Intel Arc B390 vs NVIDIA H20 NVL16 Comparison

The Verdict

The recorded data presents a stark contrast between two products designed for entirely different purposes. The Intel Arc B390 is an integrated graphics processor (IGP) built into the Panther Lake mobile platform, while the NVIDIA H20 NVL16 is a server-grade accelerator module based on the Hopper architecture. The benchmark database contains a single recorded 3DMark Steel Nomad DX12 score for the Intel Arc B390 (1482 points), placing it at the 9th percentile of all GPUs. The NVIDIA H20 NVL16 has no recorded benchmark scores in the database, leaving its average benchmark score at zero and its percentile at 50 by default. This absence of comparative data means the database cannot provide a direct head-to-head performance verdict.

The Intel Arc B390 is positioned for portable devices where graphics capability is secondary to power efficiency. Its 80 W TDP, integrated form factor, and system-shared memory indicate a design for everyday computing, media playback, and light gaming. The data shows its nearest rivals are all legacy NVIDIA discrete GPUs: the GeForce GT 520MX (1.3% slower), GeForce 800M (1.5% slower), GeForce GT 625 OEM (2.5% slower), and GeForce GT 710 (2.7% slower). These margins are narrow, suggesting the Arc B390 delivers performance in the same class as those older entry-level parts.

The NVIDIA H20 NVL16 targets a completely different market segment. With 96 GB of HBM3 memory, a 6144-bit memory bus, 4.03 TB/s of bandwidth, 9984 shading units, and 312 tensor cores, the data describes a compute accelerator for large-scale AI and scientific workloads. Its 400 W TDP and SXM module form factor confirm this is not a consumer graphics product. The H20 NVL16 has no display outputs and no DirectX, OpenGL, or Vulkan API support, which indicates the database treats it as a non-rendering compute device.

The verdict from the available data is straightforward: the Intel Arc B390 suits mobile devices requiring integrated graphics with modest 3D performance, while the NVIDIA H20 NVL16 serves server environments prioritizing massive memory capacity and tensor throughput. These products do not compete in any measurable way within this database.

Where Each One Wins

The Intel Arc B390 wins in the only benchmark category where data exists. Its 3DMark Steel Nomad DX12 score of 1482 points exceeds the scores of all four recorded nearest rivals. The GeForce GT 520MX scores 1463 (1.3% lower), the GeForce 800M scores 1460 (1.5% lower), the GeForce GT 625 OEM scores 1446 (2.5% lower), and the GeForce GT 710 scores 1443 (2.7% lower). This pattern shows the Arc B390 edges out each competitor by a small but consistent margin, suggesting its Xe3-LPG architecture delivers slightly better DirectX 12 performance than those older NVIDIA parts.

The NVIDIA H20 NVL16 wins in every specification category that matters for compute workloads. Its shading unit count of 9984 is 6.5 times higher than the Arc B390's 1536. Its texture rate of 617.8 GTexel/s is over five times the Arc B390's 120.0 GTexel/s. Its FP32 throughput of 39.54 TFLOPS is more than five times the Arc B390's 7.680 TFLOPS. Its FP16 throughput of 79.07 TFLOPS is also more than five times the Arc B390's 15.36 TFLOPS. The H20 NVL16's 96 GB memory capacity dwarfs the Arc B390's system-shared memory, and its 4.03 TB/s bandwidth is far beyond what an integrated processor can access.

The use-case split is clear from the data. The Arc B390 wins in portable, power-constrained environments where integrated graphics are the only option. The H20 NVL16 wins in server racks where power consumption up to 400 W is acceptable and where memory capacity and tensor core count determine suitability for large model training or inference workloads.

Architecture Differences

The Intel Arc B390 uses the Panther Lake chip built on Intel's 3 nm process node, with the Xe3-LPG architecture belonging to the Arc Graphics-M (Panther Lake) generation. The NVIDIA H20 NVL16 uses the GH100 chip built on TSMC's 5 nm process node, with the Hopper architecture belonging to the Server Hopper (Hxx) generation. These process differences reflect the distinct design goals: the 3 nm node targets mobile power efficiency, while the 5 nm node on a 814 mm² die with 80,000 million transistors targets maximum compute density.

