NVIDIA GB10 vs NVIDIA GeForce RTX 5090 SE Comparison

NVIDIA GB10 and NVIDIA GeForce RTX 5090 SE both use the Blackwell 2.0 architecture and a 5 nm process from TSMC, but they are engineered for entirely different workloads. The GB10 is a server-focused integrated graphics processor with a 140 W power envelope, while the RTX 5090 SE is a dual-slot discrete GeForce card with a 500 W TDP. The recorded data shows the GB10 has an average benchmark score of 117,393 across two tests, while the RTX 5090 SE has no recorded benchmarks and an average score of 0. This gap in available measurements shapes the analysis, but the architectural differences provide clear direction.

FAQ

Q: What is the GB10’s benchmark performance relative to its nearest rivals?

A: The GB10 scores 120,137 in Geekbench OpenCL and 114,648 in Geekbench Vulkan, yielding an average of 117,393. It sits 0.3% above the NVIDIA RTX 4000 SFF Ada Generation (117,088), 1.3% below the AMD Radeon PRO W7700 (118,976), 2.6% above the NVIDIA Tesla V100 SXM2 16 GB (114,395), and 3% above the NVIDIA RTX A5500 Mobile (113,944). Its percentile rank among all GPUs is 95.

Q: Does the RTX 5090 SE have any benchmark scores in the database?

A: No. The database lists an empty benchmarks array for the RTX 5090 SE, an average score of 0, and a percentile rank of 50. No head-to-head benchmark results exist between the GB10 and the RTX 5090 SE.

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

A: The GB10 has 128 GB of LPDDR5X on a 256-bit bus, delivering 273.2 GB/s bandwidth. The RTX 5090 SE has 24 GB of GDDR7 on a 384-bit bus, delivering 1.34 TB/s bandwidth. The GB10’s capacity is roughly 5.3 times larger, while the RTX 5090 SE’s bandwidth is about 4.9 times higher.

Q: What is the transistor count and die size for each chip?

A: The GB10 uses the GB20B chip with a 382 mm² die and unknown transistor count. The RTX 5090 SE uses the GB202 chip with 92,200 million transistors on a 750 mm² die, giving a transistor density of 122.9 million per mm². The RTX 5090 SE’s die is approximately 1.96 times larger.

Q: Which card supports modern graphics APIs?

A: The RTX 5090 SE supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The GB10 lists N/A for DirectX, OpenGL, and Vulkan, indicating no consumer graphics API support in the recorded data.

Q: What are the power connector requirements?

A: The GB10 has no power connectors and a suggested PSU of 300 W. The RTX 5090 SE requires a single 16-pin connector and a suggested PSU of 900 W. The TDP is 140 W for the GB10 and 500 W for the RTX 5090 SE.

Architecture Differences

The GB10 and RTX 5090 SE share the same Blackwell 2.0 architecture and 5 nm TSMC process, but their implementations diverge sharply. The GB10’s chip is the GB20B, a 382 mm² die with an unknown transistor count. The RTX 5090 SE uses the GB202, a 750 mm² die with 92,200 million transistors and a density of 122.9 million per mm². The die area difference of 368 mm² indicates a much larger silicon footprint for the GeForce part.

The GB10 is an integrated graphics processor (IGP) with a slot width of IGP, while the RTX 5090 SE is a dual-slot discrete card. This physical distinction aligns with their intended deployments: the GB10 targets server environments where space and power are constrained, while the RTX 5090 SE occupies a standard expansion slot in a desktop.

Compute resources differ substantially. The GB10 has 6,144 shading units, 384 texture mapping units, 48 raster output units, 48 ray tracing cores, and 384 tensor cores. The RTX 5090 SE has 14,080 shading units, 440 TMUs, 160 ROPs, 110 RT cores, and 440 tensor cores. The shading unit count is 2.29 times higher on the RTX 5090 SE, and the ROP count is 3.33 times higher.

