Intel Arc B390 vs Intel Data Center GPU Max Subsystem Comparison

The Verdict

The database presents a stark contrast between two Intel GPUs that occupy entirely different segments of the market. The Intel Arc B390 is an integrated graphics solution built on the Panther Lake chip with Xe3-LPG architecture, designed for portable devices where power efficiency and compactness are paramount. The Intel Data Center GPU Max Subsystem is a massive accelerator built on the Ponte Vecchio chip with Generation 12.5 architecture, engineered for data center workloads requiring enormous memory capacity and compute throughput.

The recorded benchmark data shows the Arc B390 scores 1482 points in the 3DMark Steel Nomad DX12 test, placing it in the 9th percentile among all GPUs. Its nearest rivals in the database are legacy NVIDIA parts: the GeForce GT 520MX with an average score of 1463, the GeForce 800M with 1460, the GeForce GT 625 OEM with 1446, and the GeForce GT 710 with 1443. The Arc B390 leads these comparisons by margins ranging from 1.3 percent to 2.7 percent. These are narrow victories against very old hardware, which indicates the Arc B390 delivers entry-level performance appropriate for basic graphics tasks in a portable form factor.

The Data Center GPU Max Subsystem has no recorded benchmark scores and no nearest rivals in the database. Its percentile ranking of 50 reflects an absence of comparative data rather than a performance assessment. The subsystem targets a completely different workload profile, one measured by training throughput, inference latency, and memory bandwidth rather than gaming frame rates. The absence of benchmarks for this part means the database cannot place it on the same performance scale as the Arc B390.

The data indicates the Arc B390 suits users who need integrated graphics for a lightweight portable device, with performance modestly ahead of decade-old discrete GPUs. The Data Center GPU Max Subsystem serves organizations requiring massive parallel compute, 128 GB of HBM2e memory, and 3.21 TB/s of bandwidth. These products do not compete with each other, and the benchmark results confirm they should not be compared directly.

Head-to-Head Benchmarks

The head-to-head benchmark table in the database contains no entries, meaning no direct comparison tests exist between the Arc B390 and the Data Center GPU Max Subsystem. The Arc B390 has exactly one benchmark result: 1482 in 3DMark Steel Nomad DX12. The Data Center GPU Max Subsystem has zero recorded benchmark scores.

The Arc B390's single score places it in the 9th percentile of all GPUs tracked by the database. Its nearest rival, the GeForce GT 520MX, averages 1463 points, a difference of 1.3 percent. The GeForce 800M follows at 1460, a 1.5 percent gap. The GeForce GT 625 OEM scores 1446, trailing by 2.5 percent, and the GeForce GT 710 scores 1443, trailing by 2.7 percent. These margins are small, indicating the Arc B390 performs at a level comparable to early-2010s entry-level discrete graphics.

The Data Center GPU Max Subsystem has no comparable data. Its average benchmark score of 0 in the database reflects the absence of submitted results, not a measured performance of zero. The subsystem's specifications suggest it would dominate compute-oriented workloads, but without recorded scores, the database cannot quantify that advantage relative to any other product.

The pixel rate figures illustrate the different design philosophies. The Arc B390 achieves 60.00 GPixel/s, while the Data Center GPU Max Subsystem records 0 MPixel/s. This disparity stems from the subsystem having zero ROPs, meaning it is not designed to rasterize traditional graphics output. The Arc B390 has 24 ROPs, enabling display output for portable devices. The subsystem offers no display outputs at all, confirming it is a compute accelerator rather than a graphics card.

Texture rates also diverge sharply. The Arc B390 delivers 120.0 GTexel/s from its 48 texture mapping units. The Data Center GPU Max Subsystem achieves 1,638.4 GTexel/s from 1024 TMUs, a 13.6 times higher texture throughput. This enormous difference reflects the subsystem's 16384 shading units versus the Arc B390's 1536, a factor of 10.7.

Architecture Differences

The Arc B390 uses the Panther Lake chip built on Intel's 3 nm process node with Xe3-LPG architecture, belonging to the Arc Graphics-M (Panther Lake) generation. The Data Center GPU Max Subsystem uses the Ponte Vecchio chip on a 10 nm Intel process with Generation 12.5 architecture, belonging to the Data Center GPU (Ponte Vecchio) generation. The process node difference is substantial: 3 nm versus 10 nm, giving the Arc B390 a significant density and efficiency advantage per transistor.

