Intel Arc Pro B370 vs NVIDIA B300 Comparison

Where Each One Wins

The Intel Arc Pro B370 and NVIDIA B300 occupy entirely different corners of the GPU landscape, and the recorded data shows almost no overlap in their intended roles. The Arc Pro B370 is an integrated graphics processor (IGP) built for portable devices, while the B300 is a server-class SXM module designed for data center compute. The benchmark wins are not close contests; they are category separations.

The Intel part wins in any scenario that demands a self-contained graphics solution within a mobile or compact system. Its 25 W thermal design power makes it suitable for thin-and-light devices where power delivery is constrained. The Arc Pro B370 uses system-shared memory, meaning it pulls from the host's main memory rather than requiring dedicated VRAM. This simplifies the bill of materials for a portable device and allows the GPU to scale with available system memory. Its display outputs are listed as portable device dependent, reinforcing that this chip is meant to drive an integrated laptop or handheld display rather than external monitors in a rack.

The NVIDIA B300 wins in every compute-heavy scenario. Its 76.99 TFLOPS of FP32 performance is more than twelve times the Arc Pro B370's 6.144 TFLOPS. In FP16 workloads, the gap becomes enormous: the B300 delivers 1,231.8 TFLOPS (16:1 ratio) versus 12.29 TFLOPS (2:1 ratio) for the Intel part. The B300's 144 GB of HBM3e memory with a 4096-bit bus and 4.10 TB/s bandwidth is a massive advantage for large model inference, training, and scientific simulation. The Arc Pro B370's memory bandwidth is listed as system dependent, which in practice means it is limited by the host platform's memory subsystem, a fraction of the B300's dedicated high-bandwidth stack.

The B300 also wins on raw throughput metrics. Its texture rate of 1,202.9 GTexel/s dwarfs the Arc Pro B370's 96.00 GTexel/s. Its pixel rate of 48.77 GPixel/s slightly edges the Intel part's 48.00 GPixel/s, a rare metric where the two are comparable. The B300 packs 18,944 shading units, 592 tensor cores, and 592 texture mapping units, while the Arc Pro B370 has 1,280 shading units, 10 ray tracing cores, 40 TMUs, and 20 ROPs. The B300 reports no RT core count in the database, but its tensor core count alone (592) exceeds the Intel chip's total shading units by a wide margin.

The integrated Intel part has no power connectors, draws 25 W, and uses an IGP bus interface. The B300 requires a 1800 W suggested PSU, uses PCIe 5.0 x16, and has no display outputs. The Arc Pro B370 has a base clock of 300 MHz and a boost of 2400 MHz. The B300 runs at a base of 1665 MHz and boosts to 2032 MHz. The Intel chip is built on Intel's 3 nm process, while the B300 uses TSMC's 5 nm node. The B300 packs 104,000 million transistors on its die. The Intel part's transistor count and die size are not recorded in the database.

The Verdict

The data indicates that these two products should not be cross-shopped. The Arc Pro B370 is an integrated GPU for portable systems, evidenced by its IGP slot width, system-shared memory, and portable device dependent display outputs. It is active in production and released on 2026-01-26, succeeding HD Graphics-WM. Its predecessor lineage places it firmly in the integrated graphics family.

The NVIDIA B300 is a server accelerator module, released on 2025-09-10, succeeding Server Hopper and preceding Server Rubin. It has no display outputs, uses an SXM Module slot width, and targets PCIe 5.0 x16 host integration. Its 1400 W TDP and 1800 W suggested PSU make it a data center part through and through.

For a portable device builder, the Arc Pro B370 is the only viable option between the two, because the B300 physically cannot fit into such a system and offers no display interface. For a server operator running FP16 or FP32 compute workloads, the B300 is the clear choice, with 1,231.8 TFLOPS of FP16 throughput versus 12.29 TFLOPS for the Intel part. The B300's 144 GB HBM3e memory capacity and 4.10 TB/s bandwidth enable workloads that the Arc Pro B370's system-shared memory cannot approach.