Clock behavior also separates the two. The GB10 has a base clock of 1665 MHz and a boost clock of 2418 MHz. The RTX 5090 SE has a base clock of 1740 MHz and a boost clock of 2377 MHz. The base clock is 75 MHz higher on the RTX 5090 SE, but the boost clock is 41 MHz higher on the GB10. Memory clocks show a larger gap: the GB10 runs at 1067 MHz with 8.5 Gbps effective, while the RTX 5090 SE runs at 1750 MHz with 28 Gbps effective.

The GB10’s memory type is LPDDR5X with a 256-bit bus, whereas the RTX 5090 SE uses GDDR7 with a 384-bit bus. The bus width difference of 128 bits contributes to the bandwidth gap. The GB10’s bandwidth is 273.2 GB/s; the RTX 5090 SE’s is 1.34 TB/s, which is 4.9 times higher.

Power delivery reflects the performance targets. The GB10 has a 140 W TDP, no power connectors, and a suggested PSU of 300 W. The RTX 5090 SE has a 500 W TDP, one 16-pin connector, and a suggested PSU of 900 W. The TDP difference is 360 W, indicating a much larger thermal budget for the GeForce card.

Display outputs also differ. The GB10 has only 1x HDMI. The RTX 5090 SE has 1x HDMI 2.1b and 3x DisplayPort 2.1b. The GB10 lists N/A for DirectX, OpenGL, and Vulkan, while the RTX 5090 SE supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

The Verdict

The data indicates the GB10 is for server-side compute tasks that prioritize memory capacity and low power. Its 128 GB of LPDDR5X memory, 140 W TDP, and lack of consumer API support point to workloads like large-scale inference or data processing where graphics output is secondary. The 95th percentile rank among all GPUs and an average score of 117,393 confirm it competes with workstation-class accelerators.

The RTX 5090 SE is for desktop graphics and gaming. It has 24 GB of GDDR7, 1.34 TB/s bandwidth, 66.94 TFLOPS FP32, and full DirectX 12 Ultimate support. The 500 W TDP and 900 W suggested PSU indicate a high-end consumer card. Its lack of recorded benchmarks means no direct performance comparison is possible, but the architectural numbers show a clear advantage in raw throughput.

Users needing large memory pools and lower power consumption should select the GB10. Users needing high bandwidth, high shading unit counts, and modern graphics APIs should select the RTX 5090 SE. The GB10’s release date of October 14, 2025, and the RTX 5090 SE’s December 31, 2025, place both in the same generation era, but their design goals are opposite.

Specification Differences

The GB10 and RTX 5090 SE differ across every major specification field except architecture, process node, foundry, and bus interface.

• Chip: GB20B vs GB202
• Generation: Server Blackwell (Bxx) vs GeForce 50
• Die size: 382 mm² vs 750 mm²
• Transistors: unknown vs 92,200 million
• Transistor density: not listed vs 122.9M / mm²
• Base clock: 1665 MHz vs 1740 MHz
• Boost clock: 2418 MHz vs 2377 MHz
• Memory clock: 1067 MHz, 8.5 Gbps effective vs 1750 MHz, 28 Gbps effective
• Memory size: 128 GB vs 24 GB
• Memory type: LPDDR5X vs GDDR7
• Memory bus width: 256 bit vs 384 bit
• Memory bandwidth: 273.2 GB/s vs 1.34 TB/s
• Shading units: 6,144 vs 14,080
• TMUs: 384 vs 440
• ROPs: 48 vs 160
• RT cores: 48 vs 110
• Tensor cores: 384 vs 440
• Pixel rate: 116.1 GPixel/s vs 380.3 GPixel/s
• Texture rate: 928.5 GTexel/s vs 1,045.9 GTexel/s
• FP32: 29.71 TFLOPS vs 66.94 TFLOPS
• FP16: 29.71 TFLOPS (1:1) vs 66.94 TFLOPS (1:1)
• TDP: 140 W vs 500 W
• Slot width: IGP vs Dual-slot
• Power connectors: None vs 1x 16-pin
• Suggested PSU: 300 W vs 900 W
• Display outputs: 1x HDMI vs 1x HDMI 2.1b, 3x DisplayPort 2.1b
• DirectX: N/A vs 12 Ultimate (12_2)
• OpenGL: N/A vs 4.6
• Vulkan: N/A vs 1.4
• Dimensions: 150 mm x 51 mm x 150 mm vs 267 mm x 111 mm x 40 mm
• Release date: 2025-10-14 vs 2025-12-31
• Predecessor: Server Hopper vs GeForce 40
• Successor: Server Rubin vs GeForce 60
• Launch MSRP: 3,999 USD vs 1,499 USD

