AMD Ryzen Z2 Extreme GPU vs NVIDIA H20 Comparison

Where Each One Wins

The AMD Ryzen Z2 Extreme GPU and the NVIDIA H20 occupy completely different segments of the hardware spectrum, and the recorded data makes this split explicit. The Ryzen Z2 Extreme GPU is a mobile-class integrated graphics solution built for compact, low-power systems. Its single recorded benchmark result in 3DMark Steel Nomad DX12 places it at 516 points, which puts it in the 1st percentile of all GPUs in the database. That percentile figure indicates this part sits near the very bottom of the performance distribution, surrounded by older discrete GPUs from over a decade ago.

The NVIDIA H20, by contrast, is a server-focused accelerator with no recorded gaming or graphics benchmarks in the database. Its average benchmark score is listed as zero, and its percentile rank is 50, which places it at the median of all GPUs tracked. The absence of benchmark entries does not mean the H20 is slow; it means the database has no gaming-oriented measurements for it, which is consistent with a part designed for compute workloads rather than rasterization. The wins are therefore split by category: the Ryzen Z2 Extreme GPU wins in any scenario that requires a DirectX 12 Ultimate graphics pipeline, while the H20 wins in raw compute throughput, memory capacity, and bandwidth.

In terms of pixel processing, the Ryzen Z2 Extreme GPU posts 129.6 GPixel/s, which is substantially higher than the H20's 47.52 GPixel/s. This is a meaningful advantage for the AMD part in traditional graphics rendering, where fill rate directly impacts resolution and detail levels. The H20's lower pixel rate, combined with its lack of display outputs, confirms it is not designed for direct framebuffer work. Texture rate tells the opposite story: the H20 delivers 617.8 GTexel/s versus the Ryzen Z2 Extreme GPU's 172.8 GTexel/s, an advantage that matters in shader-heavy compute and texture sampling workloads.

The data indicates these two products win in different domains because they were built for different jobs. The AMD part wins in client-side graphics, low-power operation, and API compatibility. The NVIDIA part wins in server-side compute, memory throughput, and FP16 processing. Neither product is a substitute for the other, and the benchmark results reflect that fundamental separation.

Architecture Differences

The architectural gap between these two chips is wide. The AMD Ryzen Z2 Extreme GPU uses the Strix Point chip built on RDNA 3.5 architecture, manufactured on a 4 nm process at TSMC. The NVIDIA H20 uses the GH100 chip based on Hopper architecture, built on a 5 nm process, also at TSMC. The process node difference gives AMD a density advantage: the Ryzen Z2 Extreme GPU packs 34,000 million transistors into 233 mm², producing a transistor density of 145.9 million per mm². The H20 holds 80,000 million transistors across 814 mm², resulting in 98.3 million per mm². The AMD die is smaller and denser, while the NVIDIA die is physically enormous and carries more than twice the transistor count.

Compute resources differ sharply. The Ryzen Z2 Extreme GPU has 1024 shading units, 64 texture mapping units, 48 render output units, and 16 ray tracing cores. The H20 has 9984 shading units, 312 texture mapping units, 24 render output units, and 312 tensor cores. The H20 does not list ray tracing cores in the database, while the AMD part does. The H20's tensor core count is a defining feature; the Ryzen Z2 Extreme GPU has no tensor cores listed at all. This distinction points to divergent design goals: the AMD chip targets graphics and ray tracing, while the NVIDIA chip targets tensor-heavy compute workloads.

Memory architecture is another major divider. The Ryzen Z2 Extreme GPU uses 16 GB of LPDDR5X on a 128-bit bus, delivering 128.0 GB/s of bandwidth. The H20 uses 96 GB of HBM3 on a 6144-bit bus, delivering 4.03 TB/s, which is roughly 31 times the bandwidth of the AMD part. The clock speeds differ as well: the AMD GPU runs at a base of 800 MHz and boosts to 2700 MHz, while the H20 runs at a base of 1830 MHz and boosts to 1980 MHz. The AMD part has a higher boost clock, but the H20 compensates with far more parallel hardware.

