Intel Arc B580 vs NVIDIA H20 NVL16 Comparison

Where Each One Wins

The Intel Arc B580 and NVIDIA H20 NVL16 occupy entirely different segments of the GPU market, and the recorded data makes that split explicit. The Arc B580 is a client-focused graphics card with a full suite of benchmark results across DirectX 10, 11, 12, 9, OpenCL, Vulkan, and compute workloads. The H20 NVL16 is a server accelerator with no display outputs, no graphics API support, and an empty benchmark array in the database.

The Arc B580 wins every recorded graphics benchmark because it is the only one of the two with graphics benchmarks recorded. Its 3DMark Steel Nomad DX12 score of 3068, Geekbench OpenCL score of 92821, and Geekbench Vulkan score of 109672 all stand as measurable outcomes. The H20 NVL16 has no recorded scores in any of these tests, so the data cannot show a single head-to-head win for it.

The H20 NVL16 wins on raw compute specification. Its FP32 throughput of 39.54 TFLOPS is roughly 2.9 times the Arc B580's 13.67 TFLOPS. Its FP16 throughput of 79.07 TFLOPS (2:1) is also roughly 2.9 times the Arc B580's 27.34 TFLOPS (2:1). The H20 NVL16 carries 312 tensor cores while the Arc B580 lists none, which positions the server part for AI and machine learning workloads. Its memory capacity of 96 GB HBM3 across a 6144-bit bus delivers 4.03 TB/s of bandwidth, compared to the Arc B580's 12 GB GDDR6 on a 192-bit bus at 456.0 GB/s.

The percentile rankings reflect their different positioning. The Arc B580 sits at the 68th percentile against all GPUs in the database, with an average benchmark score of 23021. The H20 NVL16 is at the 50th percentile with an average benchmark score of 0, which is a direct consequence of having no benchmark entries recorded.

Architecture Differences

The two parts come from different architectural lineages. The Arc B580 uses the BMG-G21 chip built on Intel's Xe2-HPG architecture, part of the Battlemage generation within the Arc 5 series. The H20 NVL16 uses the GH100 chip built on NVIDIA's Hopper architecture, part of the Server Hopper generation. Both are fabricated by TSMC on a 5 nm process, but the similarities end there.

Transistor counts and die sizes differ substantially. The GH100 packs 80,000 million transistors on a 814 mm² die, while the BMG-G21 has 19,600 million transistors on a 272 mm² die. The resulting transistor densities are 98.3 million per mm² for the H20 NVL16 and 72.1 million per mm² for the Arc B580. The H20 NVL16 is the denser and dramatically larger chip.

The compute architectures diverge in their core organization. The Arc B580 has 2560 shading units, 160 texture mapping units, 80 ROPs, and 20 RT cores. The H20 NVL16 has 9984 shading units, 312 texture mapping units, only 24 ROPs, and 312 tensor cores. The H20 NVL16 lists no RT cores in the database, which is consistent with its server positioning where rasterization and ray tracing are not the primary duties. The Arc B580 has no tensor cores listed, marking it as a graphics-first part.

Clock behavior also separates the two. The Arc B580 runs at a base and boost clock of 2670 MHz, with memory at 2375 MHz (19 Gbps effective). The H20 NVL16 has a base clock of 1830 MHz and a boost of 1980 MHz, with memory at 1313 MHz (5.3 Gbps effective). The Arc B580's higher clocks help it reach a pixel rate of 213.6 GPixel/s and a texture rate of 427.2 GTexel/s. The H20 NVL16 manages 47.52 GPixel/s but a higher texture rate of 617.8 GTexel/s, due to its larger TMU count.

The H20 NVL16 is an SXM module with no display outputs, no power connectors listed, and no DirectX, OpenGL, or Vulkan API support. The Arc B580 is a dual-slot card with a single 8-pin power connector, one HDMI 2.1a output, and three DisplayPort 2.1 outputs, with DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 support.

Head-to-Head Benchmarks

The database contains no head-to-head benchmark entries for these two parts, so the comparison must rely on the separately recorded results and specification data. The Arc B580's recorded benchmark suite shows a mix of DirectX and compute results. Its Passmark scores range from 76 in DirectX 10 to 183 in DirectX 9, with DirectX 11 at 128 and DirectX 12 at 76. The Passmark G2D score is 709, the G3D score is 15748, and the GPU compute score is 7729. Geekbench OpenCL returns 92821 and Vulkan returns 109672.

