AMD Radeon Pro W6900X vs NVIDIA GB10 Comparison

The AMD Radeon Pro W6900X and NVIDIA GB10 occupy very different positions in the GPU landscape, with the former being an end-of-life Apple MPX workstation card and the latter an active, low-power server IGP. Benchmark data shows the AMD card leads in both shared tests, but the GB10 counters with a massive memory capacity advantage and a far more efficient architecture. The W6900X wins on raw compute and graphics API performance, while the GB10 is defined by its 128 GB unified memory and 140 W power envelope.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Pro W6900X scores 168,574 on average, placing it in the 97th percentile of all GPUs. The NVIDIA GB10 averages 117,393, which puts it in the 95th percentile.

Q: How much faster is the AMD card in the shared OpenCL test?

A: The W6900X scores 130,035 in Geekbench OpenCL versus 120,137 for the GB10, a lead of 8.2 percent.

Q: What is the margin in the Vulkan test?

A: The AMD card achieves 148,865 in Geekbench Vulkan, while the GB10 manages 114,648. That is a 29.8 percent advantage for the W6900X.

Q: Does the NVIDIA GB10 have more memory than the AMD card?

A: Yes, the GB10 has 128 GB of LPDDR5X, which is four times the 32 GB of GDDR6 found on the W6900X.

Q: Which GPU has the higher boost clock?

A: The NVIDIA GB10 boosts to 2418 MHz, which is higher than the AMD card’s 2171 MHz boost clock.

Q: What is the production status of each product?

A: The AMD Radeon Pro W6900X is end-of-life, while the NVIDIA GB10 is listed as active.

Architecture Differences

The two GPUs are built on fundamentally different architectures and process nodes. The AMD Radeon Pro W6900X uses the Navi 21 chip with RDNA 2.0 architecture, fabricated on TSMC’s 7 nm process. It packs 26,800 million transistors into a 520 mm² die, giving a transistor density of 51.5M per mm². The NVIDIA GB10 uses the GB20B chip with Blackwell 2.0 architecture, built on a smaller 5 nm TSMC process. Its die size is 382 mm², though transistor count is listed as unknown.

The shading engine configuration differs significantly. The AMD card has 5120 shading units, 320 TMUs, and 128 ROPs. The GB10 has more shading units at 6144, more TMUs at 384, but far fewer ROPs at just 48. Ray tracing hardware also differs: the W6900X has 80 RT cores, while the GB10 has 48. The NVIDIA part includes 384 tensor cores; the AMD card has no tensor core field listed.

Memory architecture is another major split. The W6900X uses 32 GB of GDDR6 on a 256-bit bus, delivering 512.0 GB/s of bandwidth. The GB10 uses 128 GB of LPDDR5X on the same 256-bit bus, but bandwidth drops to 273.2 GB/s. Clock behavior also differs: the AMD card has a base of 1825 MHz and boost of 2171 MHz, while the GB10 has a lower base of 1665 MHz but a higher boost of 2418 MHz.

API support is completely different. The AMD card supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The GB10 lists N/A for DirectX, OpenGL, and Vulkan, indicating its server-focused role. The W6900X uses an Apple MPX bus interface, while the GB10 uses PCIe 5.0 x16 and is classified as an IGP with no power connectors.

Head-to-Head Benchmarks

The data includes two direct comparisons, and the AMD Radeon Pro W6900X wins both. In Geekbench OpenCL, the AMD card scores 130,035 against the GB10’s 120,137. That 8.2 percent delta is a clear but not overwhelming margin. The Vulkan test shows a much larger gap: the W6900X scores 148,865 versus 114,648 for the GB10, a 29.8 percent advantage.

Looking at the broader context, the W6900X’s average score of 168,574 puts it 1.5 percent ahead of the NVIDIA RTX 4500 Ada Generation and 2 percent ahead of the RTX A5500. The GB10’s average of 117,393 is nearly tied with the RTX 4000 SFF Ada Generation, which trails by just 0.3 percent. The GB10 is 1.3 percent behind the AMD Radeon PRO W7700 and 2.6 percent ahead of the Tesla V100 SXM2 16 GB.

The Vulkan result is particularly telling. The W6900X’s 29.8 percent lead in that test aligns with its higher pixel rate of 277.9 GPixel/s and texture rate of 694.7 GTexel/s, versus the GB10’s 116.1 GPixel/s and 928.5 GTexel/s. The AMD card also has double the FP16 throughput at 44.46 TFLOPS (2:1) compared to the GB10’s 29.71 TFLOPS (1:1). In raw FP32, however, the GB10 leads with 29.71 TFLOPS versus 22.23 TFLOPS for the AMD card.