Neither part wins on price, as neither has a launch MSRP recorded in the database. The production status for both is Active, so both remain available for their respective markets. The verdict from the recorded data is simple: the Arc Pro B370 serves integrated portable graphics, and the B300 serves high-performance server compute. There is no scenario in the data where one substitutes for the other.

Head-to-Head Benchmarks

The database records no direct head-to-head benchmark results between these two parts, and the win counts are zero for each. However, the architectural specifications provide a clear quantitative comparison across every measurable metric.

The most significant gap is in FP16 throughput. The B300 delivers 1,231.8 TFLOPS with a 16:1 ratio, while the Arc Pro B370 delivers 12.29 TFLOPS with a 2:1 ratio. That is a 100-fold difference in raw FP16 compute. In FP32, the B300's 76.99 TFLOPS is roughly 12.5 times the Arc Pro B370's 6.144 TFLOPS. The B300's tensor cores, 592 of them, provide the hardware path for that FP16 and FP32 throughput, while the Intel part has no tensor core count recorded.

Memory capacity and bandwidth also show a stark separation. The B300 has 144 GB of HBM3e on a 4096-bit bus, yielding 4.10 TB/s of bandwidth. The Arc Pro B370 uses system-shared memory with a system dependent bandwidth figure. In practical terms, the B300's memory subsystem is dedicated and enormous, whereas the Intel part relies on whatever memory the host device provides.

Texture rate favors the B300 at 1,202.9 GTexel/s versus 96.00 GTexel/s for the Intel part, a ratio of about 12.5 to 1, mirroring the FP32 gap. Pixel rate is nearly identical: 48.77 GPixel/s for the B300 and 48.00 GPixel/s for the Arc Pro B370. This close pixel rate suggests that for pure rasterization output to a display, the two parts have similar pixel-fill capabilities, though the B300 has no display outputs to use it.

Clock speeds tell an interesting story. The B300 has a higher base clock (1665 MHz versus 300 MHz) but a lower boost clock (2032 MHz versus 2400 MHz). The Intel part boosts 368 MHz higher than the B300, though it starts from a much lower base. The B300's memory clock is recorded as 2000 MHz with 8 Gbps effective, while the Arc Pro B370's memory clock is system shared.

Transistor count is a major differentiator. The B300 packs 104,000 million transistors, while the Intel part's transistor count is listed as unknown. The B300 uses a 5 nm TSMC process, and the Intel part uses a 3 nm Intel process. The smaller node for the Intel chip helps explain its 25 W TDP, versus the B300's 1400 W TDP. The B300 requires a suggested PSU of 1800 W, while the Arc Pro B370 needs no power connectors at all.

API support also differs. The Arc Pro B370 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The B300 has no recorded API support, consistent with a server part that does not target gaming or graphics rendering. The B300 has no display outputs, while the Arc Pro B370's display outputs are portable device dependent.

FAQ

Q: Which GPU has higher FP32 performance?

A: The NVIDIA B300 delivers 76.99 TFLOPS of FP32 compute, while the Intel Arc Pro B370 delivers 6.144 TFLOPS. The B300 is roughly 12.5 times faster in FP32.

Q: What memory configurations do these GPUs use?

A: The NVIDIA B300 uses 144 GB of HBM3e memory on a 4096-bit bus with 4.10 TB/s bandwidth. The Intel Arc Pro B370 uses system-shared memory with a system dependent bandwidth and bus width.

Q: Can the NVIDIA B300 output to a display?

A: No. The B300 has no display outputs. The Intel Arc Pro B370 has display outputs that are portable device dependent, meaning it can drive integrated displays on portable devices.

Q: What are the power requirements for each GPU?

A: The Intel Arc Pro B370 has a 25 W TDP and requires no power connectors. The NVIDIA B300 has a 1400 W TDP and a suggested PSU of 1800 W.

Q: Which GPU supports DirectX 12 Ultimate?