The Arc B390 contains 1536 shading units, 48 texture mapping units, 24 raster operation units, and 12 ray tracing cores. It has no dedicated tensor cores listed in the database. The H20 NVL16 contains 9984 shading units, 312 texture mapping units, 24 raster operation units, and 312 tensor cores. It has no ray tracing cores listed. The H20 NVL16's transistor density of 98.3M per mm² contrasts with the Arc B390's unknown transistor count and die size.

Clock behavior differs significantly. The Arc B390 runs at a 300 MHz base clock and boosts to 2500 MHz, which is typical for an integrated part with power limits. The H20 NVL16 runs at a 1830 MHz base clock and boosts to 1980 MHz, a much narrower boost range. The memory architectures are entirely different: the Arc B390 uses system-shared memory with system-dependent bandwidth, while the H20 NVL16 uses dedicated HBM3 memory running at 1313 MHz with 5.3 Gbps effective speed across a 6144-bit bus.

API support separates the two as well. The Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it a full-featured graphics solution. The H20 NVL16 lists all three APIs as N/A, confirming its role as a compute-only accelerator with no display outputs. The bus interface also differs: the Arc B390 uses an integrated graphics processor connection, while the H20 NVL16 uses PCIe 5.0 x16.

The power envelopes reflect these architectural choices. The Arc B390 draws 80 W with no power connectors, fitting its IGP form factor. The H20 NVL16 draws 400 W, requires an 800 W suggested power supply, and mounts as an SXM module.

FAQ

Q: Does the database show any direct benchmark comparison between the Intel Arc B390 and the NVIDIA H20 NVL16?

A: No. The head-to-head benchmark list is empty, and there are no recorded benchmark scores for the NVIDIA H20 NVL16. The Intel Arc B390 has one recorded score in 3DMark Steel Nomad DX12, but no comparative data exists for the H20 NVL16.

Q: Which GPU has a higher FP32 compute throughput?

A: The NVIDIA H20 NVL16 delivers 39.54 TFLOPS of FP32 performance, which is over five times the Intel Arc B390's 7.680 TFLOPS. This difference reflects the H20 NVL16's server-class compute design versus the Arc B390's integrated mobile graphics role.

Q: What memory configuration does each product use?

A: The Intel Arc B390 uses system-shared memory with a system-dependent bandwidth, meaning it draws from the host system's RAM. The NVIDIA H20 NVL16 uses 96 GB of dedicated HBM3 memory with a 6144-bit bus and 4.03 TB/s bandwidth.

Q: Can the NVIDIA H20 NVL16 output video to a display?

A: No. The database lists display outputs as "No outputs" for the H20 NVL16, and its DirectX, OpenGL, and Vulkan API support are all marked N/A. It is a compute-only accelerator.

Q: How does the Intel Arc B390 compare to its nearest recorded rivals?

A: The Arc B390 scores 1482 in 3DMark Steel Nomad DX12, which is 1.3% higher than the GeForce GT 520MX (1463), 1.5% higher than the GeForce 800M (1460), 2.5% higher than the GeForce GT 625 OEM (1446), and 2.7% higher than the GeForce GT 710 (1443).

Q: What are the power requirements for each product?

A: The Intel Arc B390 has a 80 W TDP and requires no power connectors. The NVIDIA H20 NVL16 has a 400 W TDP and a suggested power supply of 800 W.

Head-to-Head Benchmarks

The database contains no head-to-head benchmark results between the Intel Arc B390 and the NVIDIA H20 NVL16. The wins counter shows zero for both products. The absence of data is itself informative: these two accelerators serve such different markets that no common benchmark suite has produced comparable results.