FP32 throughput favors the H20 at 39.54 TFLOPS versus 5.530 TFLOPS for the AMD part, a gap of roughly 7.1 times. FP16 performance shows an even larger divergence: the H20 reaches 79.07 TFLOPS with a 2:1 ratio, while the AMD part stays at 5.530 TFLOPS with a 1:1 ratio. The power envelope is also extreme: the Ryzen Z2 Extreme GPU is rated at 28 W with no power connectors, while the H20 is rated at 500 W and uses an SXM module with a suggested PSU of 900 W. The AMD part outputs video through a single USB Type-C port; the H20 has no display outputs.

The Verdict

The recorded data supports a clear verdict for each product based on use case. The AMD Ryzen Z2 Extreme GPU is a DirectX 12 Ultimate graphics solution with Vulkan 1.4 and OpenGL 4.6 support, a 28 W power draw, and a compact 233 mm² die. It is positioned for systems where graphics output, low power consumption, and standard API compatibility are required. Its 3DMark Steel Nomad score of 516 places it in the 1st percentile, and its nearest rivals in the database include the Intel HD Graphics P4000 at 534 points (3.4% higher), the AMD Radeon HD 6870 at 536 points (3.7% higher), the AMD Radeon HD 6750M at 484 points (6.6% lower), and the AMD Radeon HD 6770M at 569 points (9.3% higher). These comparisons place the Ryzen Z2 Extreme GPU in the company of legacy integrated and entry-level discrete parts.

The NVIDIA H20, with no recorded benchmarks, cannot be evaluated on the same graphics scale. Its 50th percentile rank and zero average score indicate the database treats it as a compute part outside the gaming benchmark scope. Its 96 GB HBM3 memory, 4.03 TB/s bandwidth, 312 tensor cores, and 39.54 TFLOPS FP32 make it a server accelerator. The 500 W TDP, SXM form factor, and absence of display outputs reinforce this classification.

Users who need a graphics output, ray tracing cores, DirectX 12 Ultimate, and low power consumption should select the AMD Ryzen Z2 Extreme GPU. Users who need massive memory capacity, extreme bandwidth, tensor cores, and high FP16 throughput for server workloads should select the NVIDIA H20. The data does not support using either part in the other's domain.

FAQ

Q: Which GPU has higher FP32 performance?

A: The NVIDIA H20 delivers 39.54 TFLOPS FP32, while the AMD Ryzen Z2 Extreme GPU delivers 5.530 TFLOPS, a difference of roughly 7.1 times in favor of the H20.

Q: Does the AMD Ryzen Z2 Extreme GPU support modern graphics APIs?

A: Yes, the database lists DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 support for the AMD part. The NVIDIA H20 lists N/A for DirectX, OpenGL, and Vulkan.

Q: How much memory does each GPU have?

A: The AMD Ryzen Z2 Extreme GPU has 16 GB of LPDDR5X on a 128-bit bus with 128.0 GB/s bandwidth. The NVIDIA H20 has 96 GB of HBM3 on a 6144-bit bus with 4.03 TB/s bandwidth.

Q: What is the power consumption of each part?

A: The AMD Ryzen Z2 Extreme GPU is rated at 28 W and requires no power connectors. The NVIDIA H20 is rated at 500 W and uses an SXM module with a suggested PSU of 900 W.

Q: Which GPU has ray tracing cores?

A: The AMD Ryzen Z2 Extreme GPU lists 16 ray tracing cores. The NVIDIA H20 does not list ray tracing cores in the database, but it does list 312 tensor cores.

Q: What are the nearest competitors to the AMD Ryzen Z2 Extreme GPU in the database?

A: The nearest rivals include the Intel HD Graphics P4000 at 534 points (3.4% higher), the AMD Radeon HD 6870 at 536 points (3.7% higher), the AMD Radeon HD 6750M at 484 points (6.6% lower), and the AMD Radeon HD 6770M at 569 points (9.3% higher).

Head-to-Head Benchmarks

The database records no direct head-to-head benchmark entries between the AMD Ryzen Z2 Extreme GPU and the NVIDIA H20. However, the individual measurements allow for a comparative analysis across several key metrics.