Head-to-Head Benchmarks

No head-to-head benchmark results exist in the database for these two GPUs. The GB10 has two recorded tests: Geekbench OpenCL at 120,137 and Geekbench Vulkan at 114,648. The RTX 5090 SE has no benchmark entries, so a direct numerical comparison of measured performance is impossible.

The GB10’s average score of 117,393 places it in the 95th percentile of all GPUs. Its nearest rival is the NVIDIA RTX 4000 SFF Ada Generation, which scores 117,088 and trails by 0.3%. The AMD Radeon PRO W7700 leads the GB10 by 1.3% with a score of 118,976. The GB10 outperforms the NVIDIA Tesla V100 SXM2 16 GB (114,395) by 2.6% and the NVIDIA RTX A5500 Mobile (113,944) by 3%.

For the RTX 5090 SE, the database shows no rivals, no scores, and an average of 0. The percentile rank of 50 is a neutral placeholder, not a measured performance indicator. Any discussion of its benchmark performance must rely on specification-derived expectations rather than recorded results.

The closest proxy for the RTX 5090 SE’s potential is its FP32 throughput. At 66.94 TFLOPS, it is 2.25 times higher than the GB10’s 29.71 TFLOPS. Pixel rate is 380.3 GPixel/s versus 116.1 GPixel/s, a 3.27 times advantage. Texture rate is 1,045.9 GTexel/s versus 928.5 GTexel/s, a 1.13 times advantage. These numbers suggest the RTX 5090 SE would dominate in compute-heavy workloads, but no measured score confirms this.

Where Each One Wins

The GB10 wins in memory capacity and power efficiency. Its 128 GB of LPDDR5X is 5.33 times larger than the RTX 5090 SE’s 24 GB. Its 140 W TDP is 28% of the RTX 5090 SE’s 500 W. The suggested PSU of 300 W versus 900 W further emphasizes the GB10’s suitability for dense server environments. The GB10 also has a smaller physical footprint: 150 mm by 51 mm by 150 mm versus 267 mm by 111 mm by 40 mm. The GB10 requires no power connectors, simplifying installation.

The RTX 5090 SE wins in nearly every raw performance metric. Memory bandwidth is 1.34 TB/s versus 273.2 GB/s, a 4.9 times advantage. Shading units are 14,080 versus 6,144, a 2.29 times advantage. ROPs are 160 versus 48, a 3.33 times advantage. FP32 is 66.94 TFLOPS versus 29.71 TFLOPS, a 2.25 times advantage. Pixel rate is 380.3 GPixel/s versus 116.1 GPixel/s, a 3.27 times advantage. The RTX 5090 SE also has 110 RT cores versus 48, and 440 tensor cores versus 384.

The RTX 5090 SE supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, enabling consumer gaming and graphics applications. The GB10 has no API support listed, meaning it cannot run standard graphics workloads. Display connectivity also favors the RTX 5090 SE with three DisplayPort outputs versus the GB10’s single HDMI.

The GB10’s release date is earlier (October 14, 2025) than the RTX 5090 SE (December 31, 2025). Both are active in production. The GB10’s predecessor is Server Hopper and successor is Server Rubin, while the RTX 5090 SE’s predecessor is GeForce 40 and successor is GeForce 60. These lineage differences confirm separate product families.

In summary, the GB10 is the choice for memory-bound server tasks with tight power limits. The RTX 5090 SE is the choice for bandwidth-bound graphics and compute tasks with ample power delivery. The absence of benchmark data for the RTX 5090 SE means final performance claims rest on its specification advantage, which is substantial but unmeasured in the current database.