Intel Arc Pro B370 vs NVIDIA H20 Comparison

Head-to-Head Benchmarks

The database does not contain any head-to-head benchmark results for these two processors, and neither device has any individual benchmark scores recorded. The recorded data shows winsA and winsB both at zero, and the headToHeadBenchmarks array is empty. Consequently, any direct performance comparison must be derived from the architectural and specification data available in the database, rather than from measured application or game frame rates.

When the raw compute capabilities are examined, the NVIDIA H20 dominates in raw throughput metrics. The H20 delivers 39.54 TFLOPS of FP32 compute, which is 6.4 times the 6.144 TFLOPS produced by the Intel Arc Pro B370. In FP16 calculations, the H20 reaches 79.07 TFLOPS (2:1), compared to 12.29 TFLOPS (2:1) for the Intel part, a factor of roughly 6.4 as well. The texture rate difference is even larger: the H20 sustains 617.8 GTexel/s against 96.00 GTexel/s for the Arc Pro B370, a 6.4x advantage. Pixel rates are nearly identical, with the Intel device at 48.00 GPixel/s and the NVIDIA device at 47.52 GPixel/s, a negligible 0.48 GPixel/s gap in favor of the Intel part.

The absence of measured benchmark scores means the percentileVsAllGpus field is identical for both, sitting at 50 for each device. That equal percentile does not reflect the massive compute disparity, it simply indicates that neither device has recorded performance data in the database. The average benchmark score is zero for both, confirming the lack of empirical measurements.

Memory capacity and bandwidth show an enormous separation. The H20 carries 96 GB of HBM3 memory on a 6144-bit bus, delivering 4.03 TB/s of bandwidth. The Intel Arc Pro B370 uses system shared memory, with a bus width marked as system shared and bandwidth described as system dependent. No fixed memory bandwidth figure exists for the Intel part because it depends entirely on the host platform's memory subsystem. The H20's dedicated 4.03 TB/s bandwidth is a fixed, measured property of the GPU itself, whereas the Intel part's effective bandwidth is variable and unquantified in the database.

Clock speeds tell a different story. The Intel Arc Pro B370 has a base clock of 300 MHz and a boost clock of 2400 MHz, a 2.1 GHz boost range. The NVIDIA H20 operates at a base of 1830 MHz and a boost of 1980 MHz, only a 150 MHz span. The Intel part's boost clock exceeds the H20's boost clock by 420 MHz, but the H20 runs far more shading units, 9984 versus 1280, so its lower clock is multiplied across far more execution hardware.

The power envelope is starkly different. The Intel Arc Pro B370 consumes 25 W, making it an IGP with no power connectors. The NVIDIA H20 draws 500 W, uses an SXM module form factor, and carries a suggested PSU rating of 900 W. That 25 W versus 500 W difference is a 20x gap in thermal design power. The Intel part is integrated into a processor, while the H20 is a discrete server module, which explains the power disparity.

FAQ

Q: Which GPU has higher FP32 compute performance, the Intel Arc Pro B370 or the NVIDIA H20?

A: The NVIDIA H20 delivers 39.54 TFLOPS FP32, which is approximately 6.4 times the Intel Arc Pro B370's 6.144 TFLOPS FP32.

Q: How do the memory subsystems compare between these two GPUs?

A: The NVIDIA H20 uses 96 GB of HBM3 memory on a 6144-bit bus, providing a fixed 4.03 TB/s of bandwidth. The Intel Arc Pro B370 uses system shared memory with a system dependent bandwidth, so its memory performance is not fixed and depends on the host platform.

Q: What are the power consumption figures for each GPU?

A: The Intel Arc Pro B370 has a TDP of 25 W and requires no power connectors, while the NVIDIA H20 has a TDP of 500 W and is an SXM module with a suggested PSU rating of 900 W.

Q: Does either GPU support DirectX 12 Ultimate?

A: The Intel Arc Pro B370 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The NVIDIA H20 lists N/A for DirectX, OpenGL, and Vulkan, indicating it has no display outputs and no consumer graphics API support.

Q: What is the shading unit count for each GPU?

A: The NVIDIA H20 has 9984 shading units, while the Intel Arc Pro B370 has 1280 shading units. The H20 also has 312 texture mapping units and 24 ROPs, compared to 40 TMUs and 20 ROPs on the Intel part.

Q: Which GPU has a higher pixel fill rate?

A: The Intel Arc Pro B370 has a pixel rate of 48.00 GPixel/s, slightly above the NVIDIA H20's 47.52 GPixel/s. The difference is only 0.48 GPixel/s in favor of the Intel part.

Where Each One Wins

The Intel Arc Pro B370 wins in scenarios that require low power, integrated graphics, and display output. Its 25 W TDP means it can operate without any power connectors, fitting into an IGP slot width. It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it suitable for graphics workloads on portable devices. The display outputs are described as portable device dependent, so it is designed to drive screens. Its pixel rate of 48.00 GPixel/s edges out the H20, which is the only raw performance metric where the Intel part leads. The 300 MHz base clock and 2400 MHz boost clock indicate a wide dynamic range for power management. The architecture is Xe3-LPG on a 3 nm process from Intel, part of the Arc Graphics-WM (Panther Lake) generation. It has 10 ray tracing cores, a feature that the NVIDIA H20 does not list in the database. For a mobile or embedded device that needs graphics API support and ray tracing, the Arc Pro B370 is the only one of the two with those capabilities.

