AMD Radeon Pro Vega 20 vs NVIDIA P106-100 Comparison

The AMD Radeon Pro Vega 20 and the NVIDIA P106-100 represent two very different approaches to GPU design, one aimed at professional mobile workstations and the other at cryptocurrency mining. The recorded benchmark data shows a clear split in performance characteristics, with the NVIDIA part taking the lead in the available cross-platform tests. However, the full picture is more complex than a simple win-loss tally, as each card’s architecture, memory subsystem, and feature set point to distinct intended workloads.

Head-to-Head Benchmarks

The database contains two direct head-to-head comparisons between these GPUs, and the NVIDIA P106-100 wins both. In the Geekbench OpenCL test, the P106-100 scores 35,951 against the Radeon Pro Vega 20’s 26,679. That is a delta of negative 25.8 percent when viewed from the AMD card’s perspective, meaning the NVIDIA part is roughly 34.8 percent faster in this compute-oriented workload. The margin is substantial and reflects the P106-100’s higher raw shader throughput.

The second direct comparison, Geekbench Vulkan, shows a similar outcome. The P106-100 scores 32,897 while the Radeon Pro Vega 20 trails at 26,410. The delta here is negative 19.7 percent from the AMD side, translating to an NVIDIA lead of about 24.6 percent. Vulkan is a low-level graphics API that often rewards raw geometry and pixel throughput, so this result hints at the P106-100’s advantage in those areas.

It is notably the Radeon Pro Vega 20 has a third recorded benchmark, Geekbench Metal, where it scores 30,427. There is no comparable Metal result for the P106-100, so this cannot be used as a direct head-to-head metric. Still, the Metal score is higher than the card’s OpenCL and Vulkan scores, suggesting that the AMD part performs relatively better in Apple’s API environment, which is consistent with its design as a Mac-oriented professional GPU.

Overall, the head-to-head tally is 0 wins for the Radeon Pro Vega 20 and 2 wins for the P106-100. But the average benchmark scores tell a slightly different story. The Radeon Pro Vega 20 has an average score of 27,839 across all its recorded tests, while the P106-100 sits at 23,249. This discrepancy arises because the P106-100’s average includes a much lower 3DMark Steel Nomad DX12 score of 899, which drags its overall figure down. The Radeon Pro Vega 20 has no comparable DX12 result in the database, so its average is based solely on the three Geekbench tests. This means the two metrics are not directly comparable, but they do highlight how different benchmark suites can shift the perceived ranking.

Where Each One Wins

The P106-100 is the clear winner in raw compute performance as measured by OpenCL and Vulkan. Its OpenCL score of 35,951 puts it in a different class from the Radeon Pro Vega 20’s 26,679. This suggests the NVIDIA card is better suited to tasks that leverage general-purpose GPU compute, such as rendering, physics simulation, or data processing, where the higher FP32 throughput of 4.375 TFLOPS can be fully utilized.

The P106-100 also wins in pixel and texture throughput. Its pixel rate is 82.03 GPixel/s, which is almost exactly double the Radeon Pro Vega 20’s 41.06 GPixel/s. Its texture rate of 136.7 GTexel/s is about 33 percent higher than the AMD card’s 102.6 GTexel/s. These are important metrics for rasterization-heavy workloads, and they help explain why the NVIDIA part wins the Vulkan benchmark. The P106-100 also has 48 ROPs compared to the AMD card’s 32, and its memory bandwidth of 192.2 GB/s is slightly higher than the Radeon’s 189.4 GB/s.

The Radeon Pro Vega 20, however, has strengths that do not show up in the head-to-head tests. Its FP16 performance is 6.569 TFLOPS, achieved through a 2:1 ratio with FP32. This is a massive advantage over the P106-100, which only manages 68.36 GFLOPS in FP16, a 1:64 ratio. For workloads that support FP16 arithmetic, such as certain machine learning inference tasks or image processing pipelines, the AMD card is fundamentally better equipped. This is a case where the benchmark data does not capture the full capability of the hardware.