Specification Differences

The two GPUs differ across nearly every measurable field. Process node: 7 nm for AMD, 5 nm for NVIDIA. Die size: 520 mm² versus 382 mm². Transistor count: 26,800 million for AMD, unknown for NVIDIA. The AMD card has a base clock of 1825 MHz and boost of 2171 MHz; the GB10 has a base of 1665 MHz and boost of 2418 MHz.

Memory configuration is starkly different. The W6900X has 32 GB GDDR6 at 2000 MHz (16 Gbps effective) with 512.0 GB/s bandwidth. The GB10 has 128 GB LPDDR5X at 1067 MHz (8.5 Gbps effective) with 273.2 GB/s bandwidth. Both use a 256-bit bus, but the AMD card has nearly double the bandwidth.

Compute unit counts also diverge. Shading units: 5120 for AMD, 6144 for NVIDIA. TMUs: 320 versus 384. ROPs: 128 versus 48. RT cores: 80 versus 48. Tensor cores: none listed for AMD, 384 for NVIDIA. Pixel rate: 277.9 GPixel/s for AMD, 116.1 GPixel/s for NVIDIA. Texture rate: 694.7 GTexel/s versus 928.5 GTexel/s. FP32: 22.23 TFLOPS versus 29.71 TFLOPS. FP16: 44.46 TFLOPS (2:1) versus 29.71 TFLOPS (1:1).

Power and physical specs are also distinct. The W6900X has a 300 W TDP and suggests a 700 W PSU, while the GB10 has a 140 W TDP and suggests a 300 W PSU. The AMD card is 267 mm long and 120 mm high; the GB10 is 150 mm long, 51 mm high, and 150 mm wide. The GB10 is an IGP with no power connectors, while the AMD card uses Apple MPX and offers 1x HDMI 2.1 plus 4x Thunderbolt outputs. The GB10 has only 1x HDMI.

Where Each One Wins

The AMD Radeon Pro W6900X wins in every direct benchmark comparison available. Its 29.8 percent Vulkan lead suggests it is the stronger choice for graphics-heavy workloads that use that API. The 8.2 percent OpenCL margin is smaller but still consistent. The card’s higher pixel rate (277.9 GPixel/s) and larger ROP count (128) point to superior rasterization throughput. Its 512.0 GB/s memory bandwidth is nearly double the GB10’s, which helps in bandwidth-bound tasks. The W6900X also supports DirectX 12 Ultimate and Vulkan 1.4, making it viable for modern graphics applications.

The NVIDIA GB10 wins on memory capacity and efficiency. Its 128 GB of LPDDR5X is unmatched by the AMD card’s 32 GB, making it suitable for very large datasets that fit in memory. The GB10 also has a lower TDP of 140 W versus 300 W, and a suggested PSU of 300 W versus 700 W. Its higher FP32 throughput (29.71 TFLOPS) and texture rate (928.5 GTexel/s) indicate strong compute and texture processing capability. The 384 tensor cores give it a hardware advantage for tensor-based workloads, though no benchmark data confirms this. The GB10’s active production status and newer release date (2025-10-14 versus 2021-08-02) also favor it for ongoing availability.

The Verdict

The data supports a clear split: the AMD Radeon Pro W6900X is the performance leader in shared benchmarks, while the NVIDIA GB10 is the capacity and efficiency leader. If the decision rests on measured benchmark scores, the W6900X wins both OpenCL and Vulkan, with a particularly strong 29.8 percent margin in Vulkan. Its 97th percentile ranking versus the GB10’s 95th percentile confirms this hierarchy in the broader GPU pool.

However, the GB10’s 128 GB memory capacity is a decisive feature for workloads that require holding massive models or datasets in memory. The AMD card’s 32 GB is a quarter of that. The GB10’s 140 W TDP and 300 W suggested PSU also make it far more practical for dense server deployments, where the W6900X’s 300 W TDP and 700 W PSU requirement would be a burden.

For users prioritizing raw graphics performance, API support, and memory bandwidth, the W6900X is the data-backed choice. For users prioritizing memory size, power efficiency, and tensor core availability, the GB10 is the logical pick. The GB10’s lack of DirectX, OpenGL, and Vulkan support is a major caveat for graphics workloads, but its server classification and IGP form factor suggest those are not its intended use cases. The launch MSRP difference is notable: the W6900X launched at 5,999 USD, while the GB10 launched at 3,999 USD.