Intel Arc Pro B390 vs NVIDIA GB10 Comparison

FAQ

Q: What are the benchmark scores for the NVIDIA GB10?

A: The NVIDIA GB10 records a Geekbench OpenCL score of 120137 and a Geekbench Vulkan score of 114648. Its average benchmark score across recorded tests is 117393, placing it in the 95th percentile of all GPUs in the database.

Q: How does the NVIDIA GB10 compare to its nearest rivals in the database?

A: The GB10 is 0.3% ahead of the NVIDIA RTX 4000 SFF Ada Generation (average score 117088), 1.3% behind the AMD Radeon PRO W7700 (average score 118976), 2.6% ahead of the NVIDIA Tesla V100 SXM2 16 GB (average score 114395), and 3% ahead of the NVIDIA RTX A5500 Mobile (average score 113944).

Q: What is the Intel Arc Pro B390's benchmark standing?

A: The Intel Arc Pro B390 has no recorded benchmark scores in the database, and its average benchmark score is 0. It sits at the 50th percentile of all GPUs, indicating it has not been measured against the same workload suite used for the GB10.

Q: What are the clock speeds of each GPU?

A: The Intel Arc Pro B390 has a base clock of 300 MHz and a boost clock of 2500 MHz. The NVIDIA GB10 has a base clock of 1665 MHz and a boost clock of 2418 MHz, with a memory clock of 1067 MHz (8.5 Gbps effective).

Q: What memory configurations do these two GPUs use?

A: The Intel Arc Pro B390 uses system shared memory, with system dependent bandwidth. The NVIDIA GB10 has 128 GB of LPDDR5X memory on a 256-bit bus, delivering 273.2 GB/s of bandwidth.

Q: What is the power draw and form factor for each?

A: The Intel Arc Pro B390 has an 80 W TDP and is an integrated graphics processor (IGP) with no power connectors. The NVIDIA GB10 has a 140 W TDP, is also an IGP in slot width, has no power connectors, and carries a suggested PSU rating of 300 W.

The Verdict

The data shows a complete separation in purpose and performance tier between these two parts. The NVIDIA GB10 is a server-class Blackwell 2.0 part with recorded benchmark results, a 95th percentile standing, and an average score of 117393. The Intel Arc Pro B390 is an integrated graphics solution in the Arc Graphics-WM (Panther Lake) generation, with no recorded benchmarks and a 50th percentile placement. Any selection between them must rest on workload compatibility and platform requirements.

For compute-heavy server or workstation tasks, the GB10 is the only option with measured evidence of performance. Its nearest rivals are professional GPUs such as the RTX 4000 SFF Ada Generation and Radeon PRO W7700, and it trades within 1.3% of both. The GB10 also carries 384 tensor cores, which the Intel part lacks entirely. For applications that rely on NVIDIA's Blackwell architecture, CUDA-capable tensor operations, or the 128 GB LPDDR5X memory pool, the GB10 is the clear choice.

The Intel Arc Pro B390 targets a different segment. It is an IGP with no dedicated memory, no power connectors, and a 3 nm process node from Intel's own foundry. Its Xe3-LPG architecture supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it suitable for graphics workloads on systems without a discrete GPU. The B390 has 1536 shading units, 48 TMUs, 24 ROPs, and 12 ray tracing cores, with FP32 throughput of 7.680 TFLOPS. For an integrated solution, these are substantial resources, but without benchmark data, no direct performance comparison to the GB10 is possible.

The verdict is straightforward: the GB10 is the performance leader by every recorded metric, and the B390 is the low-power integrated alternative for systems where a discrete card is not an option. The GB10's 140 W TDP and suggested 300 W PSU make it unsuitable for the same compact platforms that would use an IGP. Conversely, the B390's shared memory and system dependent bandwidth mean it cannot match the GB10's dedicated 273.2 GB/s memory subsystem. Neither part replaces the other; they serve different hardware configurations and software ecosystems.

Head-to-Head Benchmarks

The head-to-head benchmark table in the database contains no entries, and neither GPU records a win in direct comparison tests. The only measurable performance data belongs to the NVIDIA GB10, which has two recorded Geekbench results: an OpenCL score of 120137 and a Vulkan score of 114648. The Intel Arc Pro B390 has no comparable scores, so a direct apples-to-apples comparison cannot be made from the recorded data.

What the data does show is the GB10's position among its nearest rivals. Its average score of 117393 puts it 0.3% above the RTX 4000 SFF Ada Generation at 117088, a margin of roughly 305 points. It trails the AMD Radeon PRO W7700 by 1.3%, a gap of 1583 points. Against the Tesla V100 SXM2 16 GB, the GB10 leads by 2.6%, or 2998 points. Against the RTX A5500 Mobile, it leads by 3%, or 3449 points. These deltas place the GB10 in a tight cluster of high-end professional GPUs, none of which it dominates by more than 3% in either direction.

The FP32 compute figures offer a separate comparison point. The GB10 delivers 29.71 TFLOPS, which is 3.9 times the 7.680 TFLOPS of the Arc Pro B390. The GB10's texture rate of 928.5 GTexel/s is 7.7 times the B390's 120.0 GTexel/s. Its pixel rate of 116.1 GPixel/s is 1.9 times the B390's 60.00 GPixel/s. These are raw specification comparisons, not benchmark results, but they illustrate the scale of the compute gap between a server GPU and an integrated part.

