Intel Arc B390 vs NVIDIA H100 PCIe 96 GB Comparison

Head-to-Head Benchmarks

The recorded data shows only one benchmark entry for the Intel Arc B390: a 3DMark Steel Nomad DX12 score of 1482. The NVIDIA H100 PCIe 96 GB has no benchmark scores in the database, so direct numerical comparison is limited to the Arc B390's positioning against other GPUs. The Arc B390 sits at the 9th percentile of all GPUs, placing it near the bottom of the performance distribution. Its nearest rivals illustrate its performance tier: the NVIDIA GeForce GT 520MX scores 1463, a 1.3% gap; the GeForce 800M scores 1460, a 1.5% gap; the GT 625 OEM scores 1446, a 2.5% gap; and the GT 710 scores 1443, a 2.7% gap. The Arc B390 leads each of these by a narrow margin, with the largest advantage being 2.7% over the GT 710.

The H100 PCIe 96 GB, by contrast, has an average benchmark score of 0 and no nearest rivals listed, meaning the database contains no comparative performance data for it. The H100 does hold a 50th percentile position among all GPUs, but this percentile is not backed by any recorded benchmark scores. The absence of head-to-head benchmark entries between the Arc B390 and the H100 means the database cannot produce a direct score comparison. What the data does show is a massive architectural and specification gap: the H100 delivers 62.08 TFLOPS of FP32 performance versus 7.680 TFLOPS for the Arc B390, a difference of roughly 8x in raw compute throughput. The H100's FP16 output reaches 248.3 TFLOPS (4:1 ratio), while the Arc B390 manages 15.36 TFLOPS (2:1 ratio), a 16x difference. The H100 also has 528 tensor cores, a feature the Arc B390 lacks entirely, with no tensor core count listed.

Texture and pixel rates reinforce the divide. The H100 achieves 969.9 GTexel/s versus 120.0 GTexel/s for the Arc B390. Pixel rates are closer: the H100 produces 44.09 GPixel/s, while the Arc B390 produces 60.00 GPixel/s, meaning the Arc B390 actually leads in pixel fill rate by 36%. This is an anomaly driven by the Arc B390's higher clock behavior in a low-power integrated design, but it does not offset the H100's overwhelming advantage in every other measured compute metric.

Where Each One Wins

The Arc B390 wins in scenarios that prioritize pixel output and power efficiency. Its 60.00 GPixel/s pixel rate exceeds the H100's 44.09 GPixel/s, so tasks that depend heavily on rasterization fill, such as basic 2D rendering or lightweight graphics workloads, favor the Intel part. The Arc B390 also draws only 80 W of power, compared to the H100's 700 W TDP, and it requires no power connectors, whereas the H100 needs an 8-pin EPS connector and a suggested 1100 W power supply. The Arc B390 is an integrated graphics processor (IGP) with a slot width of IGP, so it fits into systems without a discrete GPU slot. It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it viable for modern graphics APIs in a portable device context. Its display outputs are listed as "Portable Device Dependent," meaning it is designed for systems where the display path is determined by the host device.

The H100 wins in every compute-heavy category. Its 16896 shading units dwarf the Arc B390's 1536, and its 528 texture mapping units compare to 48 on the Arc B390. The H100 has 528 tensor cores; the Arc B390 has none listed. The H100's 96 GB of HBM3 memory with a 5120-bit bus and 3.36 TB/s bandwidth stands in stark contrast to the Arc B390's system-shared memory, which has no dedicated size, type, or bus width and a bandwidth described as "System Dependent." The H100's memory clock runs at 1313 MHz with 5.3 Gbps effective speed, while the Arc B390 relies entirely on system memory. For FP32 compute, the H100 delivers 62.08 TFLOPS, which is 8.1x the Arc B390's 7.680 TFLOPS. For FP16, the H100 produces 248.3 TFLOPS, which is 16.2x the Arc B390's 15.36 TFLOPS. The H100's 80,000 million transistors on an 814 mm² die with a density of 98.3M / mm² indicate a purpose-built server compute chip, whereas the Arc B390's transistor count and die size are not recorded.

The H100 is a dual-slot PCIe 5.0 x16 card with a 268 mm length and 111 mm height. It has no display outputs, confirming its role as a compute accelerator rather than a graphics output device. Its predecessor is Server Ada and its successor is Server Blackwell, placing it in the Hopper server generation. The Arc B390 uses the Panther Lake chip with Xe3-LPG architecture, part of the Arc Graphics-M (Panther Lake) generation, and it is built on Intel's 3 nm process. The H100 uses the GH100 chip with Hopper architecture on TSMC's 5 nm process.

