AMD Radeon Pro W5700 vs NVIDIA CMP 70HX Comparison

Head-to-Head Benchmarks

The recorded data shows two benchmark comparisons between the NVIDIA CMP 70HX and the AMD Radeon Pro W5700, and in both tests the AMD card finishes ahead. The largest gap appears in Geekbench OpenCL, where the Radeon Pro W5700 scores 74,613 against the CMP 70HX’s 25,135. That is a 66.3% deficit for the NVIDIA part, meaning the AMD card delivers roughly three times the raw compute throughput in this particular workload. The OpenCL result likely reflects the Radeon Pro W5700’s dual-rate FP16 execution, which can boost certain compute tasks, though the test itself does not specify the precision used.

The second comparison, Geekbench Vulkan, narrows the gap but still favors AMD. The Radeon Pro W5700 records 70,706 points, while the CMP 70HX manages 35,817. The delta here is 49.3%, so the NVIDIA card is just under half the performance of its rival in this graphics API test. Both cards run Vulkan 1.4 per the database, so the difference stems from architecture rather than API support. The CMP 70HX has more shading units (3,840 vs. 2,304) and higher FP32 throughput (10.71 TFLOPS vs. 8.663 TFLOPS), yet it trails in both recorded benchmarks. This suggests that raw shader count does not translate into real-world wins for this mining-focused card, especially in workloads that are not tuned for its specific compute pattern.

Across the entire head-to-head set, the AMD Radeon Pro W5700 wins 2 out of 2 tests. The NVIDIA CMP 70HX records zero wins. This is a clean sweep in the database’s direct comparisons. However, the overall average benchmark score tells a slightly different story. The CMP 70HX averages 30,476 across its two tests, while the Radeon Pro W5700 averages 25,726 across ten tests. The NVIDIA card’s average is inflated because it only has two data points, both from Geekbench, and its Vulkan score is relatively strong compared to its OpenCL score. The AMD card’s average is dragged down by its Passmark results, which include several low DirectX scores (88 in DirectX 10, 104 in DirectX 11, 54 in DirectX 12, and 225 in DirectX 9). Those older API tests are not kind to the RDNA 1.0 architecture, but they are absent from the NVIDIA card’s benchmark history.

The percentile rankings place the CMP 70HX at the 75th percentile among all GPUs, while the Radeon Pro W5700 sits at the 71st percentile. Despite losing every head-to-head test, the NVIDIA card ranks slightly higher in the overall distribution. That is a consequence of its limited benchmark set: both of its scores are above 25,000, whereas the AMD card’s Passmark G2D score of 899 and DirectX scores drag its average down. If the comparison were limited to compute-heavy workloads like OpenCL and Vulkan, the AMD card would clearly rank higher. The database’s percentile metric rewards consistency across many tests, and the AMD card has more tests to pull from, including some that are not favorable to it.

FAQ

Q: Which card wins in Geekbench OpenCL performance?

A: The AMD Radeon Pro W5700 wins decisively, scoring 74,613 compared to the NVIDIA CMP 70HX’s 25,135. That is a 66.3% advantage for the AMD card.

Q: How do the two cards compare in Vulkan workloads?

A: The Radeon Pro W5700 again leads, with 70,706 points against 35,817 for the CMP 70HX. The margin is 49.3%, meaning the AMD card is nearly twice as fast in this test.

Q: Does the NVIDIA card have any benchmark wins over the AMD card?

A: No. The database records zero wins for the NVIDIA CMP 70HX in head-to-head comparisons. The AMD Radeon Pro W5700 wins both available tests.

Q: Why does the NVIDIA card have a higher overall percentile rank?

A: The CMP 70HX sits at the 75th percentile while the Radeon Pro W5700 is at the 71st. This is because the NVIDIA card has only two benchmark results, both above 25,000 points, while the AMD card has ten results, including several low Passmark DirectX scores that pull its average down.

Q: What is the average benchmark score for each card?

A: The NVIDIA CMP 70HX averages 30,476 across its two tests. The AMD Radeon Pro W5700 averages 25,726 across ten tests. The averages are not directly comparable due to different test counts and test types.

Q: Which card has more shading units and higher FP32 throughput?

A: The NVIDIA CMP 70HX has 3,840 shading units and 10.71 TFLOPS FP32 performance. The AMD Radeon Pro W5700 has 2,304 shading units and 8.663 TFLOPS FP32. Despite this hardware advantage, the NVIDIA card loses both recorded benchmarks.