In ray tracing, the GB10 has 48 RT cores versus the B390's 12, a 4x difference. The GB10 also has 384 tensor cores while the B390 has none. The GB10's shading unit count of 6144 is 4x the B390's 1536. Every parallel compute resource favors the GB10 by a factor of at least 2, and in most cases by 4x or more. The B390's only advantages are its lower 80 W TDP, its integrated form factor, and its support for graphics APIs that the GB10 does not expose, since the GB10 lists DirectX, OpenGL, and Vulkan as N/A.

Specification Differences

The two GPUs differ across nearly every specification field in the database. The Intel Arc Pro B390 uses Intel's 3 nm process node and is fabricated by Intel. The NVIDIA GB10 uses a 5 nm node from TSMC. The GB10 has a known die size of 382 mm², while the B390's die size is not recorded. Neither part has a disclosed transistor count.

Clock behavior diverges significantly. The B390 has a base clock of 300 MHz and a boost clock of 2500 MHz. The GB10 has a base clock of 1665 MHz and a boost clock of 2418 MHz. The B390's boost clock is 82 MHz higher, but its base clock is 1365 MHz lower. The GB10's memory clock is 1067 MHz with 8.5 Gbps effective transfer, while the B390's memory clock is listed as system shared.

Memory capacity and bandwidth are entirely different. The B390 has system shared memory with system dependent bandwidth. The GB10 has 128 GB of LPDDR5X on a 256-bit bus with 273.2 GB/s of bandwidth. This is a dedicated memory pool, not shared with the system.

Compute unit counts differ by large margins. The B390 has 1536 shading units, 48 TMUs, 24 ROPs, and 12 RT cores. The GB10 has 6144 shading units, 384 TMUs, 48 ROPs, and 48 RT cores. The GB10 also has 384 tensor cores; the B390 has none. The B390's FP32 throughput is 7.680 TFLOPS and its FP16 is 15.36 TFLOPS at a 2:1 ratio. The GB10's FP32 is 29.71 TFLOPS and its FP16 is 29.71 TFLOPS at a 1:1 ratio, meaning full-rate FP16 on the GB10.

Power and physical requirements differ as well. The B390 has an 80 W TDP and no power connectors. The GB10 has a 140 W TDP, no power connectors, and a suggested PSU of 300 W. The B390 uses an IGP bus interface. The GB10 uses PCIe 5.0 x16. The B390's display outputs are portable device dependent. The GB10 has one HDMI output. The GB10 has physical dimensions of 150 mm by 51 mm by 150 mm, while the B390 lists no dimensions.

Release timing also differs. The GB10 was released on 2025-10-14, and the B390 on 2026-01-26. The B390's predecessor is HD Graphics-WM, and the GB10's predecessor is Server Hopper with a successor of Server Rubin. The GB10 has a launch MSRP of 3,999 USD. The B390 has no launch MSRP recorded.

Architecture Differences

The Intel Arc Pro B390 is built on the Xe3-LPG architecture, part of the Arc Graphics-WM (Panther Lake) generation, using the Panther Lake chip. The NVIDIA GB10 uses the Blackwell 2.0 architecture from the Server Blackwell (Bxx) generation, built on the GB20B chip. These are unrelated design lineages: one is Intel's integrated graphics evolution from HD Graphics-WM, the other is NVIDIA's server compute line descended from Server Hopper.

The B390's Xe3-LPG architecture includes 12 ray tracing cores and supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. It is a graphics-oriented design with no tensor cores and no dedicated memory interface. Its FP16 throughput doubles its FP32 rate, indicating a 2:1 FP16 ratio typical of consumer and integrated graphics parts. The 3 nm Intel process node is the smallest manufacturing node in this comparison.

The GB10's Blackwell 2.0 architecture includes 48 RT cores and 384 tensor cores, and it records FP16 at the same rate as FP32 (29.71 TFLOPS, 1:1 ratio). This full-rate FP16 is characteristic of compute-focused server GPUs. The architecture does not expose DirectX, OpenGL, or Vulkan APIs, which is consistent with a server part that targets compute workloads rather than graphics rendering. The GB10's die size of 382 mm² on TSMC's 5 nm process is recorded, while the B390's die size is unknown.

The memory architecture reflects the different design goals. The B390 uses system shared memory, meaning its bandwidth is system dependent and its memory size is whatever the host system provides. The GB10 has 128 GB of dedicated LPDDR5X with a fixed 273.2 GB/s bandwidth, a configuration suited to large in-memory datasets common in server workloads.

The tensor core count of 384 on the GB10 is the most significant architectural differentiator. The B390 has no tensor cores at all. This means the GB10 can execute matrix operations and AI workloads that the B390 cannot accelerate in hardware. The B390's 12 RT cores versus the GB10's 48 RT cores further separates the two in ray tracing throughput, though the B390's graphics API support means it can actually run DirectX and Vulkan workloads, which the GB10 cannot.

The bus interface also differs at the architecture level. The B390 is an IGP, integrated into the Panther Lake platform with no external bus. The GB10 uses PCIe 5.0 x16, making it a discrete-style part despite its IGP slot width classification. The GB10's 140 W TDP and suggested 300 W PSU confirm a board-level power delivery requirement, while the B390's 80 W TDP fits within a mobile or compact integrated platform. These architectural choices align with the release sequence: the GB10 launched in October 2025 as a server part, and the B390 followed in January 2026 as an integrated graphics option.