The only benchmark score available belongs to the Intel Arc B390. Its 1482 points in 3DMark Steel Nomad DX12 place it at the 9th percentile of all GPUs in the database. This low percentile indicates that the Arc B390's graphics performance lands near the bottom of the overall distribution. The nearest rivals confirm this positioning. The GeForce GT 520MX trails by 1.3%, the GeForce 800M by 1.5%, the GeForce GT 625 OEM by 2.5%, and the GeForce GT 710 by 2.7%. These are all older, low-end discrete GPUs, and the Arc B390's margins over them are small. The data suggests the integrated Arc B390 performs comparably to those parts, with a slight edge.

For the NVIDIA H20 NVL16, the database records zero benchmark scores and an average benchmark score of zero. Its percentile of 50 reflects the database's default treatment of products without measurements rather than an actual performance ranking. Without recorded results, the data cannot quantify the H20 NVL16's graphics performance, and its lack of graphics APIs makes such testing unlikely.

The specification differences provide the only meaningful comparison. The H20 NVL16's FP32 throughput of 39.54 TFLOPS is 5.15 times the Arc B390's 7.680 TFLOPS. Its texture rate of 617.8 GTexel/s is 5.15 times the Arc B390's 120.0 GTexel/s. Its shading unit count of 9984 is 6.5 times the Arc B390's 1536. The H20 NVL16's memory bandwidth of 4.03 TB/s is a different order of magnitude from the Arc B390's system-dependent shared memory. The pixel rates are closer: the Arc B390 delivers 60.00 GPixel/s while the H20 NVL16 delivers 47.52 GPixel/s, a rare category where the integrated part leads. The raster operation unit counts are identical at 24.

Clock speeds tell another story. The Arc B390 boosts to 2500 MHz, which is 26% higher than the H20 NVL16's 1980 MHz boost clock. The Arc B390's base clock of 300 MHz is far lower than the H20 NVL16's 1830 MHz base, but the integrated part's boost behavior compensates. The H20 NVL16's higher sustained clocks at 400 W enable its compute throughput, while the Arc B390's bursty boost behavior suits mobile workloads.

Specification Differences

The Intel Arc B390 and NVIDIA H20 NVL16 differ in nearly every recorded specification field.

Chip and process: The Arc B390 uses the Panther Lake chip on a 3 nm process from Intel. The H20 NVL16 uses the GH100 chip on a 5 nm process from TSMC.

Transistors and die size: The Arc B390's transistor count and die size are unknown. The H20 NVL16 contains 80,000 million transistors on an 814 mm² die, with a transistor density of 98.3M per mm².

Clock speeds: The Arc B390 runs at 300 MHz base and 2500 MHz boost. The H20 NVL16 runs at 1830 MHz base and 1980 MHz boost, with memory at 1313 MHz and 5.3 Gbps effective.

Memory: The Arc B390 uses system-shared memory with system-dependent bandwidth. The H20 NVL16 uses 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 NVL16 has 9984 shading units, 312 TMUs, 24 ROPs, and 312 tensor cores.

Rates: The Arc B390 delivers 60.00 GPixel/s and 120.0 GTexel/s. The H20 NVL16 delivers 47.52 GPixel/s and 617.8 GTexel/s.

Performance: The Arc B390 offers 7.680 TFLOPS FP32 and 15.36 TFLOPS FP16 (2:1). The H20 NVL16 offers 39.54 TFLOPS FP32 and 79.07 TFLOPS FP16 (2:1).

Power and form factor: The Arc B390 has an 80 W TDP, IGP slot width, no power connectors, and an IGP bus interface. The H20 NVL16 has a 400 W TDP, SXM Module slot width, an 800 W suggested power supply, and PCIe 5.0 x16 interface.

Outputs and APIs: The Arc B390 supports portable-device-dependent display outputs, DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The H20 NVL16 has no outputs and lists all three APIs as N/A.

Release dates: The Arc B390 was released on 2026-01-26. The H20 NVL16 was released on 2025-09-01. The H20 NVL16 lists its predecessor as Server Ada and its successor as Server Blackwell. The Arc B390 lists no predecessor or successor.