The H20 NVL16's benchmark array is empty, meaning the database has no measured performance results for it. Its average benchmark score of 0 confirms this absence. The only quantitative performance indicators for the H20 NVL16 come from its specification sheet: 39.54 TFLOPS FP32, 79.07 TFLOPS FP16, and 4.03 TB/s memory bandwidth.

The nearest rivals recorded for the Arc B580 give context for its average score of 23021. The AMD Radeon RX 580 2048SP scores 23061, which is 0.2 percent higher. The NVIDIA GeForce RTX 2080 scores 22895, which is 0.6 percent lower. The NVIDIA GeForce RTX 3080 scores 23172, which is 0.7 percent higher. The NVIDIA P106-100 scores 23249, which is 1 percent higher. The Arc B580 is within one percent of all four rivals in average score, indicating that its overall benchmark average lands in a tightly clustered performance band.

Specification Differences

The two cards differ across nearly every specification field in the database.

Memory is the most extreme difference. The H20 NVL16 uses 96 GB of HBM3 on a 6144-bit bus with 4.03 TB/s bandwidth. The Arc B580 uses 12 GB of GDDR6 on a 192-bit bus with 456.0 GB/s bandwidth. The H20 NVL16 has eight times the capacity and roughly 8.8 times the bandwidth.

Compute resources differ sharply. The H20 NVL16 has 9984 shading units versus 2560 for the Arc B580, 312 TMUs versus 160, and 312 tensor cores versus none. The Arc B580 has 80 ROPs versus 24 for the H20 NVL16, and 20 RT cores versus none listed. FP32 throughput is 39.54 TFLOPS for the H20 NVL16 versus 13.67 TFLOPS for the Arc B580. FP16 throughput is 79.07 TFLOPS versus 27.34 TFLOPS.

Power and physical format diverge completely. The H20 NVL16 is rated at 400 W TDP with a suggested PSU of 800 W, and it is an SXM module with no listed power connectors. The Arc B580 is rated at 190 W TDP with a suggested PSU of 450 W, using a single 8-pin connector and a dual-slot form factor. The Arc B580 measures 272 mm in length, 115 mm in height, and 45 mm in width. The H20 NVL16 has no dimensions recorded.

Bus interface and connectivity also differ. The H20 NVL16 uses PCIe 5.0 x16 and has no display outputs. The Arc B580 uses PCIe 4.0 x8 and has one HDMI 2.1a plus three DisplayPort 2.1 outputs. API support exists only on the Arc B580.

Release timing differs as well. The Arc B580 was released on 2024-12-12 and lists Alchemist as its predecessor. The H20 NVL16 was released on 2025-09-01, lists Server Ada as its predecessor, and Server Blackwell as its successor. Both are marked as Active in production status.

FAQ

Q: Which GPU has more memory?

A: The NVIDIA H20 NVL16 has 96 GB of HBM3 memory, while the Intel Arc B580 has 12 GB of GDDR6 memory.

Q: What is the memory bandwidth difference?

A: The H20 NVL16 delivers 4.03 TB/s over a 6144-bit bus, compared to the Arc B580's 456.0 GB/s over a 192-bit bus.

Q: Do both GPUs support DirectX?

A: No. The Arc B580 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The H20 NVL16 lists DirectX, OpenGL, and Vulkan all as N/A.

Q: How does the Arc B580's average benchmark score compare to its nearest rivals?

A: The Arc B580 has an average benchmark score of 23021. The AMD Radeon RX 580 2048SP scores 23061 (0.2 percent higher), the NVIDIA GeForce RTX 2080 scores 22895 (0.6 percent lower), the NVIDIA GeForce RTX 3080 scores 23172 (0.7 percent higher), and the NVIDIA P106-100 scores 23249 (1 percent higher).

Q: What is the FP32 performance of each card?

A: The H20 NVL16 delivers 39.54 TFLOPS FP32, while the Arc B580 delivers 13.67 TFLOPS FP32.

Q: Do both cards have tensor cores?

A: No. The H20 NVL16 has 312 tensor cores, while the Arc B580 lists no tensor cores. The Arc B580 has 20 RT cores, while the H20 NVL16 lists no RT cores.