GPU Comparison

The NVIDIA P106-090 and AMD Radeon Pro 455 are two very different mobile and mining-oriented graphics processors, separated by their target use cases and architectural generations. The data shows the NVIDIA part delivers a substantial raw performance advantage in compute workloads, while the AMD part offers a more power-efficient and portable form factor. In the head-to-head benchmarks available, the NVIDIA P106-090 wins both tests, but the AMD Radeon Pro 455 holds its own in terms of API compatibility and physical design.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA P106-090 has an average benchmark score of 13470, compared to the AMD Radeon Pro 455's 12831. This places the NVIDIA part in the 54th percentile of all GPUs, while the AMD part sits in the 52nd percentile.

Q: How large is the performance gap in the head-to-head OpenCL test?

A: In the Geekbench OpenCL test, the NVIDIA P106-090 scores 21304 against the AMD Radeon Pro 455's 10336. That is a 106.1% advantage for the NVIDIA part, meaning it more than doubles the AMD GPU's score.

Q: What is the difference in memory bandwidth between the two cards?

A: The NVIDIA P106-090 offers 192.2 GB/s of bandwidth, while the AMD Radeon Pro 455 provides 81.28 GB/s. The NVIDIA part achieves this with a 192-bit bus and 3 GB of GDDR5 memory, whereas the AMD part uses a 128-bit bus and 2 GB of GDDR5 memory.

Q: Which GPU has a higher transistor density?

A: The AMD Radeon Pro 455, built on a 14 nm process at GlobalFoundries, has a transistor density of 24.4M / mm². The NVIDIA P106-090, using TSMC's 16 nm process, has a density of 22.0M / mm².

Q: What are the power consumption differences?

A: The AMD Radeon Pro 455 has a TDP of 35 W and requires no power connectors, fitting an MXM Module form factor. The NVIDIA P106-090 has a TDP of 75 W and requires a single 6-pin power connector, with a suggested PSU of 250 W.

Q: Which GPU supports a higher version of Vulkan?

A: The NVIDIA P106-090 supports Vulkan 1.4, while the AMD Radeon Pro 455 supports Vulkan 1.3. Both support DirectX 12 and OpenGL 4.6, though the NVIDIA part supports DirectX 12_1 while the AMD part supports DirectX 12_0.

Where Each One Wins

The NVIDIA P106-090 is the clear winner in raw compute performance. Its Geekbench OpenCL score of 21304 is more than double the AMD Radeon Pro 455's 10336. In Vulkan, the NVIDIA part also leads decisively, scoring 18596 against 12240, a 51.9% advantage. The NVIDIA GPU's higher pixel rate of 73.49 GPixel/s and texture rate of 73.49 GTexel/s, compared to the AMD's 13.68 GPixel/s and 41.04 GTexel/s, suggest it is better suited for tasks that demand high fill rates and shader throughput.

The AMD Radeon Pro 455 wins in efficiency and physical integration. Its 35 W TDP is less than half of the NVIDIA's 75 W, and it draws power directly from the slot with no external connectors. Its MXM Module form factor is designed for portable devices, with display outputs that are "Portable Device Dependent," making it a natural fit for laptops where space and power are constrained. The AMD part also has a higher transistor density (24.4M / mm² vs 22.0M / mm²), indicating a more compact design despite its smaller die size of 123 mm² versus 200 mm².

Architecture Differences

The two GPUs come from different architectural lineages. The NVIDIA P106-090 is based on the Pascal architecture, using the GP106 chip, and belongs to the "Mining GPUs" generation. It is fabricated on a 16 nm process at TSMC, with 4,400 million transistors on a 200 mm² die. The AMD Radeon Pro 455 uses the GCN 4.0 architecture, built around the Baffin chip, and is part of the Radeon Pro Mac (400 Series) generation. It is manufactured on a 14 nm process at GlobalFoundries, with 3,000 million transistors on a 123 mm² die.

The FP16 capabilities differ notably. The NVIDIA P106-090 has an FP16 throughput of 36.74 GFLOPS, which is a 1:64 ratio relative to its FP32 performance of 2.352 TFLOPS. The AMD Radeon Pro 455, in contrast, has an FP16 throughput of 1,313.3 GFLOPS, matching its FP32 output at a 1:1 ratio. This means the AMD part can handle half-precision workloads at the same rate as full-precision, while the NVIDIA part is heavily optimized for FP32.

