NVIDIA B300 SXM6 AC vs NVIDIA GeForce RTX 5090 Comparison

Where Each One Wins

The recorded data shows a clear division of roles between these two NVIDIA accelerators. The NVIDIA B300 SXM6 AC wins the only shared benchmark, Geekbench OpenCL, with a score of 369,831 against 334,370 for the RTX 5090, a 10.6% advantage. That single victory, however, tells only part of the story because the two products are built for entirely different environments.

The B300 SXM6 AC is a server compute module. It has no display outputs, no graphics API support (DirectX, OpenGL, and Vulkan are all listed as N/A), and it is designed for the SXM form factor with a 1100 W power envelope and a 1500 W suggested PSU. Its benchmark presence is limited to a single OpenCL result, which reflects its role as a data center accelerator where raw compute throughput matters more than graphics rendering. The RTX 5090, by contrast, is a dual-slot consumer graphics card with 1x HDMI 2.1b and 3x DisplayPort 2.1b outputs, full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support, and a 575 W power draw.

The database contains ten benchmark results for the RTX 5090 spanning DirectX 9, 10, 11, and 12, 3DMark Steel Nomad, Geekbench OpenCL and Vulkan, Passmark 2D, 3D, and GPU compute. The B300 has only the one OpenCL entry. This asymmetry is itself informative: the B300 is not designed for the graphics workloads that dominate the RTX 5090's benchmark suite.

The percentile data reinforces the positioning. The B300 sits at the 100th percentile among all GPUs in the database, while the RTX 5090 sits at the 92nd percentile. The B300's nearest rivals are other data center parts: the NVIDIA B200 (7% behind), the H200 NVL (10.4% behind), the AMD Instinct MI300X (16.3% behind), and the L40S (25% behind). The RTX 5090's nearest rivals, by contrast, are the Tesla P100 variants and mobile/workstation parts, a much lower performance tier. The RTX 5090's average benchmark score of 79,842 is dragged down by its many lower-scoring graphics tests, while the B300's average equals its single OpenCL score.

In short, the B300 wins where raw compute density is the goal, and the RTX 5090 wins where graphics, display output, and API compatibility are required.

FAQ

Q: Which GPU has the higher Geekbench OpenCL score?

A: The NVIDIA B300 SXM6 AC scores 369,831, which is 10.6% higher than the RTX 5090's 334,370 in the same test.

Q: How much memory does each card have, and what type?

A: The B300 SXM6 AC has 288 GB of HBM3e memory on a 8192-bit bus with 8.19 TB/s bandwidth. The RTX 5090 has 32 GB of GDDR7 memory on a 512-bit bus with 1.79 TB/s bandwidth.

Q: What are the power requirements for each card?

A: The B300 SXM6 AC has a 1100 W TDP and a suggested PSU of 1500 W. The RTX 5090 has a 575 W TDP and a suggested PSU of 950 W, using a single 16-pin power connector.

Q: Does the B300 support DirectX or Vulkan?

A: No. The B300 lists DirectX, OpenGL, and Vulkan as N/A. The RTX 5090 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: Which card has higher FP32 compute?

A: The RTX 5090 delivers 104.8 TFLOPS FP32, while the B300 delivers 76.99 TFLOPS. The RTX 5090 also has more shading units (21,760 vs. 18,944) and higher clock speeds.

Q: What is the launch date and MSRP of each product?

A: The B300 SXM6 AC was released on September 10, 2025, with no launch MSRP recorded. The RTX 5090 was released on January 29, 2025, with a launch MSRP of 1,999 USD.

Head-to-Head Benchmarks

The only direct benchmark comparison available in the database is Geekbench OpenCL. The B300 SXM6 AC records 369,831 points, and the RTX 5090 records 334,370 points. The B300 leads by 10.6%. That margin places the B300 ahead of its own nearest rival, the B200, by 7%, and ahead of the H200 NVL by 10.4%. The RTX 5090's OpenCL result is closer to its own rival tier: the Tesla P100 PCIe 16 GB is only 0.3% behind the RTX 5090's average score, and the AMD Radeon Pro Vega 64X is 1.4% ahead.

The RTX 5090's other benchmark results have no B300 counterpart, but they show where the consumer card excels. In Passmark G3D, it scores 39,650. In Passmark GPU compute, it scores 26,756. Its 3DMark Steel Nomad DX12 result is 18,355. The Passmark DirectX tests show strong legacy API performance: 395 in DirectX 9, 341 in DirectX 11, 226 in DirectX 10, and 185 in DirectX 12. Passmark G2D scores 1,413, and Geekbench Vulkan scores 376,728, which actually exceeds its own OpenCL score of 334,370.

