Intel Arc Pro B60 vs NVIDIA H20 NVL16 Comparison
Intel Arc Pro B60
H20 NVL16
PERFORMANCE BENCHMARKS
Analysis: Intel Arc Pro B60 vs NVIDIA H20 NVL16
Head-to-Head Benchmarks
The direct comparison between the Intel Arc Pro B60 and the NVIDIA H20 NVL16 is constrained by a significant gap in recorded data. The Intel Arc Pro B60 has eight benchmark entries in the database, while the NVIDIA H20 NVL16 has no recorded benchmark scores at all. The database shows the Arc Pro B60's average benchmark score at 3182, placing it in the 20th percentile of all GPUs. The H20 NVL16, lacking any benchmark submissions, holds a 50th percentile ranking based on its specification profile, but that percentile is derived from hardware attributes rather than measured performance.
Looking at the Intel part's individual results, the strongest showing is in PassMark G3D with a score of 14580, followed by PassMark GPU Compute at 7029. The DirectX 9 test returns 179, DirectX 11 returns 122, and DirectX 12 returns 76. The DirectX 10 score is 61, while the 3DMark Steel Nomad DX12 test produces 2646. The PassMark G2D score is 763. These figures establish a baseline for the Arc Pro B60, but without corresponding H20 NVL16 results, a direct head-to-head numerical comparison is impossible.
The nearest rivals for the Arc Pro B60 provide context for its standing. The NVIDIA Quadro P1000 scores 3163, which is 0.6 percent below the Arc Pro B60's average. The NVIDIA GeForce GT 640 scores 3210, placing it 0.9 percent above. The GeForce 920M scores 3287, which is 3.2 percent higher. The RTX 5080 SUPER scores 3075, sitting 3.5 percent below the Arc Pro B60. These deltas show the Arc Pro B60 clustered tightly with older or lower-tier NVIDIA parts, within a range of roughly 3.5 percent either direction. The data indicates the Arc Pro B60 is not positioned as a high-performance outlier; it sits in a narrow band alongside those four competitors.
For the H20 NVL16, the absence of benchmark data means no delta percentages can be calculated against any rival. The database records zero wins for each part in head-to-head tests, because no such tests exist in the records. The analysis must therefore rely on architectural specifications and the limited performance data available for one side of the comparison.
Architecture Differences
The two GPUs come from different manufacturers and target different segments. The Intel Arc Pro B60 uses the BMG-G21 chip built on the Xe2-HPG architecture, part of the Battlemage Pro Series. The NVIDIA H20 NVL16 uses the GH100 chip built on the Hopper architecture, part of the Server Hopper series. Both are fabricated on a 5 nm process at TSMC, but the similarities end there.
The transistor counts differ dramatically. The Intel chip contains 19,600 million transistors on a 272 mm² die, yielding a transistor density of 72.1 million per square millimeter. The NVIDIA chip contains 80,000 million transistors on an 814 mm² die, yielding 98.3 million per square millimeter. The NVIDIA die is nearly three times larger and holds more than four times the transistors.
Memory configurations diverge sharply. The Arc Pro B60 has 24 GB of GDDR6 memory on a 192-bit bus, delivering 456.0 GB/s of bandwidth. The H20 NVL16 has 96 GB of HBM3 memory on a 6144-bit bus, delivering 4.03 TB/s, which is roughly 8.8 times higher bandwidth. The memory clock on the Intel part is 2375 MHz (19 Gbps effective) versus 1313 MHz (5.3 Gbps effective) on the NVIDIA part, but the vastly wider bus on the NVIDIA side dominates throughput.
Compute resources also differ substantially. The Arc Pro B60 has 2560 shading units, 160 texture mapping units, and 80 raster operation units. It includes 20 ray tracing cores and no tensor cores. The H20 NVL16 has 9984 shading units, 312 TMUs, and only 24 ROPs. It has 312 tensor cores and no dedicated ray tracing cores listed. The shading unit count on the NVIDIA part is nearly four times higher. Pixel rate favors the Intel part at 192.0 GPixel/s versus 47.52 GPixel/s for the NVIDIA part, while texture rate favors NVIDIA at 617.8 GTexel/s versus 384.0 GTexel/s.
Floating point performance shows a clear hierarchy. The Arc Pro B60 delivers 12.29 TFLOPS FP32 and 24.58 TFLOPS FP16 (2:1 ratio). The H20 NVL16 delivers 39.54 TFLOPS FP32 and 79.07 TFLOPS FP16 (2:1 ratio). The NVIDIA part is roughly 3.2 times higher in FP32 and FP16 throughput.
The power and form factor differences reflect their intended environments. The Arc Pro B60 has a 200 W TDP, uses a dual-slot design with a single 8-pin power connector, and requires a 550 W suggested PSU. It connects via PCIe 5.0 x8 and offers four mini-DisplayPort 2.1 outputs. The H20 NVL16 has a 400 W TDP, ships as an SXM module with no power connectors listed, and requires an 800 W suggested PSU. It uses PCIe 5.0 x16 and has no display outputs. The Intel part is a workstation card with display capability; the NVIDIA part is a server accelerator with no video output.
API support also separates the two. The Arc Pro B60 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The H20 NVL16 lists N/A for DirectX, OpenGL, and Vulkan, indicating it is not designed for conventional graphics API workloads.
Where Each One Wins
Based on the recorded data, the Intel Arc Pro B60 wins in any scenario involving standard graphics rendering benchmarks, because it has measurable scores and the NVIDIA H20 NVL16 has none. The Arc Pro B60's PassMark G3D score of 14580 and its 3DMark Steel Nomad DX12 score of 2646 provide concrete reference points. The H20 NVL16 cannot be compared in those tests due to missing data, but its specification profile suggests different strengths.