The transistor counts differ enormously. The Data Center GPU Max Subsystem packs 100,000 million transistors on a 1280 mm² die, yielding a transistor density of 78.1 million per square millimeter. The Arc B390's transistor count and die size are not recorded in the database, so no direct density comparison is possible. The subsystem's die size alone exceeds the physical dimensions of many entire graphics cards.

Memory configurations could not be more different. The Arc B390 uses system shared memory, with size, type, bus width, and bandwidth all dependent on the host portable device. The Data Center GPU Max Subsystem carries 128 GB of HBM2e memory on an 8192-bit bus, delivering 3.21 TB/s of bandwidth. The memory clock runs at 1565 MHz, with 3.1 Gbps effective data rate. This memory subsystem is designed for massive data movement in AI training and scientific computing, whereas the Arc B390 relies on whatever system memory the host provides.

Clock speeds show an interesting inversion. The Arc B390 has a 300 MHz base clock and a 2500 MHz boost clock. The Data Center GPU Max Subsystem runs at 900 MHz base and 1600 MHz boost. Despite the subsystem's higher base clock, its boost clock is substantially lower than the Arc B390's. The Arc B390's higher boost clock reflects its ability to ramp up quickly for burst workloads, while the subsystem maintains more conservative clocks across its massive array of compute units.

Compute throughput numbers demonstrate the performance gap. The Arc B390 delivers 7.680 TFLOPS of FP32 performance and 15.36 TFLOPS of FP16 using a 2:1 ratio. The Data Center GPU Max Subsystem achieves 52.43 TFLOPS of FP32 and 52.43 TFLOPS of FP16 at a 1:1 ratio. The subsystem offers 6.8 times more FP32 throughput and 3.4 times more FP16 throughput, with the 1:1 FP16 ratio indicating full-rate half-precision support crucial for AI workloads.

Ray tracing hardware also differs. The Arc B390 includes 12 ray tracing cores, appropriate for consumer gaming and graphical effects. The Data Center GPU Max Subsystem contains 128 ray tracing cores, but with zero ROPs and no display outputs, these cores likely serve compute-oriented ray tracing workloads rather than interactive graphics.

Power and physical specifications highlight the segment gap. The Arc B390 consumes 80 W, uses no power connectors, occupies a single IGP slot, and connects via an integrated graphics bus. The Data Center GPU Max Subsystem draws 2400 W, requires a 1x 16-pin power connector, needs a 2800 W suggested power supply, occupies a dual-slot form factor, measures 267 mm (10.5 inches) in length, and connects via PCIe 5.0 x16. The subsystem consumes 30 times more power than the Arc B390.

API support differs in important ways. The Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The Data Center GPU Max Subsystem supports DirectX 12 (12_1) and OpenGL 4.6, but the database records no Vulkan support. The Arc B390's newer DirectX feature level reflects its consumer graphics focus, while the subsystem's older DirectX version aligns with its compute-first design.

Release dates show a three-year gap. The Data Center GPU Max Subsystem launched on January 9, 2023, while the Arc B390 launched on January 26, 2026. Both remain in active production status according to the database. The subsystem has a recorded successor, the H3C Graphics, while the Arc B390 has no successor listed.

FAQ

Q: How does the Intel Arc B390 compare to its nearest rivals in benchmark performance?

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

Q: Why does the Data Center GPU Max Subsystem have no benchmark scores?

A: The database records zero benchmarks for this product, and its average benchmark score is 0. No nearest rivals are listed either. The subsystem's compute-oriented design, with zero ROPs and no display outputs, means standard graphics benchmarks likely do not apply to its workload profile.

Q: What memory does each GPU use?

A: The Arc B390 uses system shared memory, with size, type, bus width, and bandwidth all dependent on the host device. The Data Center GPU Max Subsystem has 128 GB of HBM2e memory on an 8192-bit bus with 3.21 TB/s bandwidth.

Q: What are the clock speeds of these two GPUs?

A: The Arc B390 runs at 300 MHz base and 2500 MHz boost. The Data Center GPU Max Subsystem runs at 900 MHz base and 1600 MHz boost.

Q: What power requirements does each GPU have?

A: The Arc B390 consumes 80 W, uses no power connectors, and requires no suggested power supply. The Data Center GPU Max Subsystem consumes 2400 W, uses a 1x 16-pin power connector, and requires a 2800 W suggested power supply.

Q: What is the physical size difference between the two?

