Intel Arc B390 vs NVIDIA H200 SXM 141 GB Comparison

Head-to-Head Benchmarks

The recorded data presents an unusual comparison: the Intel Arc B390 has a single benchmark result in the database, while the NVIDIA H200 SXM 141 GB has no recorded benchmark scores. The Intel part scores 1482 points in 3DMark Steel Nomad DX12, placing it in the 9th percentile of all GPUs. This indicates that the Arc B390 performs at a level that is below the vast majority of installed graphics processors. The H200 SXM, with its 50th percentile placement and an average benchmark score of 0, sits at the median of all GPUs in the database, although this percentile is derived from no actual test results, suggesting its position is based on hardware specifications rather than measured performance.

The absence of head-to-head benchmark data means there are no direct comparisons with exact numbers. However, the nearest rivals for the Arc B390 provide context for its performance tier. The Arc B390 is 1.3% ahead of the NVIDIA GeForce GT 520MX, which scores 1463. It is 1.5% ahead of the NVIDIA GeForce 800M at 1460. Against the NVIDIA GeForce GT 625 OEM, the lead expands to 2.5%, with that card scoring 1446. The Arc B390 also holds a 2.7% advantage over the NVIDIA GeForce GT 710, which scores 1443. These deltas are small, indicating that the Arc B390 clusters with entry-level discrete GPUs from over a decade ago in this specific test. The H200 SXM has no nearest rivals listed, so no comparative benchmark deltas exist for it.

The data shows that in raw FP32 throughput, the H200 SXM delivers 66.91 TFLOPS, which is 8.7 times the 7.680 TFLOPS of the Arc B390. For FP16 with a 2:1 ratio, the H200 reaches 133.8 TFLOPS versus 15.36 TFLOPS for the Arc B390, a difference of roughly 8.7 times as well. Pixel rates are closer: the Arc B390 outputs 60.00 GPixel/s, while the H200 manages 47.52 GPixel/s, meaning the Intel part is 26.3% faster in pixel fill. Texture rates show a massive inversion, with the H200 achieving 1,045.4 GTexel/s compared to 120.0 GTexel/s for the Arc B390, a difference of 8.7 times. These numbers indicate that the two products are optimized for entirely different workloads, with the H200 dominating compute-heavy tasks while the Arc B390 holds a niche advantage in pixel throughput.

Architecture Differences

The Intel Arc B390 uses the Panther Lake chip built on a 3 nm process, fabricated by Intel itself. Its architecture is Xe3-LPG, belonging to the Arc Graphics-M generation. The NVIDIA H200 SXM uses the GH100 chip built on a 5 nm process, fabricated by TSMC, with a Hopper architecture from the Server Hopper generation. The transistor counts differ substantially: the H200 has 80,000 million transistors on a 814 mm² die, yielding a density of 98.3M per mm². The Arc B390 has an unknown transistor count and die size, so no density comparison is possible.

The Arc B390 integrates 1536 shading units, 48 texture mapping units, and 24 raster operation pipelines. It includes 12 ray tracing cores but has no dedicated tensor cores. The H200 SXM houses 16896 shading units, 528 TMUs, and 24 ROPs. It has no ray tracing cores but includes 528 tensor cores. This structural difference highlights that the H200 is designed for matrix math and AI acceleration, while the Arc B390 relies on its shader array for all work. The Arc B390 has a base clock of 300 MHz and a boost clock of 2500 MHz. The H200 runs at a base of 1500 MHz and boosts to 1980 MHz.

Memory configurations are fundamentally different. The Arc B390 uses system shared memory, with its type, bus width, and bandwidth all dependent on the host platform. The H200 SXM has 141 GB of HBM3e memory on a 6144-bit bus, delivering 4.89 TB/s of bandwidth. Clock speeds for memory are listed as "System Shared" for the Arc B390, while the H200 runs at 1593 MHz with 6.4 Gbps effective. The power envelopes show a 620 W gap, with the Arc B390 rated at 80 W and the H200 at 700 W. The Arc B390 is an integrated graphics processor with no power connectors and a slot width of IGP, while the H200 is an SXM module requiring an 8-pin EPS connector and a suggested 1100 W power supply.

API support diverges sharply. The Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The H200 SXM lists N/A for DirectX, OpenGL, and Vulkan, confirming it is not intended for graphics rendering in the traditional sense. Display outputs for the Arc B390 are portable device dependent, while the H200 has no outputs at all. The bus interface for the Arc B390 is IGP, meaning it communicates through the system, while the H200 uses PCIe 5.0 x16.

FAQ

Q: Which GPU has a higher pixel fill rate?

A: The Intel Arc B390 has a pixel rate of 60.00 GPixel/s, which is 26.3% higher than the NVIDIA H200 SXM's 47.52 GPixel/s.

Q: Does the NVIDIA H200 SXM support DirectX?

A: No. The database lists DirectX as N/A for the H200 SXM, along with OpenGL and Vulkan also being N/A. The Intel Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: What is the memory bandwidth difference between the two?

A: The H200 SXM has 141 GB of HBM3e memory with 4.89 TB/s bandwidth on a 6144-bit bus. The Arc B390 uses system shared memory with bandwidth that is system dependent, so no fixed figure exists.

