AMD Instinct MI350X vs NVIDIA B300 SXM6 AC Comparison

Head-to-Head Benchmarks

The database contains one recorded benchmark for NVIDIA B300 SXM6 AC: a Geekbench OpenCL score of 369,831. AMD Instinct MI350X has no recorded benchmark scores in the database, so the head-to-head comparison relies on the B300 SXM6 AC's nearest rival data and the architectural specifications of both accelerators.

The NVIDIA B300 SXM6 AC delivers an OpenCL score that places it in the 100th percentile of all GPUs in the database. Its nearest documented rival, NVIDIA B200, averages 345,482 points, which puts the B300 SXM6 AC 7% ahead. The B300 SXM6 AC also leads the NVIDIA H200 NVL by 10.4%, with the H200 NVL averaging 334,891 points. Against AMD Instinct MI300X, the B300 SXM6 AC holds a 16.3% advantage, as the MI300X averages 317,994 points. The NVIDIA L40S trails by 25%, averaging 295,763 points.

In raw compute specifications, the B300 SXM6 AC posts an FP32 throughput of 76.99 TFLOPS, while the MI350X delivers 72.09 TFLOPS. That is a 6.8% gap in favor of the NVIDIA part. Both accelerators maintain a 1:1 FP16 to FP32 ratio, meaning the B300 SXM6 AC also delivers 76.99 TFLOPS for FP16 workloads, while the MI350X manages 72.09 TFLOPS. The MI350X does counter with a higher texture rate at 2,252.8 GTexel/s versus 1,202.9 GTexel/s for the B300 SXM6 AC, a difference of roughly 87%. The B300 SXM6 AC, however, is the only one with a documented pixel rate at 48.77 GPixel/s, while the MI350X lists 0 MPixel/s.

Memory configurations are identical: both use 288 GB of HBM3e with an 8192-bit bus and 8.19 TB/s of bandwidth. Clock behavior differs meaningfully. The B300 SXM6 AC runs a base clock of 1665 MHz and a boost clock of 2032 MHz. The MI350X has a 1000 MHz base and a 2200 MHz boost, so the AMD part has a higher peak clock but a much lower idle or sustained base. The B300 SXM6 AC also carries more shading units at 18,944 versus 16,384 for the MI350X, a 15.6% advantage. The MI350X has more texture mapping units at 1,024 versus 592 for the B300 SXM6 AC. The NVIDIA part includes 24 ROPs and 592 tensor cores, while the MI350X lists no ROPs and no tensor cores in the recorded data.

Power draw figures place the B300 SXM6 AC at 1100 W TDP, with a suggested PSU of 1500 W. The MI350X draws 1000 W TDP with a suggested PSU of 1400 W. The B300 SXM6 AC uses a PCIe 6.0 x16 interface, while the MI350X uses PCIe 5.0 x16. Form factors differ as well: the B300 SXM6 AC mounts in an SXM module, and the MI350X is an OAM module. Neither accelerator has display outputs, and both have no applicable DirectX, OpenGL, or Vulkan support in the database.

The Verdict

The recorded data favors the NVIDIA B300 SXM6 AC in every measurable performance category. Its Geekbench OpenCL score of 369,831 places it at the 100th percentile of all GPUs in the database, and it leads each of its four nearest rivals by margins ranging from 7% to 25%. The AMD Instinct MI350X, by contrast, has no benchmark score and sits at the 50th percentile, which in the database reflects the absence of recorded performance data rather than a measured result.

The B300 SXM6 AC also has a higher FP32 throughput at 76.99 TFLOPS versus 72.09 TFLOPS for the MI350X, a larger shading unit count at 18,944 versus 16,384, and a more advanced bus interface with PCIe 6.0 x16 versus PCIe 5.0 x16. The MI350X does have a higher boost clock at 2200 MHz versus 2032 MHz, but the B300 SXM6 AC compensates with a substantially higher base clock at 1665 MHz versus 1000 MHz. The MI350X also leads in texture fill rate at 2,252.8 GTexel/s versus 1,202.9 GTexel/s, but that does not translate into a higher shading or compute score in the available data.

For buyers selecting strictly on recorded benchmark performance, the B300 SXM6 AC is the clear choice. It is faster in the single measured test, has more compute cores, and delivers higher FP32 and FP16 throughput. The MI350X is not without merit, but its advantages are confined to texture throughput and peak boost clock, neither of which appears in a recorded benchmark result. The B300 SXM6 AC also holds a higher percentile ranking, which indicates that among all GPUs in the database, it sits at the top, while the MI350X sits at the median.

Where Each One Wins

The NVIDIA B300 SXM6 AC wins in general compute workloads as measured by Geekbench OpenCL. Its score of 369,831 puts it ahead of the nearest rival B200 by 7%, and ahead of the MI300X by 16.3%. The B300 SXM6 AC also wins in FP32 and FP16 compute throughput, delivering 76.99 TFLOPS in both, which surpasses the MI350X's 72.09 TFLOPS. The B300 SXM6 AC has more shading units at 18,944, which supports higher parallel throughput in shader-heavy tasks. It also includes tensor cores, with 592 available, which the MI350X does not list in the database, suggesting an advantage for workloads that utilize tensor operations.

