Intel Arc B390 vs NVIDIA RTX PRO 2000 Blackwell Comparison

# Intel Arc B390 vs NVIDIA RTX PRO 2000 Blackwell

The Intel Arc B390 and NVIDIA RTX PRO 2000 Blackwell occupy sharply different positions in the database, as their benchmark scores and percentiles demonstrate. The NVIDIA part sits in the 70th percentile of all GPUs, while the Intel integrated solution lands in the 9th percentile. Across the single recorded head-to-head benchmark, the NVIDIA card wins decisively, yet the two parts target fundamentally different use cases: one is a mobile integrated processor, the other a dedicated workstation card.

Where Each One Wins

The Intel Arc B390 is an integrated graphics processor (IGP) designed for portable devices. Its benchmark results place it near the very bottom of the database, but that is not a failure of execution; it is a design constraint. An IGP shares system memory, has no dedicated cooling solution, and must fit within the power envelope of a mobile processor. The Arc B390's 3DMark Steel Nomad DX12 score of 1482 puts it just ahead of a cluster of older discrete low-end cards, including the NVIDIA GeForce GT 520MX (1463, 1.3% behind), the NVIDIA GeForce 800M (1460, 1.5% behind), and the NVIDIA GeForce GT 625 OEM (1446, 2.5% behind). These are not modern gaming GPUs; they are legacy parts from over a decade ago. The Arc B390's win here is modest, but it shows the integrated part can outrun some ancient dedicated hardware in a modern DirectX 12 workload.

The NVIDIA RTX PRO 2000 Blackwell wins where it matters for professional work. Its 3DMark Steel Nomad score of 2374.5 is 60.1% higher than the Arc B390's 1482, a delta of 37.6% in NVIDIA's favor. In Geekbench compute tests, the NVIDIA card posts 106087 in OpenCL and 113865 in Vulkan, scores that reflect serious parallel processing capability. Passmark results tell a similar story: 20049 in G3D, 8396 in GPU compute, and 241 in DirectX 9. The NVIDIA part also sits near a group of much more powerful GPUs, within 2% of the AMD Radeon RX 6700M (25633), the AMD Radeon Pro W5700 (25726), and the NVIDIA GeForce RTX 3080 Ti Mobile (25740). The Arc B390 has no equivalent standing; its nearest rivals are all far less capable.

The database records zero benchmark wins for the Intel part and one for the NVIDIA part. For a user seeking a discrete workstation GPU, the RTX PRO 2000 is the only rational choice. For a user who needs graphics on a thin, fanless portable device, the Arc B390 is the only option that fits.

Architecture Differences

The Intel Arc B390 uses the Panther Lake chip built on Intel's 3 nm process, employing the Xe3-LPG architecture. It belongs to the Arc Graphics-M (Panther Lake) generation. The NVIDIA RTX PRO 2000 Blackwell uses the GB206 chip on TSMC's 5 nm process, with the Blackwell 2.0 architecture, part of the Blackwell PRO W (x000) generation. The transistor counts could not be more different: NVIDIA lists 21,900 million transistors on a 181 mm² die, yielding a density of 121.0M per mm². Intel's transistor count and die size are not recorded, but the process node difference (3 nm vs 5 nm) suggests Intel is chasing density on paper, while NVIDIA has shipped a much larger, more complex chip.

The shading unit counts reflect the fundamental design split. The Arc B390 has 1536 shading units, 48 texture mapping units, and 24 raster output units. The RTX PRO 2000 has 4352 shading units, 136 TMUs, and 48 ROPs. That is 2.83 times more shading units, 2.83 times more TMUs, and exactly double the ROPs. The NVIDIA part also carries 34 ray tracing cores and 136 tensor cores, while the Intel part has 12 ray tracing cores and no tensor core count listed. For machine learning or AI workloads, the tensor cores are essential; the Arc B390 offers no such hardware.

Memory architecture diverges completely. The Arc B390 uses system shared memory, with a system-dependent bandwidth and no dedicated VRAM. The RTX PRO 2000 has 16 GB of GDDR7 on a 128-bit bus, delivering 288.0 GB/s of bandwidth. That dedicated memory pool is a critical advantage for professional applications that need to hold large datasets or render complex scenes without spilling to system RAM. The NVIDIA card's memory clock is listed at 1125 MHz (18 Gbps effective), while the Intel part's memory clock is "System Shared", meaning it fluctuates with the host system.

Clock speeds also tell a story of different design philosophies. The Arc B390 has a base clock of 300 MHz and a boost of 2500 MHz, a wide range that lets it idle at very low power and ramp up when needed. The RTX PRO 2000 has a base of 982 MHz and a boost of 1957 MHz, a narrower range that reflects a dedicated card running at more sustained clocks. The NVIDIA card's pixel rate of 93.94 GPixel/s and texture rate of 266.2 GTexel/s dwarf the Arc B390's 60.00 GPixel/s and 120.0 GTexel/s. In raw FP32 throughput, the NVIDIA card delivers 17.03 TFLOPS versus 7.680 TFLOPS for Intel, more than double. For FP16, the NVIDIA part does 17.03 TFLOPS (1:1 ratio), while the Intel part does 15.36 TFLOPS (2:1 ratio), meaning NVIDIA's FP16 is not just faster in absolute terms but also does not sacrifice rate for precision.