The H20 NVL16 wins on raw compute capacity. Its FP32 throughput of 39.54 TFLOPS is more than triple the Arc Pro B60's 12.29 TFLOPS. Its FP16 throughput of 79.07 TFLOPS similarly dwarfs the Intel part's 24.58 TFLOPS. For workloads that rely on dense floating-point math, such as AI training or scientific simulation, the NVIDIA part has a clear theoretical advantage. The presence of 312 tensor cores on the NVIDIA part further indicates acceleration for matrix operations, while the Intel part has no tensor cores at all.
Memory capacity and bandwidth favor the NVIDIA part decisively. The 96 GB HBM3 pool with 4.03 TB/s bandwidth supports large models and data sets that would never fit in the Arc Pro B60's 24 GB GDDR6 with 456.0 GB/s. For server workloads holding large working sets, the H20 NVL16 is the only viable option between the two.
The Arc Pro B60 wins in graphics-oriented tasks. Its 80 ROPs and 20 ray tracing cores give it a hardware path for rasterization and ray-traced rendering. Its 192.0 GPixel/s pixel rate and 384.0 GTexel/s texture rate support display-driven workflows. The H20 NVL16's 24 ROPs and 47.52 GPixel/s pixel rate are low for graphics, and its lack of display outputs confirms it is not intended for rendering to a screen.
The Arc Pro B60 also wins on physical footprint and power efficiency in a practical sense. At 200 W TDP versus 400 W, and with a 167 mm length dual-slot card versus an SXM module, the Intel part fits into conventional workstation chassis. The database does not provide efficiency ratios, but the power figures alone indicate a lower operating requirement.
FAQ
Q: Which GPU has higher FP32 performance?
A: The NVIDIA H20 NVL16 delivers 39.54 TFLOPS FP32, compared to the Intel Arc Pro B60's 12.29 TFLOPS.
Q: How much memory does each GPU have?
A: The Intel Arc Pro B60 has 24 GB of GDDR6 on a 192-bit bus, while the NVIDIA H20 NVL16 has 96 GB of HBM3 on a 6144-bit bus.
Q: Does the Intel Arc Pro B60 support ray tracing?
A: Yes, it includes 20 ray tracing cores. The NVIDIA H20 NVL16 does not list any ray tracing cores in the database.
Q: Which GPU has display outputs?
A: The Intel Arc Pro B60 has four mini-DisplayPort 2.1 outputs. The NVIDIA H20 NVL16 has no display outputs.
Q: What is the power draw difference?
A: The Intel Arc Pro B60 has a 200 W TDP, while the NVIDIA H20 NVL16 has a 400 W TDP.
Q: Are there any benchmark scores for the NVIDIA H20 NVL16?
A: No. The database contains no benchmark entries for the H20 NVL16. Its percentile ranking of 50 is based on specifications, not measured performance.
Specification Differences
The two GPUs differ across every major specification category in the database.
Process node and foundry are identical: both use 5 nm at TSMC. Transistor counts differ: the Intel chip has 19,600 million transistors on a 272 mm² die, while the NVIDIA chip has 80,000 million transistors on an 814 mm² die. Transistor density is 72.1M per mm² for Intel and 98.3M per mm² for NVIDIA.
Clock speeds differ. The Intel part runs at 2000 MHz base and 2400 MHz boost. The NVIDIA part runs at 1830 MHz base and 1980 MHz boost. Memory clocks are 2375 MHz (19 Gbps effective) for Intel and 1313 MHz (5.3 Gbps effective) for NVIDIA.
Memory size, type, bus width, and bandwidth all differ. Intel uses 24 GB GDDR6 with a 192-bit bus and 456.0 GB/s. NVIDIA uses 96 GB HBM3 with a 6144-bit bus and 4.03 TB/s.
Compute unit counts differ. Intel has 2560 shading units, 160 TMUs, 80 ROPs, 20 RT cores, and no tensor cores. NVIDIA has 9984 shading units, 312 TMUs, 24 ROPs, no RT cores listed, and 312 tensor cores.
Pixel rate and texture rate differ. Intel has 192.0 GPixel/s and 384.0 GTexel/s. NVIDIA has 47.52 GPixel/s and 617.8 GTexel/s.
FP32 and FP16 performance differ. Intel has 12.29 TFLOPS FP32 and 24.58 TFLOPS FP16. NVIDIA has 39.54 TFLOPS FP32 and 79.07 TFLOPS FP16.
TDP differs: 200 W for Intel, 400 W for NVIDIA. Slot width differs: dual-slot for Intel, SXM module for NVIDIA. Power connectors exist only on Intel (1x 8-pin). Suggested PSU is 550 W for Intel and 800 W for NVIDIA.
Bus interface differs: PCIe 5.0 x8 for Intel, PCIe 5.0 x16 for NVIDIA. Display outputs differ: four mini-DisplayPort 2.1 for Intel, none for NVIDIA.
API support differs: Intel supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. NVIDIA lists N/A for all three.
Physical dimensions exist only for Intel: 167 mm length, 69 mm height, 40 mm width. The NVIDIA dimensions are null in the database.
Release dates are close: Intel on 2025-09-04 and NVIDIA on 2025-09-01. The NVIDIA part has a predecessor (Server Ada) and successor (Server Blackwell) listed, while the Intel part has neither. The Intel part has a launch MSRP of 499 USD, which is stated here once as recorded. The NVIDIA part has no launch MSRP in the database.