Architecture Differences

The Intel Arc B390 uses the Xe3-LPG architecture on the Panther Lake chip, fabricated by Intel on a 3 nm process. It belongs to the Arc Graphics-M (Panther Lake) generation, indicating a mobile or integrated graphics focus. The NVIDIA H100 PCIe 96 GB uses the Hopper architecture on the GH100 chip, fabricated by TSMC on a 5 nm process. It belongs to the Server Hopper (Hxx) generation. The process node difference is significant: Intel's 3 nm node is newer than TSMC's 5 nm node, but the H100 compensates with a massive die. The H100 has 80,000 million transistors on an 814 mm² die, yielding a density of 98.3M transistors per mm². The Arc B390's transistor count and die size are unknown, so density cannot be compared.

The H100's memory subsystem is entirely dedicated: 96 GB of HBM3 on a 5120-bit bus, delivering 3.36 TB/s of bandwidth. The memory clock is 1313 MHz with 5.3 Gbps effective speed. The Arc B390 has no dedicated memory; it uses system-shared memory with a system-dependent bandwidth. This architectural choice makes the Arc B390 suitable for low-power integrated systems where memory is shared with the CPU, but it cannot match the H100's dedicated high-bandwidth memory for large data transfers.

The H100 includes 528 tensor cores, which are absent from the Arc B390's specification. Tensor cores accelerate matrix math for AI and deep learning workloads. The H100's FP16 performance of 248.3 TFLOPS at a 4:1 ratio reflects its tensor core throughput. The Arc B390's FP16 of 15.36 TFLOPS at a 2:1 ratio is a shader-based rate without tensor acceleration. The H100 also has 528 TMUs versus 48 on the Arc B390, and 16896 shading units versus 1536. Both have 24 ROPs, so the pixel output difference is purely clock-driven: the Arc B390's boost clock is 2500 MHz, while the H100's boost is 1837 MHz. The Arc B390's base clock is 300 MHz, and the H100's base is 1665 MHz. The Arc B390's 60.00 GPixel/s pixel rate comes from its 2500 MHz boost and 24 ROPs; the H100's 44.09 GPixel/s comes from 1837 MHz and 24 ROPs.

The H100's API support is not listed for DirectX, OpenGL, or Vulkan, which is typical for server compute cards that do not target graphics APIs. The Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it a full-featured graphics solution for consumer or portable devices. The H100 has no display outputs, while the Arc B390's outputs are portable-device dependent, meaning it can drive displays in the host system.

The H100 requires 700 W of power, an 8-pin EPS connector, and a suggested 1100 W power supply. The Arc B390 has a TDP of 80 W, no power connectors, and no suggested PSU. The H100 is a dual-slot card with dimensions of 268 mm by 111 mm, while the Arc B390 is an IGP with no slot width beyond IGP. The H100's bus interface is PCIe 5.0 x16, while the Arc B390's is IGP. The release dates differ: the H100 launched on 2023-03-20, and the Arc B390 launched on 2026-01-26. Both are listed as Active in production status.

FAQ

Q: Which GPU has higher FP32 performance?

A: The NVIDIA H100 PCIe 96 GB delivers 62.08 TFLOPS of FP32, which is 8.1x the Intel Arc B390's 7.680 TFLOPS.

Q: Does the Arc B390 have tensor cores?

A: No tensor core count is listed for the Arc B390. The H100 has 528 tensor cores.

Q: How much memory does each GPU have?

A: The H100 has 96 GB of HBM3 memory with a 5120-bit bus and 3.36 TB/s bandwidth. The Arc B390 uses system-shared memory with no dedicated size, type, or bus width, and its bandwidth is system dependent.

Q: What is the pixel fill rate difference?

A: The Arc B390 achieves 60.00 GPixel/s, which exceeds the H100's 44.09 GPixel/s. The Arc B390 leads by 36% in pixel fill rate.

Q: What are the power requirements?

A: The Arc B390 has a TDP of 80 W and requires no power connectors. The H100 has a TDP of 700 W, requires an 8-pin EPS connector, and has a suggested power supply of 1100 W.

Q: What is the process node difference?

A: The Arc B390 is fabricated by Intel on a 3 nm process. The H100 is fabricated by TSMC on a 5 nm process.

Q: Which GPU has better texture processing?

A: The H100 has 528 TMUs and a texture rate of 969.9 GTexel/s. The Arc B390 has 48 TMUs and a texture rate of 120.0 GTexel/s. The H100's texture rate is 8.1x higher.

