mm width. The Rubin has no listed dimensions. Release dates differ: the RTX 5090 launched in January 2025, the Rubin is dated for December 2025. The RTX 5090 has a launch MSRP of 1,999 USD. The Rubin has no launch MSRP listed. The Rubin’s predecessor is Server Blackwell, while the RTX 5090’s predecessor is GeForce 40. The RTX 5090 has a successor listed as GeForce 60, while the Rubin has none. The RTX 5090 belongs to the GeForce 50-series, while the Rubin has no series designation.

FAQ

Q: Why does the Rubin GPU have no benchmark scores in the database?

A: The database lists an empty benchmarks array for the Rubin GPU, an average benchmark score of 0, and a 50th percentile rank. This indicates no measured test results exist for this product in the database.

Q: Which GPU has higher FP32 compute?

A: The Rubin GPU lists 130.0 TFLOPS FP32, while the RTX 5090 lists 104.8 TFLOPS FP32. The Rubin is approximately 24% higher in raw FP32 throughput.

Q: How do the memory bandwidth figures compare?

A: The Rubin GPU uses HBM4 with a 16,384-bit bus and 22.1 TB/s bandwidth. The RTX 5090 uses GDDR7 with a 512-bit bus and 1.79 TB/s bandwidth. The Rubin’s bandwidth is roughly twelve times higher.

Q: What is the transistor density difference?

A: The RTX 5090 has 122.9 million transistors per square millimeter on a 5 nm process. The Rubin GPU has 230.8 million transistors per square millimeter on a 3 nm process. The Rubin’s density is about 88% higher.

Q: Does the Rubin GPU have display outputs?

A: The database records no outputs for the Rubin GPU. The RTX 5090 has 1x HDMI 2.1b and 3x DisplayPort 2.1b outputs.

Q: What is the power consumption difference?

A: The RTX 5090 has a TDP of 575 W with a suggested PSU of 950 W. The Rubin GPU has a TDP of 2300 W with a suggested PSU of 2700 W. The Rubin draws four times the power.

Where Each One Wins

The RTX 5090 wins in any scenario that requires verified benchmark performance. Its measured scores across DirectX, OpenCL, Vulkan, and compute workloads place it in the 92nd percentile. The nearest rivals in the database are within 1.5% of its average score, indicating it sits at the top of a competitive cluster. For rasterization, the RTX 5090’s 176 ROPs and 423.6 GPixel/s pixel rate vastly exceed the Rubin’s 24 ROPs and 54.41 GPixel/s. The RTX 5090 also supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, while the Rubin lists N/A for all graphics APIs. Any workload that relies on conventional graphics rendering, display output, or consumer-facing benchmarks favors the RTX 5090.

The Rubin GPU wins on raw compute capacity and memory scale. Its 28,672 shading units, 896 tensor cores, and 260.0 TFLOPS FP16 throughput exceed the RTX 5090’s corresponding figures. The 288 GB HBM4 memory with 22.1 TB/s bandwidth provides a memory environment suited to large data sets, far beyond the RTX 5090’s 32 GB and 1.79 TB/s. The Rubin’s 3 nm process and 336,000 million transistors indicate a more advanced silicon implementation. Its PCIe 6.0 x16 interface is a generation ahead of the RTX 5090’s PCIe 5.0 x16. For server compute, tensor operations, or FP16-heavy workloads, the Rubin’s specifications are superior. The absence of display outputs and graphics API support confirms it is not meant for desktop graphics.

The database does not provide measured results for the Rubin, so any performance advantage is inferred from specifications, not verified scores. The RTX 5090 remains the only product of the two with confirmed benchmark data. The Rubin’s higher FP32 and FP16 throughput, larger memory pool, and higher bandwidth suggest it targets compute workloads where the RTX 5090’s graphics-oriented design would be limiting. The RTX 5090 wins in every scenario requiring tested performance, graphics APIs, or display connectivity. The Rubin wins in scenarios prioritizing raw compute, memory capacity, and bandwidth, assuming its specifications translate to real-world performance, which the database does not currently confirm.