The B300's single result is a compute-only measurement, and it beats the RTX 5090's OpenCL result by a substantial margin. But the B300's advantage narrows when compared to the RTX 5090's Vulkan score: 369,831 versus 376,728, meaning the RTX 5090 is actually 1.9% ahead in Vulkan, though that is a different API and not a direct head-to-head comparison.

The data shows a 1-0 win count in favor of the B300 for shared tests, but the practical interpretation is that these cards should not be cross-shopped. The B300's 288 GB HBM3e memory and 8.19 TB/s bandwidth are server-class resources, while the RTX 5090's 32 GB GDDR7 and 1.79 TB/s bandwidth are suited for high-end desktop workloads.

Specification Differences

The two cards diverge sharply on almost every measurable specification. The B300 SXM6 AC uses the GB110 chip on the Blackwell Ultra architecture, while the RTX 5090 uses the GB202 chip on Blackwell 2.0. Both are fabricated on a 5 nm process at TSMC, but the B300 packs 208,000 million transistors on a 1628 mm² die, giving a transistor density of 127.8M per mm². The RTX 5090 has 92,200 million transistors on a 750 mm² die, with a density of 122.9M per mm².

Clock speeds favor the RTX 5090. Its base clock is 2017 MHz and boost clock is 2407 MHz, compared to the B300's 1665 MHz base and 2032 MHz boost. Memory clocks also differ: the B300 runs at 2000 MHz (8 Gbps effective), while the RTX 5090 runs at 1750 MHz (28 Gbps effective). The effective memory speed is much higher on the RTX 5090, but the B300's bus width of 8192 bit versus 512 bit and its HBM3e stack deliver far more total bandwidth.

The RTX 5090 has more shading units (21,760 vs. 18,944), more TMUs (680 vs. 592), and far more ROPs (176 vs. 24). It also has 170 dedicated RT cores, while the B300 lists none. Both have tensor cores, with the RTX 5090 at 680 and the B300 at 592. Pixel rate heavily favors the RTX 5090 at 423.6 GPixel/s versus 48.77 GPixel/s, and texture rate favors it as well at 1,636.8 GTexel/s versus 1,202.9 GTexel/s. FP32 and FP16 compute both favor the RTX 5090 at 104.8 TFLOPS versus 76.99 TFLOPS.

Power and physical design are also opposed. The B300 is an SXM module with a 1100 W TDP and no power connector listed, while the RTX 5090 is a dual-slot card with a 1x 16-pin connector and a 575 W TDP. The B300 has no display outputs and no dimensions recorded; the RTX 5090 measures 304 mm in length, 137 mm in height, and 40 mm in width. The B300 uses PCIe 6.0 x16, while the RTX 5090 uses PCIe 5.0 x16.

Architecture Differences

The B300 SXM6 AC belongs to the Server Blackwell (Bxx) generation, built on the Blackwell Ultra architecture with the GB110 chip. The RTX 5090 belongs to the GeForce 50 series, built on Blackwell 2.0 with the GB202 chip. Both share the same 5 nm TSMC process and the same foundry, but the B300's die is more than twice the size and holds more than twice the transistor count.

Memory architecture is the largest divergence. The B300 uses HBM3e across an 8192-bit bus, yielding 8.19 TB/s of bandwidth and 288 GB of capacity. The RTX 5090 uses GDDR7 across a 512-bit bus, yielding 1.79 TB/s and 32 GB. The B300's memory subsystem is designed for massive data center workloads where capacity and bandwidth dominate. The RTX 5090's GDDR7 is optimized for lower latency and higher clock speeds in a consumer context.

The RTX 5090 is the only one of the two with ray tracing cores (170 of them) and full graphics API support, including DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The B300 has no RT cores listed and no API support, reinforcing its role as a compute-only accelerator. Both have tensor cores, but the B300's are configured for server workloads, while the RTX 5090's tensor cores support consumer features like DLSS and ray tracing acceleration.

The B300's predecessor is listed as Server Hopper, and its successor is Server Rubin. The RTX 5090's predecessor is GeForce 40, and its successor is GeForce 60. The B300 is an active production part released in September 2025, while the RTX 5090 is also active, released in January 2025.

The B300's 208,000 million transistors and 1628 mm² die represent the largest silicon in this comparison, and its 1100 W TDP reflects that scale. The RTX 5090, at 92,200 million transistors and 750 mm², is a large consumer GPU in its own right but operates in a different thermal and power class. The transistor densities are close (127.8M vs. 122.9M per mm²), indicating similar manufacturing maturity, but the B300 simply uses far more silicon.

Architecturally, the B300 is a data center accelerator with no graphics path, while the RTX 5090 is a full-featured graphics card with compute capabilities. The benchmark data reflects this: the B300 posts one strong OpenCL score, while the RTX 5090 shows a broad profile across 2D, 3D, compute, and ray-traced workloads.