Intel Arc Pro B390 vs Intel Arc Pro B65 Comparison

Where Each One Wins

The Intel Arc Pro B390 and Intel Arc Pro B65 occupy different positions in the professional graphics lineup, and the recorded data shows a clear split in their intended workloads. The B390 is an integrated graphics processor built for compact, power-limited systems where the CPU package must carry the display and compute duties. The B65 is a discrete, dual-slot add-in card with dedicated memory and a substantially higher performance ceiling.

Based on the database measurements, the B65 wins in every scenario that demands sustained raw throughput, large frame buffers, or high-bandwidth memory access. Its 12.29 TFLOPS FP32 throughput is roughly 60% higher than the B390's 7.680 TFLOPS, which translates directly into faster rendering, simulation, and compute workloads. The B65 also carries 32 GB of dedicated GDDR6 memory on a 256-bit bus, delivering 608.0 GB/s of bandwidth. The B390, by contrast, shares system memory with the host CPU, and its bandwidth is listed as "System Dependent," meaning performance scales with whatever memory the host platform provides.

The B390 wins in scenarios where power draw, physical footprint, and system simplicity matter more than absolute performance. Its 80 W TDP is 120 W lower than the B65's 200 W figure. It requires no power connectors, occupies no expansion slot, and uses the IGP bus interface. For thin-and-light professional laptops, mini PCs, or embedded workstations where a discrete card cannot physically fit, the B390 is the only option of the two. The B65 needs a PCIe 5.0 x16 slot, a dual-slot chassis position, and a single 8-pin power connector, plus a suggested power supply of 550 W.

The B65 also wins on output flexibility for multi-display professional setups. It provides 4x DisplayPort 2.1 outputs, while the B390's display outputs are "Portable Device Dependent," meaning the number and type of ports depend entirely on the host device's design. A workstation with the B65 can drive four monitors directly from the card. A system with the B390 can only drive whatever the laptop or motherboard manufacturer chose to expose.

Architecture Differences

The two GPUs come from different architectural families and are built on different process nodes at different foundries. The B390 uses the Xe3-LPG architecture and is fabricated on Intel's 3 nm process at Intel's own fabs. Its chip is codenamed Panther Lake, and it belongs to the Arc Graphics-WM (Panther Lake) generation. The B65 uses the Xe2-HPG architecture, fabricated on a 5 nm process at TSMC, with the BMG-G21 chip from the Battlemage (Pro Series) generation.

The transistor counts tell a story of scale. The B65's die contains 19,600 million transistors across a 272 mm² die, giving a transistor density of 72.1M per mm². The B390's transistor count and die size are listed as unknown in the database, so a direct density comparison is not possible from the recorded data. What is clear is that the B65 is a large, high-power discrete GPU, while the B390 is an integrated part designed to share the Panther Lake processor's power envelope.

Core counts differ across every execution unit type. The B390 has 1536 shading units, 48 texture mapping units, 24 ROPs, and 12 ray tracing cores. The B65 has 2560 shading units, 160 TMUs, 80 ROPs, and 20 ray tracing cores. The B65's advantage in texture and pixel throughput is proportionally larger than its shader advantage: it delivers 384.0 GTexel/s and 192.0 GPixel/s, versus 120.0 GTexel/s and 60.00 GPixel/s for the B390. That means the B65 has 3.2x the texture rate and 3.2x the pixel rate, while its FP32 rate is only about 1.6x higher. The B65 is built for fill-rate-bound rasterization work, not just compute.

Clock behavior also differs. The B390 has a base clock of 300 MHz and a boost clock of 2500 MHz, a wide dynamic range typical of integrated GPUs that need to idle at very low power. The B65 runs at a flat 2400 MHz for both base and boost, with no dynamic spread. Its memory clock is 2375 MHz, which translates to 19 Gbps effective on the GDDR6 interface.

Both GPUs support the same API feature set: DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Neither lists tensor cores in the database. The B390's memory type is listed as "System Shared," and its bus width is likewise "System Shared," meaning the integrated design borrows the host's memory controller and bus.

Head-to-Head Benchmarks

The database's head-to-head benchmark array is empty for this pair, so no direct per-application scores exist. However, the specification-level data provides a clear quantitative picture of where each part wins.

The largest single gap is memory bandwidth. The B65 delivers 608.0 GB/s from its dedicated GDDR6 memory. The B390's bandwidth is "System Dependent," so it cannot be assigned a fixed number in the database. In a laptop with dual-channel LPDDR5X memory, the available bandwidth is typically a fraction of what the B65 provides. For workloads that stream large textures, run neural network inference on GPU tensors, or manipulate large datasets in VRAM, the B65's fixed 608.0 GB/s is a decisive advantage.

The FP32 compute gap is 4.61 TFLOPS in favor of the B65 (12.29 versus 7.680). That is a 60% difference. In FP16, the gap widens: the B65 reaches 24.58 TFLOPS (2:1) versus 15.36 TFLOPS (2:1) for the B390, a 9.22 TFLOPS advantage. Applications that use FP16 for machine learning or image processing will see a proportionally larger benefit from the B65.

Pixel and texture rates show the B65's largest relative wins. The B65's 192.0 GPixel/s is 3.2x the B390's 60.00 GPixel/s. Its 384.0 GTexel/s is likewise 3.2x the B390's 120.0 GTexel/s. For 4K rendering, heavy post-processing chains, or high-resolution viewport work, the B65 will finish these stages in roughly one-third of the time.