The Radeon Pro Vega 20 also uses HBM2 memory with a 1024-bit bus, which is a very wide interface. While the resulting bandwidth of 189.4 GB/s is only slightly lower than the P106-100’s GDDR5 setup, HBM2 typically offers better power efficiency and a smaller physical footprint. The AMD card is an integrated graphics processor (IGP) with a 100 W TDP, whereas the P106-100 is a dual-slot discrete card with a 120 W TDP. In systems where space and power are constrained, the AMD part has a structural advantage.

Architecture Differences

The two GPUs come from different architectural generations and foundries. The Radeon Pro Vega 20 is built on GCN 5.0 architecture, also known as Vega, using a 14 nm process from GlobalFoundries. The P106-100 uses NVIDIA’s Pascal architecture, manufactured by TSMC on a 16 nm process. The node difference is small, but the architectural philosophies are not.

The Radeon Pro Vega 20 uses the Vega 12 chip, which is a compact implementation of Vega’s compute-centric design. It has 1280 shading units, 80 texture mapping units, and 32 ROPs. The P106-100, based on the GP106 chip, also has 1280 shading units and 80 TMUs, but it has 48 ROPs. The NVIDIA part’s higher ROP count directly contributes to its doubled pixel rate.

Transistor counts and die sizes are only recorded for the P106-100. It packs 4,400 million transistors into a 200 mm² die, giving a transistor density of 22.0 million per square millimeter. The Radeon Pro Vega 20’s die size and transistor count are not listed, so a direct comparison is not possible. However, the P106-100’s figures show a relatively dense chip for its era, which aligns with Pascal’s efficiency-focused design.

Memory technology is another major divergence. The Radeon Pro Vega 20 uses 4 GB of HBM2 on a 1024-bit bus, with memory clocked at 740 MHz and an effective data rate of 1480 Mbps. The P106-100 uses 6 GB of GDDR5 on a 192-bit bus, with memory clocked at 2002 MHz and an effective data rate of 8 Gbps. Despite these very different approaches, the resulting bandwidth figures are nearly identical: 189.4 GB/s for the AMD card and 192.2 GB/s for the NVIDIA card. This is a good illustration of how HBM2’s wide bus compensates for its lower clock speed.

The FP16 capability gap is stark. The Radeon Pro Vega 20 delivers 6.569 TFLOPS of FP16 performance, while the P106-100 manages only 68.36 GFLOPS. This is because GCN 5.0 supports packed FP16 arithmetic at a 2:1 ratio with FP32, while Pascal’s GP106 does not have dedicated FP16 units and executes it at a heavily reduced 1:64 ratio. For any application that can use half-precision math, the AMD card is orders of magnitude faster.

Specification Differences

The two cards differ in nearly every major specification category. The Radeon Pro Vega 20 has a base clock of 815 MHz and a boost clock of 1283 MHz, while the P106-100 runs at 1506 MHz base and 1709 MHz boost. The NVIDIA card’s clocks are substantially higher, which contributes to its FP32 advantage.

Memory configurations diverge on capacity, type, bus width, and clock speed. The AMD card has 4 GB of HBM2, while the NVIDIA card has 6 GB of GDDR5. The bus widths are 1024-bit and 192-bit, respectively. The effective memory speeds are 1480 Mbps for the AMD part and 8 Gbps for the NVIDIA part. Bandwidth is similar, as noted, but the capacity difference of 2 GB could matter for workloads that need more memory residency.

Compute throughput also favors the P106-100. The NVIDIA card delivers 4.375 TFLOPS of FP32, compared to the AMD card’s 3.284 TFLOPS. The P106-100 also has higher pixel and texture rates. Its TDP of 120 W is 20 W higher than the AMD card’s 100 W. The P106-100 requires a 6-pin power connector and a 300 W power supply, while the Radeon Pro Vega 20 is an IGP with no listed power connector or PSU requirement.

The form factors are fundamentally different. The Radeon Pro Vega 20 is an integrated graphics processor, meaning it is designed to be soldered onto a motherboard, likely in a laptop or compact workstation. The P106-100 is a dual-slot, 250 mm (9.8 inches) expansion card. The NVIDIA card has no display outputs, which is a clear indicator of its mining-focused design. The AMD card’s display outputs are listed as “Portable Device Dependent,” which is typical for a mobile IGP.