The AMD Instinct MI350X wins in texture throughput, with 2,252.8 GTexel/s versus 1,202.9 GTexel/s for the B300 SXM6 AC. That makes the MI350X better suited for texture-bound operations, where the rate of texture mapping unit output is the limiting factor. The MI350X also has a higher boost clock at 2200 MHz versus 2032 MHz, which could provide an edge in single-threaded or lightly threaded workloads that scale with peak clock speed. The MI350X uses a 3 nm process node, while the B300 SXM6 AC uses 5 nm, both from TSMC. The smaller node may offer efficiency advantages, though the database does not record power efficiency metrics beyond TDP, where the MI350X draws 100 W less.

Power consumption is a differentiator. The MI350X has a TDP of 1000 W, while the B300 SXM6 AC draws 1100 W. The suggested PSU is 1400 W for the MI350X and 1500 W for the B300 SXM6 AC. For deployments with strict power limits, the MI350X requires less power headroom, which may allow denser packing or lower cooling demands. The MI350X also has a larger die at 2380 mm² versus 1628 mm² for the B300 SXM6 AC, though the B300 SXM6 AC has a higher transistor density at 127.8M per mm² versus 77.7M per mm² for the MI350X.

FAQ

Q: What is the recorded benchmark score for the NVIDIA B300 SXM6 AC?

A: The B300 SXM6 AC has a Geekbench OpenCL score of 369,831, which places it at the 100th percentile of all GPUs in the database.

Q: Does the AMD Instinct MI350X have any recorded benchmark scores?

A: No. The MI350X has an empty benchmark list and an average benchmark score of 0 in the database.

Q: How does the B300 SXM6 AC compare to its nearest rival in the database?

A: The B300 SXM6 AC is 7% ahead of the NVIDIA B200, 10.4% ahead of the NVIDIA H200 NVL, 16.3% ahead of the AMD Instinct MI300X, and 25% ahead of the NVIDIA L40S.

Q: What memory configuration do both accelerators use?

A: Both the MI350X and the B300 SXM6 AC use 288 GB of HBM3e memory with an 8192-bit bus and 8.19 TB/s bandwidth.

Q: Which accelerator has a higher FP32 throughput?

A: The B300 SXM6 AC has a higher FP32 throughput at 76.99 TFLOPS, compared to 72.09 TFLOPS for the MI350X.

Q: What are the power requirements for each accelerator?

A: The MI350X has a TDP of 1000 W with a suggested PSU of 1400 W. The B300 SXM6 AC has a TDP of 1100 W with a suggested PSU of 1500 W.

Architecture Differences

The two accelerators use different manufacturing processes. The AMD Instinct MI350X is built on a 3 nm node at TSMC, while the NVIDIA B300 SXM6 AC uses a 5 nm node, also at TSMC. The MI350X has a larger die at 2380 mm², while the B300 SXM6 AC has a smaller die at 1628 mm². Transistor counts differ as well: the B300 SXM6 AC has 208,000 million transistors, while the MI350X has 185,000 million. The B300 SXM6 AC achieves a higher transistor density at 127.8M per mm², versus 77.7M per mm² for the MI350X.

The core architectures are distinct. The MI350X uses AMD's CDNA 4.0 architecture with the MI350 256CU chip, while the B300 SXM6 AC uses NVIDIA's Blackwell Ultra architecture with the GB110 chip. The MI350X belongs to the Instinct (MIx) generation, while the B300 SXM6 AC is part of the Server Blackwell (Bxx) generation. The B300 SXM6 AC lists a production status of Active, and its predecessor is Server Hopper with a successor of Server Rubin. The MI350X lists Radeon Instinct as its predecessor.

Shading unit counts differ significantly. The B300 SXM6 AC has 18,944 shading units, while the MI350X has 16,384. Texture mapping units also differ, with the MI350X having 1,024 TMUs and the B300 SXM6 AC having 592. The B300 SXM6 AC has 24 ROPs and 592 tensor cores, while the MI350X lists no ROPs and no tensor cores. The MI350X has a pixel rate of 0 MPixel/s, while the B300 SXM6 AC has 48.77 GPixel/s.

Clock behavior shows a split. The MI350X has a base clock of 1000 MHz and a boost clock of 2200 MHz. The B300 SXM6 AC has a base clock of 1665 MHz and a boost clock of 2032 MHz. Memory clocks are identical at 2000 MHz with 8 Gbps effective. Both use HBM3e memory with identical capacity, bus width, and bandwidth.

The bus interface differs, with the B300 SXM6 AC using PCIe 6.0 x16 and the MI350X using PCIe 5.0 x16. Form factors also differ: the B300 SXM6 AC is an SXM module, while the MI350X is an OAM module. The MI350X has recorded dimensions of 102 mm in length and 165 mm in width, while the B300 SXM6 AC has no recorded dimensions. Neither accelerator has display outputs, and neither supports DirectX, OpenGL, or Vulkan in the database. The B300 SXM6 AC has a release date of September 2025, while the MI350X has a release date of June 2025.