A: The Arc B390 is an integrated graphics processor with IGP slot width and no dimensions recorded. The Data Center GPU Max Subsystem is a dual-slot card measuring 267 mm (10.5 inches) in length.

Where Each One Wins

The Arc B390 wins in scenarios where integrated graphics are required. Its IGP form factor, 80 W power consumption, and lack of external power connectors make it suitable for portable devices where space and battery life constrain the design. The 300 MHz base clock and 2500 MHz boost clock allow it to idle efficiently and ramp quickly when needed. Its 12 ray tracing cores and DirectX 12 Ultimate support enable modern graphics features in a low-power package. The 60.00 GPixel/s pixel rate and 120.0 GTexel/s texture rate provide adequate throughput for basic rendering tasks on a portable display. Its 1536 shading units and 24 ROPs deliver the 1482 Steel Nomad score, placing it in the 9th percentile but ahead of its nearest legacy rivals by modest margins. The Vulkan 1.4 and OpenGL 4.6 support broaden its compatibility across applications.

The Data Center GPU Max Subsystem wins in compute-intensive data center workloads. Its 16384 shading units, 1024 TMUs, and 128 ray tracing cores provide massive parallel throughput. The 52.43 TFLOPS of FP32 and FP16 performance, with the latter at 1:1 ratio, supports AI training and inference workloads that benefit from full-rate half-precision arithmetic. The 128 GB HBM2e memory with 3.21 TB/s bandwidth enables processing of large datasets that would exceed the memory capacity of consumer GPUs. The 1,638.4 GTexel/s texture rate indicates strong fill-rate capability for compute kernels that leverage texture units. The 900 MHz base clock across such a large chip provides sustained throughput for long-running workloads. The PCIe 5.0 x16 interface allows high-bandwidth communication with host systems. The 2400 W power envelope, while enormous, enables the subsystem to maintain high utilization across its compute array. The 267 mm physical length and dual-slot design fit standard server chassis.

The Arc B390 also wins in software compatibility for graphics applications. Its DirectX 12 Ultimate (12_2) support represents the latest DirectX feature level, while the Data Center GPU Max Subsystem's DirectX 12 (12_1) is an earlier version. The Arc B390's Vulkan 1.4 support expands its reach to Vulkan-based applications, where the subsystem has no recorded Vulkan support. For tasks involving display output, the Arc B390's 24 ROPs and 60.00 GPixel/s pixel rate provide actual rendering capability, while the subsystem's 0 MPixel/s pixel rate and absence of display outputs preclude any visual output function.

The Data Center GPU Max Subsystem wins in memory capacity and bandwidth by an enormous margin. Its 128 GB HBM2e memory dwarfs the Arc B390's system shared memory, and its 3.21 TB/s bandwidth exceeds what any integrated solution can access. The 8192-bit memory bus enables this bandwidth, while the Arc B390's memory bus width is system dependent. The subsystem's 100,000 million transistors on a 1280 mm² die provide the hardware resources necessary for large-scale parallel computing, whereas the Arc B390's transistor count is not recorded.

In raw compute throughput, the Data Center GPU Max Subsystem dominates. Its 52.43 TFLOPS FP32 is 6.8 times the Arc B390's 7.680 TFLOPS. Its FP16 performance of 52.43 TFLOPS at 1:1 ratio exceeds the Arc B390's 15.36 TFLOPS at 2:1 ratio by a factor of 3.4, and the 1:1 ratio means no throughput penalty for half-precision workloads. The subsystem's 128 ray tracing cores versus the Arc B390's 12 represents a 10.7 times advantage in ray tracing hardware, though the subsystem lacks the ROPs to render the results to a display.

The Arc B390 wins on efficiency per watt. Its 80 W power draw delivers 7.680 TFLOPS of FP32, yielding 96 GFLOPS per watt. The Data Center GPU Max Subsystem's 2400 W power draw delivers 52.43 TFLOPS of FP32, yielding 21.8 GFLOPS per watt. The Arc B390 is approximately 4.4 times more efficient in FP32 throughput per watt, a direct consequence of its 3 nm process node versus the subsystem's 10 nm node.

The Data Center GPU Max Subsystem wins in raw capability, the Arc B390 wins in efficiency and integration. The database records no head-to-head benchmarks between them, and their specifications place them in different product categories entirely. The Arc B390 serves the portable device market, while the Data Center GPU Max Subsystem serves the data center compute market. Each product wins in its respective domain.