In pixel rate, the AMD Ryzen Z2 Extreme GPU posts 129.6 GPixel/s, which is 2.7 times the H20's 47.52 GPixel/s. This is the AMD part's largest categorical advantage. The H20's render output unit count of 24 is half of the AMD part's 48, which explains the lower pixel throughput despite the H20's much larger overall chip.

In texture rate, the NVIDIA H20 delivers 617.8 GTexel/s, which is 3.6 times the AMD part's 172.8 GTexel/s. The H20's 312 texture mapping units versus the AMD part's 64 drive this result. Texture-heavy compute workloads would favor the H20 significantly.

In FP32 throughput, the H20's 39.54 TFLOPS is 7.1 times the AMD part's 5.530 TFLOPS. The FP16 comparison is even more lopsided: the H20 reaches 79.07 TFLOPS at a 2:1 ratio, while the AMD part stays at 5.530 TFLOPS at a 1:1 ratio. The H20's FP16 advantage is roughly 14.3 times.

Memory bandwidth shows the largest absolute gap. The H20's 4.03 TB/s is approximately 31 times the AMD part's 128.0 GB/s. The H20's 6144-bit bus and HBM3 memory are the contributing factors, while the AMD part relies on a 128-bit LPDDR5X interface.

The 3DMark Steel Nomad DX12 result of 516 for the AMD part is the only synthetic graphics score in the database for either product. The H20 has no comparable entry. The AMD part's nearest rival, the AMD Radeon HD 6770M, scores 569, which is 9.3% higher. The AMD Radeon HD 6750M scores 484, which is 6.6% lower. These deltas situate the Ryzen Z2 Extreme GPU in a narrow performance band among older mobile and desktop GPUs.

The transistor and die size gap is also notable. The H20 contains 80,000 million transistors on an 814 mm² die, while the AMD part contains 34,000 million on 233 mm². The H20's die is 3.5 times larger, and its transistor count is 2.4 times higher. The AMD part's 4 nm process gives it a higher transistor density at 145.9M per mm² versus 98.3M per mm² for the H20.

Specification Differences

The following specifications differ between the AMD Ryzen Z2 Extreme GPU and the NVIDIA H20:

• Chip: Strix Point versus GH100
• Architecture: RDNA 3.5 versus Hopper
• Generation: Console GPU (AMD) versus Server Hopper (Hxx)
• Process Node: 4 nm versus 5 nm
• Transistors: 34,000 million versus 80,000 million
• Die Size: 233 mm² versus 814 mm²
• Transistor Density: 145.9M / mm² versus 98.3M / mm²
• Base Clock: 800 MHz versus 1830 MHz
• Boost Clock: 2700 MHz versus 1980 MHz
• Memory Clock: 1000 MHz 8 Gbps effective versus 1313 MHz 5.3 Gbps effective
• Memory Size: 16 GB versus 96 GB
• Memory Type: LPDDR5X versus HBM3
• Memory Bus Width: 128 bit versus 6144 bit
• Memory Bandwidth: 128.0 GB/s versus 4.03 TB/s
• Shading Units: 1024 versus 9984
• TMUs: 64 versus 312
• ROPs: 48 versus 24
• RT Cores: 16 versus not listed
• Tensor Cores: not listed versus 312
• Pixel Rate: 129.6 GPixel/s versus 47.52 GPixel/s
• Texture Rate: 172.8 GTexel/s versus 617.8 GTexel/s
• FP32: 5.530 TFLOPS versus 39.54 TFLOPS
• FP16: 5.530 TFLOPS (1:1) versus 79.07 TFLOPS (2:1)
• TDP: 28 W versus 500 W
• Slot Width: not listed versus SXM Module
• Power Connectors: None versus not listed
• Suggested PSU: not listed versus 900 W
• Bus Interface: not listed versus PCIe 5.0 x16
• Display Outputs: 1x USB Type-C versus No outputs
• DirectX: 12 Ultimate (12_2) versus N/A
• OpenGL: 4.6 versus N/A
• Vulkan: 1.4 versus N/A
• Release Date: 2025-07-08 versus 2024-01-31
• Predecessor: not listed versus Server Ada
• Successor: not listed versus Server Blackwell
• Percentile vs All GPUs: 1 versus 50
• Average Benchmark Score: 516 versus 0