Head-to-Head Benchmarks

The database contains one direct comparison: 3DMark Steel Nomad DX12. The NVIDIA RTX PRO 2000 Blackwell scores 2374.5, the Intel Arc B390 scores 1482. The delta is 37.6% in NVIDIA's favor, which is a substantial margin for a single workload. To frame it differently, NVIDIA's score is 60.1% higher than Intel's absolute number. This is not a near-run contest; the NVIDIA card finishes far ahead.

The Geekbench and Passmark results for NVIDIA provide additional context, though no direct Intel equivalents are recorded. The Geekbench OpenCL score of 106087 and Vulkan score of 113865 indicate strong compute performance across different APIs. The Passmark DirectX 11 score of 174 versus DirectX 9 score of 241 shows that the NVIDIA card performs relatively better on older APIs, which is common for professional cards optimized for legacy software compatibility. The Passmark G2D score of 1303 reflects 2D desktop work, where the card is clearly competent.

The nearest rival data reinforces the NVIDIA card's tier. It sits within 2% of the AMD Radeon RX 6700M (25633, 1.4% ahead of NVIDIA), the AMD Radeon Pro W5700 (25726, 1.8% ahead), and the NVIDIA GeForce RTX 3080 Ti Mobile (25740, 1.8% ahead). It is 2% ahead of the NVIDIA RTX A5000 Mobile (24763). These are all midrange-to-high-end mobile or workstation parts, not ancient integrated graphics. The Arc B390's nearest rivals, by contrast, are the NVIDIA GeForce GT 520MX, GeForce 800M, GT 625 OEM, and GT 710, all of which score between 1443 and 1463. The Arc B390 beats them by 1.3% to 2.7%, a tiny margin against hardware that is generations old.

FAQ

Q: Which GPU has a higher 3DMark Steel Nomad DX12 score?

A: The NVIDIA RTX PRO 2000 Blackwell scores 2374.5, while the Intel Arc B390 scores 1482. NVIDIA leads by 37.6%.

Q: How much memory does each GPU have?

A: The Intel Arc B390 uses system shared memory with no dedicated VRAM. The NVIDIA RTX PRO 2000 Blackwell has 16 GB of GDDR7 on a 128-bit bus with 288.0 GB/s bandwidth.

Q: What are the power consumption figures?

A: The Intel Arc B390 has a TDP of 80 W, while the NVIDIA RTX PRO 2000 Blackwell has a TDP of 70 W. NVIDIA's suggested PSU is 250 W; Intel lists no suggested PSU.

Q: Which GPU has more shading units?

A: The NVIDIA RTX PRO 2000 Blackwell has 4352 shading units, compared to 1536 on the Intel Arc B390. NVIDIA also has 136 TMUs and 48 ROPs versus Intel's 48 TMUs and 24 ROPs.

Q: What is the difference in FP32 compute?

A: The NVIDIA RTX PRO 2000 Blackwell delivers 17.03 TFLOPS, while the Intel Arc B390 delivers 7.680 TFLOPS. NVIDIA is more than double.

Q: Do both GPUs support DirectX 12 Ultimate?

A: Yes, both list DirectX 12 Ultimate (12_2) support, along with OpenGL 4.6 and Vulkan 1.4.

Specification Differences

| Specification | Intel Arc B390 | NVIDIA RTX PRO 2000 Blackwell |

|---|---|---|

| Process node | 3 nm | 5 nm |

| Foundry | Intel | TSMC |

| Transistors | unknown | 21,900 million |

| Die size | unknown | 181 mm² |

| Shading units | 1536 | 4352 |

| TMUs | 48 | 136 |

| ROPs | 24 | 48 |

| RT cores | 12 | 34 |

| Tensor cores | null | 136 |

| Base clock | 300 MHz | 982 MHz |

| Boost clock | 2500 MHz | 1957 MHz |

| Memory size | System Shared | 16 GB |

| Memory type | System Shared | GDDR7 |

| Memory bus width | System Shared | 128 bit |

| Memory bandwidth | System Dependent | 288.0 GB/s |

| FP32 | 7.680 TFLOPS | 17.03 TFLOPS |

| FP16 | 15.36 TFLOPS (2:1) | 17.03 TFLOPS (1:1) |

| Pixel rate | 60.00 GPixel/s | 93.94 GPixel/s |

| Texture rate | 120.0 GTexel/s | 266.2 GTexel/s |

| TDP | 80 W | 70 W |

| Slot width | IGP | Dual-slot |

| Power connectors | None | None |

| Suggested PSU | null | 250 W |

| Bus interface | IGP | PCIe 5.0 x8 |

| Display outputs | Portable Device Dependent | 4x mini-DisplayPort 2.1b |

| Dimensions | null | 167 mm 6.6 inches length, 69 mm 2.7 inches height, 20 mm 0.8 inches width |

| Release date | 2026-01-26 | 2025-08-10 |

| Predecessor | null | Workstation Ada |