Intel Arc Graphics 32EU vs NVIDIA H20 NVL16 Comparison

Where Each One Wins

The data presents a stark contrast between these two parts, but the wins are defined by completely different workloads. The Intel Arc Graphics 32EU is an integrated graphics processor embedded in the Arrow Lake-S chip. Its only recorded benchmark result is a 3DMark Steel Nomad DX12 score of 733, placing it at the 3rd percentile of all GPUs in the database. That score ties it exactly with the Intel Arc Graphics 24EU and Intel Arc Graphics 64EU, both also at 733, and puts it 1.4% ahead of the AMD Radeon HD 6470M. This is a part built for basic display output and light rendering, and the benchmark data confirms that positioning.

The NVIDIA H20 NVL16 sits at the opposite end of the compute spectrum. It is a server accelerator in the Hopper generation with no recorded gaming or graphics benchmarks at all. Its percentile ranking is 50, but its average benchmark score is listed as zero because the database contains no test results for it. The wins for the H20 are not measured in 3DMark runs. They are measured in raw compute throughput, memory bandwidth, and tensor operations. With 39.54 TFLOPS of FP32 performance and 79.07 TFLOPS of FP16 performance, the H20 operates in a performance class that the Intel integrated part cannot approach. The Intel part delivers 998.4 GFLOPS of FP32 and 1.997 TFLOPS of FP16, which means the H20 is roughly 39 times faster in FP32 and roughly 39 times faster in FP16 based on the recorded figures.

The use-case split is clean. The Intel Arc Graphics 32EU wins in the category of everyday desktop graphics, where its DirectX 12 Ultimate support, OpenGL 4.6, and Vulkan 1.4 compatibility allow it to function as a modern IGP for a mainstream system. The NVIDIA H20 NVL16 wins in server compute, where its 96 GB of HBM3 memory, 4.03 TB/s bandwidth, and 312 tensor cores make it a data center part with no display outputs at all. There is no overlap in their intended roles, and the benchmark records reflect that divergence.

Architecture Differences

The two processors come from different manufacturing nodes and design philosophies. The Intel Arc Graphics 32EU uses the Xe-LPG architecture on a 3 nm process from TSMC. The chip is Arrow Lake-S, and the graphics block is part of a larger die that measures 243 mm² with 17,800 million transistors. The transistor density is 73.3 million per square millimeter. The NVIDIA H20 NVL16 uses the Hopper architecture on a 5 nm process, also from TSMC. Its GH100 chip measures 814 mm² and contains 80,000 million transistors, giving it a density of 98.3 million per square millimeter. The H20 die is more than three times larger and carries more than four times the transistor count.

The execution resources differ by orders of magnitude. Intel's 32EU part has 256 shading units, 16 texture mapping units, and 8 raster output units. It has no dedicated ray tracing cores and no tensor cores listed. The NVIDIA part has 9,984 shading units, 312 TMUs, 24 ROPs, and 312 tensor cores. The H20 also lists no dedicated RT cores, but its tensor core array is a central feature for server workloads.

Memory architecture is another major divide. The Intel IGP uses system shared memory, with the bus width also marked as system shared and bandwidth described as system dependent. The NVIDIA H20 carries 96 GB of HBM3 on a 6144 bit bus, delivering 4.03 TB/s of bandwidth. The memory clock is recorded as 1313 MHz with 5.3 Gbps effective. The Intel part has no dedicated memory clock, relying entirely on the host system's RAM.

Clock behavior differs as well. The Intel part runs at a 300 MHz base clock and boosts to 1950 MHz. The NVIDIA part starts at 1830 MHz base and boosts to 1980 MHz. Despite the much higher base clock on the H20, its boost margin is small, reflecting a design that runs near its maximum continuously under server loads. The Intel IGP has a much larger boost range, from a very low idle base to a 1950 MHz peak.

Power and physical format separate them further. The Intel Arc Graphics 32EU is an IGP with a 65 W TDP, integrated into the processor package with a Ring Bus interface and motherboard-dependent display outputs. The NVIDIA H20 NVL16 is an SXM module with a 400 W TDP and a suggested power supply of 800 W. It uses a PCIe 5.0 x16 bus interface and has no display outputs. The Intel part is active in production, and so is the NVIDIA part, but their physical implementations could not be more different.

Head-to-Head Benchmarks

There are no direct head-to-head benchmark entries in the database, so the comparison must rely on the individual recorded measurements. The only benchmark available for the Intel Arc Graphics 32EU is 3DMark Steel Nomad DX12, where it scored 733. That score places it at the 3rd percentile of all GPUs. Its nearest rivals in the database are the Intel Arc Graphics 24EU and Intel Arc Graphics 64EU, both with identical average scores of 733, meaning a 0% delta. The AMD Radeon HD 6470M sits 1.4% behind at 723, while the NVIDIA GeForce GT 415M is 2.4% ahead at 751.

The NVIDIA H20 NVL16 has no benchmark scores recorded, so its average benchmark score is zero. Its percentile ranking of 50 is not derived from a graphics benchmark but from the database's overall classification. The absence of test data is itself informative: this is not a part that the database evaluates through conventional GPU tests. The massive compute figures in its specification sheet stand in place of benchmark results.

Looking at the raw throughput numbers, the H20's FP32 output of 39.54 TFLOPS compares to the Intel part's 998.4 GFLOPS. The difference is a factor of roughly 39.6. In FP16, the H20 delivers 79.07 TFLOPS versus the Intel part's 1.997 TFLOPS, a factor of roughly 39.6 as well. Texture rate tells a similar story: the H20 processes 617.8 GTexel/s while the Intel part manages 31.20 GTexel/s. Pixel rate is 47.52 GPixel/s for the H20 and 15.60 GPixel/s for the Intel part. The H20 is roughly 3 times faster in pixel throughput and roughly 19.8 times faster in texture throughput.

