Intel Arc B390 vs NVIDIA H20 Comparison

FAQ

Q: What is the Intel Arc B390, and what architecture does it use?

A: The Intel Arc B390 is an integrated graphics processor based on the Xe3-LPG architecture, built on Intel's 3 nm process. It belongs to the Arc Graphics-M (Panther Lake) generation and uses the Panther Lake chip.

Q: What kind of memory does the Intel Arc B390 use?

A: The Intel Arc B390 uses system shared memory. Its memory size, type, bus width, and bandwidth are all listed as system dependent, meaning the performance relies on the host system's main memory.

Q: What is the NVIDIA H20's memory configuration?

A: The NVIDIA H20 features 96 GB of HBM3 memory on a 6144-bit bus, delivering 4.03 TB/s of bandwidth. Its memory clock is 1313 MHz, which is 5.3 Gbps effective.

Q: What are the benchmark scores for these two GPUs?

A: The Intel Arc B390 has a recorded 3DMark Steel Nomad DX12 score of 1482. The NVIDIA H20 has no benchmark entries in the database, and its average benchmark score is listed as 0.

Q: What is the process node difference between the two?

A: The Intel Arc B390 is fabricated on a 3 nm process by Intel, while the NVIDIA H20 uses a 5 nm process from TSMC. The H20 also has a known die size and transistor count, whereas those details are unknown for the Arc B390.

Q: What is the power consumption difference?

A: The Intel Arc B390 has a TDP of 80 W, while the NVIDIA H20 has a TDP of 500 W. The H20 also lists a suggested PSU of 900 W, whereas the Arc B390 is an integrated GPU with no power connectors.

Architecture Differences

The Intel Arc B390 and NVIDIA H20 diverge sharply in their architectural foundations. The Arc B390 uses the Xe3-LPG architecture on a Panther Lake chip, built at 3 nm by Intel. It is an integrated graphics processor (IGP) intended for mobile or compact systems. The H20, by contrast, uses the Hopper architecture with a GH100 chip, fabricated at 5 nm by TSMC, and is designed as a server module (SXM).

The transistor counts and die sizes tell a story of scale. The H20 packs 80,000 million transistors on an 814 mm² die, with a transistor density of 98.3 million per square millimeter. The Arc B390 has no listed transistor count or die size, indicating it is a much smaller integrated part.

Compute resources also differ. The Arc B390 has 1536 shading units, 48 texture mapping units, and 24 ROPs. It includes 12 RT cores but no tensor cores. The H20 has 9984 shading units, 312 TMUs, and 24 ROPs. It has 312 tensor cores but no listed RT cores. That asymmetry means the H20 is built for massively parallel compute workloads, while the Arc B390 retains ray tracing hardware.

Clock speeds show a different design philosophy. The Arc B390 has a base clock of 300 MHz and a boost clock of 2500 MHz, which is a wide range typical of integrated parts that scale up under load. The H20 has a base clock of 1830 MHz and a boost of 1980 MHz, a much narrower range for sustained server operation.

API support also separates them. The Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The H20 lists N/A for DirectX, OpenGL, and Vulkan, confirming its role as a compute-focused accelerator with no consumer graphics API stack.

Where Each One Wins

Based on the recorded data, the Intel Arc B390 wins in consumer graphics workloads. It is the only one with a benchmark score, achieving 1482 in 3DMark Steel Nomad DX12. It also has a full graphics API stack, display outputs (portable device dependent), and a low 80 W TDP, making it suitable for integrated graphics in mobile devices.

The NVIDIA H20 wins in raw compute capacity. Its fp32 performance is 39.54 TFLOPS, and fp16 performance is 79.07 TFLOPS (2:1), both far above the Arc B390's 7.680 TFLOPS fp32 and 15.36 TFLOPS fp16. The H20's 4.03 TB/s memory bandwidth and 96 GB HBM3 capacity are in a different class entirely, suited for large-scale server inference or training tasks.

The Arc B390 has a higher pixel rate at 60.00 GPixel/s versus the H20's 47.52 GPixel/s, which suggests it may be more efficient at rasterization per clock. However, the H20's texture rate of 617.8 GTexel/s dwarfs the Arc B390's 120.0 GTexel/s, reflecting the H20's much larger TMU count.

Where the H20 wins is in memory capacity, bandwidth, and compute throughput. Where the Arc B390 wins is in being a functional graphics solution with display outputs, consumer APIs, and lower power requirements. The data shows they occupy opposite ends of the hardware spectrum.

Specification Differences

The two GPUs differ in nearly every measurable specification. The Intel Arc B390 is built on a 3 nm process by Intel, while the NVIDIA H20 uses a 5 nm process by TSMC. The H20 has 80,000 million transistors and an 814 mm² die, while those figures are unknown for the Arc B390.