Architecture Differences

The two cards come from different foundries and process nodes. The NVIDIA CMP 70HX uses the GA104 chip built on Samsung’s 8 nm process, while the AMD Radeon Pro W5700 uses the Navi 10 chip fabricated by TSMC on 7 nm. The transistor counts differ substantially: NVIDIA packs 17,400 million transistors into a 392 mm² die, giving a density of 44.4 million transistors per square millimeter. AMD fits 10,300 million transistors into a smaller 251 mm² die, with a density of 41.0 million per square millimeter. Despite having fewer transistors, the AMD chip runs on a more advanced process node, which helps explain its higher clock speeds.

Clock behavior is another major divergence. The AMD card has a base clock of 1400 MHz and a boost clock of 1880 MHz, both higher than the NVIDIA card’s 1365 MHz base and 1395 MHz boost. The NVIDIA card’s boost clock is only 30 MHz above its base, suggesting a very narrow boost range, likely due to power or thermal constraints on a mining-oriented board. The AMD card has a 480 MHz gap between base and boost, indicating aggressive dynamic clocking. Memory clocks also differ: the NVIDIA card runs GDDR6X at 1188 MHz, translating to 19 Gbps effective, while the AMD card uses GDDR6 at 1750 MHz, which is 14 Gbps effective. The NVIDIA card’s memory bus is 256 bits, same as AMD’s, but the faster GDDR6X yields 608.3 GB/s of bandwidth versus 448.0 GB/s for the AMD card.

The compute architecture is fundamentally different. NVIDIA’s Ampere generation includes 30 RT cores and 120 tensor cores, features that are absent from the AMD card (the database lists null for both RT and tensor cores on the Radeon Pro W5700). The NVIDIA card’s FP16 throughput is 10.71 TFLOPS, identical to its FP32 rate, meaning a 1:1 ratio. The AMD card’s FP16 is 17.33 TFLOPS, double its FP32 rate of 8.663 TFLOPS, indicating a 2:1 ratio. This explains why the AMD card excels in compute-heavy benchmarks: its FP16 performance is nearly double that of the NVIDIA card, even though its FP32 is lower.

Texture and pixel rates also favor AMD. The Radeon Pro W5700 achieves 270.7 GTexel/s and 120.3 GPixel/s, while the CMP 70HX manages 167.4 GTexel/s and 89.28 GPixel/s. The AMD card has more texture mapping units (144 vs. 120) and the same number of ROPs (64), but its higher clock speed pushes both rates well above NVIDIA’s figures. The NVIDIA card has more shading units (3,840 vs. 2,304), but this advantage does not show up in benchmark results, likely because the card’s low boost clock and mining-specific design limit its general compute efficiency.

Power and connectivity differ sharply. The AMD card is rated at 205 W TDP, while the NVIDIA card has no TDP listed in the database. The AMD card uses 1x 6-pin plus 1x 8-pin power connectors and suggests a 550 W power supply. The NVIDIA card uses a single 12-pin connector and suggests only 200 W. The AMD card supports PCIe 4.0 x16, while the NVIDIA card is limited to PCIe 1.0 x4, an unusual interface for a modern GPU and likely a reflection of its mining focus. Display outputs are also absent on the NVIDIA card, listed as "No outputs," while the AMD card provides 5x mini-DisplayPort 1.4a and 1x USB Type-C.

The Verdict

The data points in one direction for general compute and graphics workloads: the AMD Radeon Pro W5700 is the faster card. It wins both head-to-head benchmarks with margins of 66.3% in OpenCL and 49.3% in Vulkan. It also has a higher boost clock, more texture units, better pixel throughput, and twice the FP16 performance. For any user running OpenCL or Vulkan workloads, the AMD card is the clear choice from this dataset.

The NVIDIA CMP 70HX, however, has its own strengths that are not captured in the benchmark suite. It has more shading units, higher FP32 throughput, much higher memory bandwidth (608.3 GB/s vs. 448.0 GB/s), and hardware ray tracing and tensor cores. Its lower power draw and single 12-pin connector suggest it was designed for dense mining rigs, not for workstation tasks. The absence of display outputs confirms this: it is not meant to drive monitors. The PCIe 1.0 x4 interface further limits its utility in normal systems, as it would bottleneck data transfer in many workloads.

The percentile ranks show the NVIDIA card at 75th overall, slightly ahead of the AMD card’s 71st. This is a statistical artifact of the benchmark set, but it does indicate that the CMP 70HX is not a weak performer in absolute terms. Its Vulkan score of 35,817 is respectable, and its average of 30,476 is higher than the AMD card’s 25,726. However, the head-to-head results are unambiguous: the AMD card beats it by a wide margin in every direct comparison.

For users who need a display output, the NVIDIA card is immediately disqualified. For users who need ray tracing or tensor cores, the AMD card lacks those features entirely. For users who need maximum OpenCL or Vulkan performance, the AMD card wins by a large margin. The NVIDIA card’s only potential advantage lies in its higher memory bandwidth and FP32 compute, but those do not translate into wins in the recorded tests. The database’s verdict is straightforward: the AMD Radeon Pro W5700 is the superior card in measured performance.