The bus interfaces also differ. The Radeon Pro Vega 20 uses PCIe 3.0 x16, while the P106-100 uses PCIe 1.0 x16. The older PCIe standard is a notable limitation for the NVIDIA card, though it may not matter much in mining workloads that primarily use local memory. The API support is similar, with both cards supporting DirectX 12 (12_1) and OpenGL 4.6. The P106-100 supports Vulkan 1.4, while the Radeon Pro Vega 20 supports Vulkan 1.3.

The P106-100 has a transistor count of 4,400 million on a 200 mm² die, with a density of 22.0M per mm². The Radeon Pro Vega 20 does not have these figures recorded. Both cards are end-of-life products. The Radeon Pro Vega 20 was released on November 13, 2018, while the P106-100 came out on June 18, 2017. Neither has a recorded launch MSRP.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Pro Vega 20 has an average score of 27,839, while the NVIDIA P106-100 averages 23,249. However, the P106-100’s average includes a low 3DMark Steel Nomad DX12 score of 899, which is a test the AMD card does not have a recorded result for.

Q: Why does the NVIDIA P106-100 win the Vulkan benchmark?

A: The P106-100 scores 32,897 in Geekbench Vulkan versus 26,410 for the Radeon Pro Vega 20. The NVIDIA card’s higher pixel rate of 82.03 GPixel/s, compared to 41.06 GPixel/s for the AMD card, likely contributes to its advantage in this graphics API.

Q: Does the AMD Radeon Pro Vega 20 have any compute advantage?

A: Yes, the Radeon Pro Vega 20 delivers 6.569 TFLOPS of FP16 performance through a 2:1 ratio, while the P106-100 only manages 68.36 GFLOPS at a 1:64 ratio. For FP16 workloads, the AMD card is vastly superior.

Q: What is the memory configuration difference?

A: The Radeon Pro Vega 20 uses 4 GB of HBM2 on a 1024-bit bus with 189.4 GB/s bandwidth. The P106-100 uses 6 GB of GDDR5 on a 192-bit bus with 192.2 GB/s bandwidth. The bandwidth is similar, but the P106-100 has 2 GB more capacity.

Q: Are both cards still in production?

A: No, both are listed as end-of-life products. The Radeon Pro Vega 20 was released on November 13, 2018, and the P106-100 was released on June 17, 2017.

Q: Can the NVIDIA P106-100 be used for display output?

A: No, the P106-100 has no display outputs. It was designed for mining GPUs. The Radeon Pro Vega 20, by contrast, has display outputs listed as “Portable Device Dependent,” which is typical for a mobile integrated GPU.

The Verdict

The data points to a clear split based on workload. For general compute and graphics tasks measured by OpenCL and Vulkan, the NVIDIA P106-100 is the stronger card. Its 34.8 percent lead in OpenCL and 24.6 percent lead in Vulkan, combined with higher FP32 throughput, pixel rate, and texture rate, make it the better choice for raw performance in those areas. Its 6 GB of memory also provides more headroom for data-heavy tasks.

However, the Radeon Pro Vega 20 is the better choice for FP16 compute workloads, where its 6.569 TFLOPS dwarfs the P106-100’s 68.36 GFLOPS. Its IGP form factor, lower 100 W TDP, and HBM2 memory make it suitable for compact systems where the P106-100’s dual-slot, 250 mm card with a 6-pin connector and 300 W PSU requirement would not fit. The AMD card also has a higher percentile ranking at 73 compared to the P106-100’s 68, indicating better standing across all GPUs in the database.

The P106-100’s lack of display outputs is a decisive limitation for any general-purpose use. It is a mining-focused card, and the data reflects that. The Radeon Pro Vega 20, despite losing both head-to-head tests, is a more versatile product with support for display output and a much stronger FP16 path. Users with FP16-dependent workloads should choose the AMD card. Users who prioritize raw FP32 and rasterization performance in a system that can accommodate a discrete card should choose the NVIDIA part. The choice is not about which card is better overall, but which one fits the intended task.