Clock speeds: the Arc B390 runs at 300 MHz base and 2500 MHz boost. The H20 runs at 1830 MHz base and 1980 MHz boost. The H20's memory clock is 1313 MHz (5.3 Gbps effective), while the Arc B390's memory clock is system shared.

Memory: the Arc B390 uses system shared memory of undefined size and type. The H20 has 96 GB of HBM3 with a 6144-bit bus and 4.03 TB/s bandwidth.

Compute units: the Arc B390 has 1536 shading units, 48 TMUs, 24 ROPs, and 12 RT cores. The H20 has 9984 shading units, 312 TMUs, 24 ROPs, and 312 tensor cores. The H20 has no listed RT cores, and the Arc B390 has no tensor cores.

Performance rates: the Arc B390 delivers 60.00 GPixel/s pixel rate and 120.0 GTexel/s texture rate. The H20 delivers 47.52 GPixel/s and 617.8 GTexel/s. FP32 is 7.680 TFLOPS for the Arc B390 and 39.54 TFLOPS for the H20. FP16 is 15.36 TFLOPS for the Arc B390 and 79.07 TFLOPS for the H20.

Power and form factor: the Arc B390 has a TDP of 80 W, is an IGP, and has no power connectors. The H20 has a TDP of 500 W, is an SXM module, and suggests a 900 W PSU. The bus interface is IGP for the Arc B390 and PCIe 5.0 x16 for the H20.

Display outputs: the Arc B390 is portable device dependent, while the H20 has no outputs. Release dates differ as well: the Arc B390 was released on 2026-01-26, and the H20 on 2024-01-31.

Head-to-Head Benchmarks

Direct head-to-head benchmark data is absent from the database. There are no entries in the head-to-head benchmarks field, and the win counts for both items are 0. This means no recorded comparison test exists between the Intel Arc B390 and NVIDIA H20.

What the data does provide is a single benchmark score for the Arc B390: 1482 in 3DMark Steel Nomad DX12. That score places it at the 9th percentile among all GPUs. Its nearest rivals are all older NVIDIA parts: the GeForce GT 520MX with an average score of 1463 (1.3% ahead), the GeForce 800M at 1460 (1.5% ahead), the GeForce GT 625 OEM at 1446 (2.5% ahead), and the GeForce GT 710 at 1443 (2.7% ahead). The Arc B390 sits just above these legacy integrated and entry-level discrete parts.

The NVIDIA H20 has no benchmark score. Its average benchmark score is 0, and it has no nearest rivals listed. The database records its percentile as 50, but with no actual measured score, that percentile cannot be interpreted as a performance result.

The biggest win for the Arc B390 is simply having a measurable graphics benchmark. The biggest win for the H20 is in compute metrics: 39.54 TFLOPS fp32 is 5.15 times the Arc B390's 7.680 TFLOPS. The H20's fp16 of 79.07 TFLOPS is 5.15 times the Arc B390's 15.36 TFLOPS. The H20's texture rate of 617.8 GTexel/s is 5.15 times the Arc B390's 120.0 GTexel/s. The H20's memory bandwidth of 4.03 TB/s is effectively unlimited compared to the Arc B390's system dependent bandwidth.

The Arc B390 counters with a pixel rate of 60.00 GPixel/s, which is 26% higher than the H20's 47.52 GPixel/s. That indicates the Arc B390 has better per-clock raster output, despite having far fewer total resources.

The Verdict

The data profiles two entirely different products. The Intel Arc B390 is an integrated GPU for consumer mobile devices. It has a full graphics API stack, display outputs, ray tracing cores, and a low 80 W power draw. Its benchmark score of 1482 places it near legacy NVIDIA parts like the GT 710 and GT 625 OEM, with deltas of 2.5% to 2.7% against those parts. Its 9th percentile ranking confirms it sits at the low end of the performance spectrum.

The NVIDIA H20 is a server accelerator with no display outputs and no consumer graphics API support. It is built for compute density: 9984 shading units, 312 tensor cores, 96 GB of HBM3, and 4.03 TB/s of bandwidth. Its 500 W TDP and SXM form factor make it unsuitable for any client workstation. Its fp32 and fp16 throughput are over 5 times that of the Arc B390.

For a user seeking a graphics solution with display capability and mainstream API support, the Arc B390 is the only functional choice. For a user seeking raw compute throughput, memory capacity, or server-side acceleration, the H20 is the only option that delivers.

The absence of shared benchmark results means no direct performance comparison exists. The recorded data shows the Arc B390 can render frames and output video, while the H20 cannot. Conversely, the H20 can process tensors and massive data sets, which the Arc B390 is not equipped to do.

The verdict follows from the specification sheets. The Arc B390 wins in graphics functionality, power efficiency, and pixel throughput. The H20 wins in compute throughput, memory bandwidth, and capacity. Each product is definitive in its own domain, and neither overlaps with the other's intended workload.