A: Only the Intel Arc Pro B370, which supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The NVIDIA B300 has no recorded API support.

Q: What process nodes are used for each GPU?

A: The Intel Arc Pro B370 is built on Intel's 3 nm process, while the NVIDIA B300 is built on TSMC's 5 nm process. The B300 contains 104,000 million transistors; the Intel part's transistor count is not recorded.

Architecture Differences

The Intel Arc Pro B370 and NVIDIA B300 are built on fundamentally different architectures with different design goals. The Intel part uses the Xe3-LPG architecture on the Panther Lake chip, part of the Arc Graphics-WM generation. The NVIDIA B300 uses the Blackwell Ultra architecture on the GB110 chip, part of the Server Blackwell (Bxx) generation.

The process nodes differ. Intel uses its own 3 nm process for the Arc Pro B370, while NVIDIA uses TSMC's 5 nm process for the B300. The B300's transistor count is recorded at 104,000 million, while the Arc Pro B370's transistor count and die size are unknown in the database.

Core configurations are wildly different. The Arc Pro B370 has 1,280 shading units, 40 TMUs, 20 ROPs, and 10 ray tracing cores. It has no tensor cores recorded. The B300 has 18,944 shading units, 592 TMUs, 24 ROPs, and 592 tensor cores. It has no RT core count recorded. The B300's shading unit count is nearly 15 times the Intel part's.

Memory architecture is another major split. The Arc Pro B370 uses system-shared memory, meaning it has no dedicated VRAM and relies on the host system's memory. Its memory type, bus width, and bandwidth are all listed as system shared or system dependent. The B300 uses 144 GB of HBM3e with a 4096-bit bus and 4.10 TB/s bandwidth, a dedicated and massive memory subsystem.

The B300's tensor cores are a defining feature, enabling its FP16 throughput of 1,231.8 TFLOPS (16:1 ratio). The Arc Pro B370 has no tensor cores but does have 10 ray tracing cores, supporting DirectX 12 Ultimate features. The B300 does not list any RT cores or API support in the database, consistent with a compute-focused server part.

Clock behavior differs as well. The Arc Pro B370 has a 300 MHz base clock and a 2400 MHz boost clock. The B300 has a 1665 MHz base clock and a 2032 MHz boost clock. Despite the Intel part's higher boost, the B300's much larger core count and higher base clock drive its performance advantage.

Power and physical form factor separate the two clearly. The Arc Pro B370 is an IGP with a 25 W TDP, no power connectors, and an IGP bus interface. The B300 is an SXM Module with a 1400 W TDP, a suggested PSU of 1800 W, and a PCIe 5.0 x16 bus interface. The B300 has no display outputs, while the Arc Pro B370's display outputs are portable device dependent.

Release timing and lineage also differ. The Arc Pro B370 was released on 2026-01-26 and succeeds HD Graphics-WM. The B300 was released on 2025-09-10, succeeds Server Hopper, and has Server Rubin as its successor. Both are listed as Active in production status.

The API support gap is notable. The Arc Pro B370 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The B300 has no DirectX, OpenGL, or Vulkan support recorded in the database. This reinforces that the B300 is not intended for graphics rendering or gaming workloads, but for compute tasks where FP16 and FP32 throughput matter.

The pixel rates are nearly identical: 48.00 GPixel/s for the Arc Pro B370 and 48.77 GPixel/s for the B300. This is a rare point of parity, though it is functionally irrelevant for the B300 since it has no display outputs. The texture rates are far apart, with the B300 at 1,202.9 GTexel/s versus 96.00 GTexel/s for the Intel part.

In summary, the architecture differences reflect two separate product categories. The Intel Arc Pro B370 is a low-power integrated GPU for portable devices, using system-shared memory and supporting modern graphics APIs. The NVIDIA B300 is a high-power server accelerator with dedicated HBM3e memory, tensor cores, and massive FP16 throughput, built for data center compute workloads.