The NVIDIA H20 wins in every heavy compute category except pixel rate. Its 39.54 TFLOPS FP32 and 79.07 TFLOPS FP16 dwarf the Intel part's 6.144 TFLOPS and 12.29 TFLOPS. The 4.03 TB/s memory bandwidth on 96 GB of HBM3 is a server-class specification, while the Intel part has no fixed memory bandwidth. The 9984 shading units, 312 TMUs, and 312 tensor cores give it massive parallel throughput. The H20 is a server Hopper (Hxx) generation device on a 5 nm TSMC process, with 80,000 million transistors on an 814 mm² die. It has no display outputs, no graphics APIs, and is built as an SXM module. Its 500 W TDP and 900 W suggested PSU indicate a data center accelerator intended for compute workloads, not graphics rendering. The tensor core count of 312, which the Intel part does not list, points to AI and deep learning workloads as the primary use case.

The use-case split is clear. For graphics, display, ray tracing, and low-power integrated operation, the Intel Arc Pro B370 is the relevant device. For raw compute, high-bandwidth memory access, tensor operations, and server deployment, the NVIDIA H20 is the only viable option. The 25 W TDP of the Intel part allows fanless or passively cooled designs, while the 500 W H20 requires a 900 W suggested PSU and server infrastructure.

Specification Differences

The two devices differ across nearly every recorded specification. The Intel Arc Pro B370 uses the Panther Lake chip with Xe3-LPG architecture, while the NVIDIA H20 uses the GH100 chip with Hopper architecture. The Intel process node is 3 nm from Intel's foundry, the NVIDIA node is 5 nm from TSMC. The H20 has 80,000 million transistors on an 814 mm² die with a transistor density of 98.3M / mm², while the Arc Pro B370 has unknown transistor count and die size.

Clock speeds differ: the Intel part runs at 300 MHz base and 2400 MHz boost, the H20 runs at 1830 MHz base and 1980 MHz boost. Memory is system shared on the Intel part, while the H20 has 96 GB HBM3 on a 6144-bit bus with 4.03 TB/s bandwidth. The shading units differ at 1280 versus 9984, TMUs at 40 versus 312, ROPs at 20 versus 24. The Intel part has 10 ray tracing cores, the H20 has no listed ray tracing cores. The H20 has 312 tensor cores, the Intel part has no listed tensor cores.

Pixel rate is nearly equal at 48.00 GPixel/s for Intel and 47.52 GPixel/s for NVIDIA. Texture rate differs greatly at 96.00 GTexel/s versus 617.8 GTexel/s. FP32 is 6.144 TFLOPS versus 39.54 TFLOPS. FP16 is 12.29 TFLOPS versus 79.07 TFLOPS. TDP is 25 W versus 500 W. Slot width is IGP versus SXM Module. The Intel part has no power connectors, the H20 lists none as well but has a 900 W suggested PSU. Bus interface is IGP versus PCIe 5.0 x16. Display outputs are portable device dependent for Intel, no outputs for NVIDIA. The Intel part supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, the H20 lists N/A for all three.

Architecture Differences

The Intel Arc Pro B370 belongs to the Arc Graphics-WM (Panther Lake) generation, using the Panther Lake chip and Xe3-LPG architecture. It is built on a 3 nm process at Intel's foundry. The H20 belongs to the Server Hopper (Hxx) generation, using the GH100 chip and Hopper architecture, built on a 5 nm process at TSMC. The transistor counts and die sizes are not comparable because the Intel data lists them as unknown, while the H20 has 80,000 million transistors on an 814 mm² die.

The Intel part is an integrated graphics processor, as indicated by its IGP slot width, IGP bus interface, and system shared memory. It carries 10 ray tracing cores, a feature absent from the H20's recorded specifications. The H20 is a discrete server accelerator with an SXM module form factor and PCIe 5.0 x16 interface. It carries 312 tensor cores, which the Intel part does not list. The H20 has no display outputs and no graphics API support, consistent with a compute-focused server accelerator. The Intel part supports a full consumer graphics API stack including DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, along with portable device dependent display outputs.

The memory architectures are fundamentally different. The Intel part shares system memory, making its bandwidth dependent on the host platform. The H20 uses dedicated HBM3 with a fixed 4.03 TB/s bandwidth and 96 GB capacity. The clock behavior also differs: the Intel part has a very low 300 MHz base clock with a 2400 MHz boost, suggesting aggressive power gating, while the H20 runs at a narrower 1830 MHz to 1980 MHz range, typical of a sustained-load server part. The 25 W TDP of the Intel part versus 500 W for the H20 reflects the integrated versus discrete design philosophy. The Intel part has no power connectors, the H20 has a 900 W suggested PSU. The predecessor and successor relationships also differ: the Intel part succeeds HD Graphics-WM, the H20 succeeds Server Ada and is succeeded by Server Blackwell.