Q: Does the Arc B390 support modern graphics APIs?

A: Yes, the Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The H100 has no API support listed for DirectX, OpenGL, or Vulkan.

Specification Differences

| Specification | Intel Arc B390 | NVIDIA H100 PCIe 96 GB |

|---|---|---|

| Chip | Panther Lake | GH100 |

| Architecture | Xe3-LPG | Hopper |

| Process Node | 3 nm (Intel) | 5 nm (TSMC) |

| Transistors | Unknown | 80,000 million |

| Die Size | Unknown | 814 mm² |

| Transistor Density | N/A | 98.3M / mm² |

| Base Clock | 300 MHz | 1665 MHz |

| Boost Clock | 2500 MHz | 1837 MHz |

| Memory Size | System Shared | 96 GB HBM3 |

| Memory Bus Width | System Shared | 5120 bit |

| Memory Bandwidth | System Dependent | 3.36 TB/s |

| Shading Units | 1536 | 16896 |

| TMUs | 48 | 528 |

| ROPs | 24 | 24 |

| Tensor Cores | None listed | 528 |

| RT Cores | 12 | None listed |

| Pixel Rate | 60.00 GPixel/s | 44.09 GPixel/s |

| Texture Rate | 120.0 GTexel/s | 969.9 GTexel/s |

| FP32 | 7.680 TFLOPS | 62.08 TFLOPS |

| FP16 | 15.36 TFLOPS (2:1) | 248.3 TFLOPS (4:1) |

| TDP | 80 W | 700 W |

| Slot Width | IGP | Dual-slot |

| Power Connectors | None | 8-pin EPS |

| Suggested PSU | None | 1100 W |

| Bus Interface | IGP | PCIe 5.0 x16 |

| Display Outputs | Portable Device Dependent | No outputs |

| DirectX | 12 Ultimate (12_2) | None |

| OpenGL | 4.6 | None |

| Vulkan | 1.4 | None |

| Length | N/A | 268 mm (10.6 inches) |

| Height | N/A | 111 mm (4.4 inches) |

| Release Date | 2026-01-26 | 2023-03-20 |

| Predecessor | None | Server Ada |

| Successor | None | Server Blackwell |

| Percentile | 9th | 50th |

| Avg Benchmark Score | 1482 | 0 |

The Verdict

The data shows two GPUs built for entirely different purposes. The Intel Arc B390 is a low-power integrated graphics processor with an 80 W TDP, no power connectors, and system-shared memory. Its only recorded benchmark score is 1482 in 3DMark Steel Nomad DX12, placing it at the 9th percentile of all GPUs and within 2.7% of low-end discrete cards like the GeForce GT 710. The Arc B390's strengths are its pixel fill rate of 60.00 GPixel/s, which exceeds the H100's 44.09 GPixel/s, and its full support for DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. It is a graphics solution for portable devices, where the display output is device dependent and power draw is minimal.

The NVIDIA H100 PCIe 96 GB is a server compute accelerator with no display outputs and no graphics API support. Its 16896 shading units, 528 tensor cores, and 96 GB of HBM3 memory with 3.36 TB/s bandwidth position it for compute workloads, not rasterization. The H100 delivers 62.08 TFLOPS of FP32, 8.1x the Arc B390's 7.680 TFLOPS, and 248.3 TFLOPS of FP16, 16.2x the Arc B390's 15.36 TFLOPS. Its 969.9 GTexel/s texture rate is 8.1x higher. The H100 requires 700 W, an 8-pin EPS connector, and a suggested 1100 W power supply, and it occupies a dual-slot with a 268 mm length.

The Arc B390 should be selected for integrated graphics duty in a portable or low-power system where modern graphics APIs and basic rendering are needed. The H100 should be selected for server or compute environments where massive FP32 and FP16 throughput, tensor core acceleration, and high-bandwidth memory are required. The database confirms they do not compete in the same segment: the Arc B390's nearest rivals are all low-end GeForce parts, while the H100 has no rivals listed because no comparable data exists. The only metric where the Arc B390 wins is pixel rate, and that advantage does not translate into general performance. The H100's 50th percentile ranking versus the Arc B390's 9th percentile, despite the H100's lack of recorded benchmark scores, reflects the different performance tiers implied by their specifications. The Arc B390 is a capable integrated graphics solution; the H100 is a compute accelerator with no graphics output. The choice depends on whether the system needs display output and low power consumption or raw compute throughput and dedicated memory.