The bus interface also differs: the NVIDIA part uses PCIe 1.0 x1, while the AMD part uses PCIe 3.0 x8. The NVIDIA GPU's limited interface reflects its mining-oriented design, where data transfer to the host is less critical than raw compute. The AMD part's wider PCIe 3.0 x8 connection is more suitable for general-purpose tasks in a portable system.

Specification Differences

The most obvious difference is memory. The NVIDIA P106-090 has 3 GB of GDDR5 on a 192-bit bus, delivering 192.2 GB/s of bandwidth. The AMD Radeon Pro 455 has 2 GB of GDDR5 on a 128-bit bus, providing 81.28 GB/s. The NVIDIA part's memory clock is rated at 2002 MHz (8 Gbps effective), while the AMD part runs at 1270 MHz (5.1 Gbps effective).

Both GPUs have 768 shading units and 48 TMUs, but they diverge on ROPs: the NVIDIA part has 48 ROPs, while the AMD part has only 16. This contributes to the large difference in pixel rate (73.49 GPixel/s vs 13.68 GPixel/s). The NVIDIA part also has a much higher texture rate of 73.49 GTexel/s versus 41.04 GTexel/s.

The physical specifications are distinct. The NVIDIA P106-090 is a dual-slot card measuring 250 mm (9.8 inches) in length, requiring a 1x 6-pin power connector and a 250 W suggested PSU. The AMD Radeon Pro 455 is an MXM Module with no length specified and no power connectors. The NVIDIA part has no display outputs, while the AMD part's outputs are listed as "Portable Device Dependent."

The release dates differ by about nine months: the AMD Radeon Pro 455 was released on 2016-10-29, and the NVIDIA P106-090 followed on 2017-07-30. Both are marked as end-of-life. Neither has a launch MSRP listed.

Head-to-Head Benchmarks

The head-to-head data includes two tests, and the NVIDIA P106-090 wins both. In Geekbench OpenCL, the NVIDIA part scores 21304 against the AMD's 10336, a delta of 106.1%. This is the largest margin in any comparison, showing the NVIDIA GPU's dominance in compute-heavy workloads. The AMD part's score here is less than half of its rival, indicating a fundamental throughput advantage for the Pascal-based card.

In Geekbench Vulkan, the NVIDIA P106-090 again leads, scoring 18596 versus 12240, a 51.9% advantage. While this is a smaller gap than in OpenCL, it still represents a substantial performance lead. The AMD Radeon Pro 455's Vulkan score is closer to its OpenCL score, but it cannot match the NVIDIA part's absolute numbers.

The average benchmark scores reflect this trend. The NVIDIA P106-090's average of 13470 is 639 points higher than the AMD's 12831. In the nearest rival comparisons, the NVIDIA part is nearly tied with the AMD Radeon Pro 555 (deltaPct 0.5) and the AMD Radeon HD 7770M (deltaPct -0.5). The AMD Radeon Pro 455 is essentially tied with the NVIDIA GeForce GTX 590 (deltaPct 0) and the AMD FirePro W5100 (deltaPct -0.1). Neither GPU is dramatically ahead of its closest competitors, but the NVIDIA part sits slightly higher in the overall distribution.

The Verdict

The data clearly favors the NVIDIA P106-090 for anyone prioritizing raw compute performance. It wins both head-to-head benchmarks by significant margins, offers more memory (3 GB vs 2 GB), higher bandwidth (192.2 GB/s vs 81.28 GB/s), and a higher pixel rate (73.49 GPixel/s vs 13.68 GPixel/s). Its average benchmark score is higher, and it ranks in a higher percentile of all GPUs. The NVIDIA part also supports a newer Vulkan version (1.4 vs 1.3) and a higher DirectX feature level (12_1 vs 12_0).

The AMD Radeon Pro 455 is the better choice for systems where power and space are the primary constraints. Its 35 W TDP is less than half the NVIDIA's 75 W, and its MXM Module form factor with no power connectors makes it suitable for portable devices. Its higher transistor density (24.4M / mm²) and smaller die (123 mm²) suggest a more efficient design per square millimeter. The AMD part also offers balanced FP16 performance at a 1:1 ratio, which could be advantageous for specific compute workloads that utilize half-precision arithmetic.

In summary, the NVIDIA P106-090 is the performance leader by a wide margin, but it is a dual-slot, power-hungry card with no display outputs and a limited PCIe 1.0 x1 interface. The AMD Radeon Pro 455 is a low-power, mobile-oriented GPU that trades performance for efficiency. The benchmark results indicate that if you need compute power, the NVIDIA part is superior; if you need an integrated, low-power solution for a laptop, the AMD part is the practical option.