The memory comparison is not a contest. The H20's 4.03 TB/s of bandwidth against the Intel part's system-dependent shared memory means the server accelerator has a dedicated, enormous memory pipeline. The 96 GB capacity versus system shared memory also highlights the different memory models. The Intel part is limited by whatever the host system provides, while the H20 has its own fixed HBM3 allocation.

Clock speeds are closer than the compute figures might suggest. The Intel part boosts to 1950 MHz, and the H20 boosts to 1980 MHz, a difference of only 30 MHz. The base clocks are further apart, with the H20 at 1830 MHz and the Intel part at 300 MHz. This shows that the H20 maintains a high floor, while the Intel IGP can drop to very low clocks when idle.

FAQ

Q: Which processor has a higher boost clock?

A: The NVIDIA H20 NVL16 boosts to 1980 MHz, while the Intel Arc Graphics 32EU boosts to 1950 MHz. The difference is 30 MHz.

Q: How much memory does the NVIDIA H20 NVL16 have?

A: The H20 NVL16 has 96 GB of HBM3 memory on a 6144 bit bus with 4.03 TB/s of bandwidth. The Intel Arc Graphics 32EU uses system shared memory.

Q: What is the transistor count of each chip?

A: The Intel Arc Graphics 32EU's Arrow Lake-S die contains 17,800 million transistors. The NVIDIA H20 NVL16's GH100 die contains 80,000 million transistors.

Q: Does the NVIDIA H20 NVL16 support DirectX?

A: No. The H20 NVL16 lists DirectX as N/A, along with OpenGL and Vulkan as N/A. The Intel Arc Graphics 32EU supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4.

Q: What is the TDP of each part?

A: The Intel Arc Graphics 32EU has a 65 W TDP and is an integrated graphics processor. The NVIDIA H20 NVL16 has a 400 W TDP and is an SXM module with a suggested power supply of 800 W.

Q: Why does the NVIDIA H20 NVL16 have no benchmark scores?

A: The database contains no recorded benchmark results for the H20 NVL16, so its average benchmark score is listed as zero. Its percentile ranking is 50, while the Intel Arc Graphics 32EU has a recorded 3DMark Steel Nomad DX12 score of 733.

Specification Differences

The two parts differ in nearly every measurable specification. The Intel Arc Graphics 32EU uses the Xe-LPG architecture on a 3 nm process, while the NVIDIA H20 NVL16 uses Hopper on a 5 nm process. Both are made by TSMC. The Intel die is 243 mm² with 17,800 million transistors, while the H20 die is 814 mm² with 80,000 million transistors. Transistor density is 73.3 million per mm² for Intel and 98.3 million per mm² for NVIDIA.

Clock speeds differ in base frequency only. The Intel part runs at 300 MHz base and 1950 MHz boost. The NVIDIA part runs at 1830 MHz base and 1980 MHz boost. Memory configuration is entirely different: the Intel part uses system shared memory with system dependent bandwidth, while the NVIDIA part uses 96 GB of HBM3 with a 6144 bit bus and 4.03 TB/s bandwidth.

Compute resources show the scale of the gap. The Intel part has 256 shading units, 16 TMUs, and 8 ROPs. The NVIDIA part has 9,984 shading units, 312 TMUs, and 24 ROPs. The NVIDIA part adds 312 tensor cores, while the Intel part lists none. Pixel rate is 15.60 GPixel/s for Intel and 47.52 GPixel/s for NVIDIA. Texture rate is 31.20 GTexel/s for Intel and 617.8 GTexel/s for NVIDIA. FP32 performance is 998.4 GFLOPS versus 39.54 TFLOPS. FP16 performance is 1.997 TFLOPS versus 79.07 TFLOPS.

Power and form factor are also different. The Intel part is an IGP with a 65 W TDP, a Ring Bus interface, and motherboard dependent display outputs. The NVIDIA part is an SXM module with a 400 W TDP, a suggested PSU of 800 W, a PCIe 5.0 x16 interface, and no display outputs. The Intel part supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The NVIDIA part lists all three APIs as N/A.

Release timing differs as well. The Intel Arc Graphics 32EU was released on 2024-10-23, with the HD Graphics-M as its predecessor. The NVIDIA H20 NVL16 was released on 2025-09-01, with Server Ada as its predecessor and Server Blackwell as its successor. The Intel part has no successor listed. Both parts are active in production.

The Verdict

The data points to a simple conclusion: these are not competing products. The Intel Arc Graphics 32EU is an integrated graphics solution for desktop processors, evidenced by its 65 W TDP, Ring Bus interface, motherboard dependent outputs, and its single 3DMark Steel Nomad DX12 score of 733. That score ties it with the Intel Arc Graphics 24EU and 64EU, and it sits just 1.4% above the AMD Radeon HD 6470M. Its 3rd percentile ranking confirms it is a basic graphics solution.

The NVIDIA H20 NVL16 is a server accelerator with no display outputs, no graphics API support, and no recorded benchmark scores. Its 400 W TDP, SXM module form factor, 96 GB HBM3 memory, and 312 tensor cores point squarely at data center compute workloads. The FP32 figure of 39.54 TFLOPS and FP16 figure of 79.07 TFLOPS place it in a completely different performance tier.

The choice between them depends on the workload. A system requiring integrated graphics with modern API support should use the Intel Arc Graphics 32EU. A server requiring massive memory bandwidth and tensor compute should use the NVIDIA H20 NVL16. The benchmark data shows the Intel part in the 3rd percentile of all GPUs, while the NVIDIA part has no graphics benchmarks because it is not a graphics product. The recorded specifications confirm that each part serves a distinct role, and the numbers do not support any crossover between those roles.