Specification Differences

The following fields differ between the two cards according to the database:

• Chip: GA104 (NVIDIA) vs. Navi 10 (AMD)
• Architecture: Ampere vs. RDNA 1.0
• Generation: Mining GPUs vs. Radeon Pro Navi (Navi Series)
• Process Node: 8 nm (Samsung) vs. 7 nm (TSMC)
• Transistors: 17,400 million vs. 10,300 million
• Die Size: 392 mm² vs. 251 mm²
• Transistor Density: 44.4M / mm² vs. 41.0M / mm²
• Base Clock: 1365 MHz vs. 1400 MHz
• Boost Clock: 1395 MHz vs. 1880 MHz
• Memory Clock: 1188 MHz (19 Gbps effective) vs. 1750 MHz (14 Gbps effective)
• Memory Type: GDDR6X vs. GDDR6
• Memory Bandwidth: 608.3 GB/s vs. 448.0 GB/s
• Shading Units: 3840 vs. 2304
• Texture Mapping Units: 120 vs. 144
• RT Cores: 30 vs. none
• Tensor Cores: 120 vs. none
• Pixel Rate: 89.28 GPixel/s vs. 120.3 GPixel/s
• Texture Rate: 167.4 GTexel/s vs. 270.7 GTexel/s
• FP32: 10.71 TFLOPS vs. 8.663 TFLOPS
• FP16: 10.71 TFLOPS (1:1) vs. 17.33 TFLOPS (2:1)
• TDP: not listed vs. 205 W
• Power Connectors: 1x 12-pin vs. 1x 6-pin + 1x 8-pin
• Suggested PSU: 200 W vs. 550 W
• Bus Interface: PCIe 1.0 x4 vs. PCIe 4.0 x16
• Display Outputs: No outputs vs. 5x mini-DisplayPort 1.4a, 1x USB Type-C
• DirectX Support: 12 Ultimate (12_2) vs. 12 (12_1)
• Release Date: not listed vs. 2019-11-18
• Predecessor: not listed vs. Radeon Pro Vega
• Launch MSRP: not listed vs. 799 USD

Both cards share 8 GB memory, 256-bit bus width, 64 ROPs, dual-slot width, identical dimensions (267 mm length, 112 mm vs. 111 mm height), OpenGL 4.6, Vulkan 1.4, and end-of-life production status.

Where Each One Wins

AMD Radeon Pro W5700 wins in every measured benchmark category available in the database. Its OpenCL score of 74,613 is 66.3% higher than the NVIDIA card’s 25,135. Its Vulkan score of 70,706 is 49.3% higher than the NVIDIA card’s 35,817. The AMD card also wins on boost clock (1880 MHz vs. 1395 MHz), texture rate (270.7 GTexel/s vs. 167.4 GTexel/s), pixel rate (120.3 GPixel/s vs. 89.28 GPixel/s), and FP16 throughput (17.33 TFLOPS vs. 10.71 TFLOPS). It has display outputs, a modern PCIe 4.0 x16 interface, and a higher DirectX feature level (12_1) compared to the NVIDIA card’s older 12_2, though the NVIDIA card supports DirectX 12 Ultimate. For any workload that relies on OpenCL, Vulkan, or FP16 compute, the AMD card is the stronger option based on the data.

NVIDIA CMP 70HX wins on raw hardware specifications in several areas. It has more shading units (3,840 vs. 2,304), a higher FP32 throughput (10.71 TFLOPS vs. 8.663 TFLOPS), more memory bandwidth (608.3 GB/s vs. 448.0 GB/s), and GDDR6X memory instead of GDDR6. It includes 30 RT cores and 120 tensor cores, which the AMD card lacks entirely. It also draws less power according to the suggested PSU rating (200 W vs. 550 W), though the NVIDIA card has no TDP listed. Its transistor count is 17,400 million, far above the AMD card’s 10,300 million. However, none of these advantages produce a benchmark win in the recorded data. The NVIDIA card’s strengths are theoretical, not measured.

For users who need ray tracing, tensor core acceleration, or the highest memory bandwidth, the NVIDIA card is the only option among the two. For users who need any kind of display output, the AMD card is the only option, as the NVIDIA card has none. For users who need maximum compute performance in OpenCL or Vulkan, the AMD card wins decisively. The database’s recorded benchmarks show no scenario where the NVIDIA card outperforms the AMD card. Its higher percentile rank (75 vs. 71) is the only metric in its favor, and that is a function of having fewer, higher-scoring tests rather than superior performance in direct comparison.