The B390 does hold one clock-speed advantage: its 2500 MHz boost clock is 100 MHz above the B65's flat 2400 MHz. With fewer cores and a much lower base clock of 300 MHz, this boost headroom helps the integrated part hit responsive burst performance in light 2D workloads, but it does not compensate for the B65's larger execution resources.

Power efficiency is not a win for either part in absolute terms, but the ratio is stark. The B65 delivers 12.29 TFLOPS at 200 W, which is 61.45 GFLOPS per watt. The B390 delivers 7.680 TFLOPS at 80 W, which is 96.0 GFLOPS per watt. The integrated part is roughly 56% more efficient per watt in FP32 throughput, though the B65 still provides 60% more absolute throughput.

Specification Differences

The table below lists only the fields where the two GPUs differ, based on the database records.

| Field | Intel Arc Pro B390 | Intel Arc Pro B65 |

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

| Chip | Panther Lake | BMG-G21 |

| Architecture | Xe3-LPG | Xe2-HPG |

| Generation | Arc Graphics-WM (Panther Lake) | Battlemage (Pro Series) |

| Process node | 3 nm | 5 nm |

| Foundry | Intel | TSMC |

| Transistors | unknown | 19,600 million |

| Die size | unknown | 272 mm² |

| Transistor density | null | 72.1M / mm² |

| Base clock | 300 MHz | 2400 MHz |

| Boost clock | 2500 MHz | 2400 MHz |

| Memory clock | System Shared | 2375 MHz, 19 Gbps effective |

| Memory size | System Shared | 32 GB |

| Memory type | System Shared | GDDR6 |

| Memory bus width | System Shared | 256 bit |

| Memory bandwidth | System Dependent | 608.0 GB/s |

| Shading units | 1536 | 2560 |

| TMUs | 48 | 160 |

| ROPs | 24 | 80 |

| Ray tracing cores | 12 | 20 |

| Pixel rate | 60.00 GPixel/s | 192.0 GPixel/s |

| Texture rate | 120.0 GTexel/s | 384.0 GTexel/s |

| FP32 | 7.680 TFLOPS | 12.29 TFLOPS |

| FP16 | 15.36 TFLOPS (2:1) | 24.58 TFLOPS (2:1) |

| TDP | 80 W | 200 W |

| Slot width | IGP | Dual-slot |

| Power connectors | None | 1x 8-pin |

| Suggested PSU | null | 550 W |

| Bus interface | IGP | PCIe 5.0 x16 |

| Display outputs | Portable Device Dependent | 4x DisplayPort 2.1 |

| Release date | 2026-01-26 | 2026-03-31 |

| Predecessor | HD Graphics-WM | null |

Both parts share the same DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 API support. Both are listed as Active production status. Neither has a recorded launch MSRP in the database, and neither lists tensor cores.

FAQ

Q: Which GPU has higher raw compute throughput?

A: The Intel Arc Pro B65. Its FP32 throughput is 12.29 TFLOPS versus 7.680 TFLOPS for the B390, a 60% advantage. In FP16, the B65 reaches 24.58 TFLOPS versus 15.36 TFLOPS.

Q: Can the B390 be installed in a desktop tower as a discrete card?

A: No. The B390 uses the IGP bus interface, has an IGP slot width, requires no power connectors, and has an 80 W TDP. It is an integrated GPU that lives inside the host processor, not a card that occupies a PCIe slot.

Q: How much memory does each GPU have?

A: The B65 has 32 GB of dedicated GDDR6 memory on a 256-bit bus with 608.0 GB/s bandwidth. The B390 has no dedicated memory; its memory size, type, and bus width are all listed as "System Shared," and its bandwidth is "System Dependent."

Q: Which GPU supports more display outputs?

A: The B65 provides 4x DisplayPort 2.1 outputs directly from the card. The B390's display outputs are "Portable Device Dependent," so the number of outputs depends on the host laptop or motherboard design.

Q: Is the B390 more power-efficient than the B65?

A: In FP32 throughput per watt, yes. The B390 delivers 7.680 TFLOPS at 80 W (96.0 GFLOPS per watt), while the B65 delivers 12.29 TFLOPS at 200 W (61.45 GFLOPS per watt). The B65 still provides more absolute performance.

Q: Do both GPUs support the same graphics APIs?

A: Yes. Both support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Neither has tensor cores listed in the database.

The Verdict

The data points to a simple split. The Intel Arc Pro B65 is the choice for any workload where performance is the binding constraint. It offers 60% more FP32 throughput, 3.2x the pixel rate, 3.2x the texture rate, 32 GB of dedicated GDDR6 memory, and 608.0 GB/s of fixed bandwidth. It needs a PCIe 5.0 x16 slot, a dual-slot chassis, one 8-pin power connector, and a 550 W suggested power supply. It released later, on 2026-03-31, and uses the Xe2-HPG architecture on a 5 nm TSMC process with 19,600 million transistors on a 272 mm² die.

The Intel Arc Pro B390 is the choice for systems that cannot physically or electrically accommodate a 200 W dual-slot card. Its 80 W TDP, connector-free power design, and IGP form factor let it fit into Panther Lake-based portable and compact systems. It delivers 7.680 TFLOPS FP32, 120.0 GTexel/s, and 60.00 GPixel/s, with a 2500 MHz boost clock. Its performance is limited by shared system memory, and its display outputs depend on the host device. For a builder assembling a full-size workstation with a discrete GPU slot, the B65 is the only sensible pick. For a thin client, laptop, or embedded system where a discrete GPU is impossible, the B390 provides a functional integrated alternative. The two parts do not compete for the same socket, chassis, or power budget; they serve different physical design constraints.