Q: How many tensor cores does each GPU have?

A: The H200 SXM has 528 tensor cores. The Arc B390 has no tensor cores listed in the database.

Q: Which GPU was released later?

A: The Intel Arc B390 has a release date of 2026-01-26, while the NVIDIA H200 SXM has a release date of 2024-11-17. The Arc B390 is the later release.

Q: What are the process nodes for each chip?

A: The Intel Arc B390 uses a 3 nm process from Intel. The NVIDIA H200 SXM uses a 5 nm process from TSMC.

Specification Differences

The two GPUs differ across nearly every recorded specification. The Intel Arc B390 uses the Panther Lake chip with Xe3-LPG architecture, while the NVIDIA H200 SXM uses the GH100 chip with Hopper architecture. Process nodes are 3 nm for Intel and 5 nm for NVIDIA. Transistor count is unknown for the Arc B390, but the H200 lists 80,000 million transistors on an 814 mm² die. Base clocks are 300 MHz versus 1500 MHz, and boost clocks are 2500 MHz versus 1980 MHz. Memory size, type, bus width, and bandwidth all differ: system shared versus 141 GB HBM3e, system shared versus 6144-bit, and system dependent versus 4.89 TB/s.

Shading units number 1536 for the Arc B390 and 16896 for the H200. TMUs are 48 versus 528. ROPs are identical at 24. Ray tracing cores exist only on the Arc B390 at 12, while tensor cores exist only on the H200 at 528. Pixel rate favors the Arc B390 at 60.00 GPixel/s versus 47.52 GPixel/s. Texture rate favors the H200 massively at 1,045.4 GTexel/s versus 120.0 GTexel/s. FP32 compute is 7.680 TFLOPS for Intel and 66.91 TFLOPS for NVIDIA. FP16 is 15.36 TFLOPS versus 133.8 TFLOPS, both with a 2:1 ratio. TDP is 80 W versus 700 W. Slot width is IGP versus SXM Module. Power connectors are None versus 8-pin EPS. The suggested PSU for the H200 is 1100 W, while the Arc B390 has no such value. Bus interface is IGP versus PCIe 5.0 x16. Display outputs are portable device dependent versus no outputs. API support includes DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 for Intel, but all are N/A for NVIDIA.

The Verdict

The data indicates that the Intel Arc B390 and NVIDIA H200 SXM are not competing products. The Arc B390 is an integrated GPU designed for portable devices, evidenced by its IGP slot width, system shared memory, and lack of power connectors. Its single benchmark score of 1482 places it at the 9th percentile, clustered with legacy entry-level GPUs. The H200 SXM is a server module with 141 GB of HBM3e memory, 528 tensor cores, and 66.91 TFLOPS of FP32 compute, placing it at the 50th percentile despite having no benchmark results. The H200 lacks any graphics API support and has no display outputs, confirming its role as a compute accelerator rather than a graphics card.

The Arc B390 wins in pixel rate and supports modern graphics APIs. The H200 wins in every compute metric, memory capacity, bandwidth, and transistor count. The power difference is substantial: 80 W versus 700 W, with the H200 requiring an 8-pin EPS connector and an 1100 W power supply. The production status for both is Active, with the Arc B390 releasing in January 2026 and the H200 in November 2024. The H200 has identified predecessors and successors in Server Ada and Server Blackwell, while the Arc B390 has none listed. For a user seeking graphics rendering in a portable device, the Arc B390 is the only option with display outputs and API support. For server-side compute, the H200 provides the necessary memory and tensor core resources.

Where Each One Wins

The Intel Arc B390 wins in scenarios involving pixel output. Its 60.00 GPixel/s rate exceeds the H200's 47.52 GPixel/s by 26.3%. It also supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, making it functional for graphics rendering. The H200 has no API support. The Arc B390 operates at 80 W with no power connectors, fitting into integrated systems, while the H200 requires a 700 W budget and an 8-pin EPS connector. The Arc B390 has 12 ray tracing cores, which the H200 lacks entirely, and its boost clock of 2500 MHz is higher than the H200's 1980 MHz.

The NVIDIA H200 SXM wins in all compute-focused categories. Its FP32 output of 66.91 TFLOPS is approximately 8.7 times that of the Arc B390. FP16 performance follows the same ratio, with 133.8 TFLOPS versus 15.36 TFLOPS. Texture rate is 1,045.4 GTexel/s versus 120.0 GTexel/s, also an 8.7 times advantage. The H200 has 141 GB of HBM3e memory with 4.89 TB/s bandwidth, while the Arc B390 depends on system memory with no fixed bandwidth. The H200 has 528 tensor cores, enabling matrix operations that the Arc B390 cannot perform. Its 16896 shading units outnumber the Arc B390's 1536 by a factor of 11. The H200 uses a 5 nm process with 80,000 million transistors, while the Arc B390's transistor count is unknown. The H200's 50th percentile placement, despite no benchmark scores, reflects its high-end positioning, whereas the Arc B390's 9th percentile with an actual score shows its entry-level status. The 3DMark Steel Nomad result of 1482 for the Arc B390, with only small deltas over GT 520MX, 800M, GT 625 OEM, and GT 710, confirms its placement near the bottom of the performance distribution.