AMD Radeon RX 470 vs NVIDIA P104-100 Comparison

The NVIDIA P104-100 and the AMD Radeon RX 470 are two graphics cards from different eras and design philosophies. The P104-100 is a mining-focused product with no display outputs, while the RX 470 is a mainstream consumer GPU. This comparison examines their recorded benchmark performance, architectural choices, and the implications for users who might encounter these cards in the secondary market or in specialized compute roles.

Head-to-Head Benchmarks

The benchmark data shows a clear and consistent advantage for the NVIDIA P104-100 across every recorded test. In the 3DMark Steel Nomad DX12 test, the P104-100 scores 1413, while the RX 470 manages 842. This represents a 67.8% difference, making it the largest gap between the two cards in any benchmark. This DX12 result suggests the P104-100 has a substantial edge in modern, low-level API workloads that are common in contemporary game engines and compute tasks.

The Vulkan benchmark tells a similar story, though with a narrower margin. The P104-100 posts a score of 45165, while the RX 470 reaches 39884. The 13.2% advantage for the NVIDIA card indicates that even in cross-platform APIs where AMD has historically been competitive, the P104-100’s higher clock speeds and memory bandwidth provide a meaningful lead. This result also shows that the gap between the two cards narrows when the API overhead is lower, but the NVIDIA card still comes out ahead.

The Geekbench OpenCL test reveals the largest absolute score difference. The P104-100 scores 52368, and the RX 470 scores 33568. That is a 56% advantage for the NVIDIA card. OpenCL is a common compute workload, and this result strongly suggests that the P104-100 is the superior choice for general-purpose GPU computing tasks such as rendering, simulation, or data processing. The RX 470 also has a Geekbench Metal score of 41690, but the P104-100 has no Metal test recorded in the database, so no direct comparison is possible for that API.

Overall, the P104-100 wins all three head-to-head tests, giving it a 3-0 record. Its average benchmark score across all recorded tests is 32982, compared to the RX 470’s 28996. This places the P104-100 in the 77th percentile of all GPUs, while the RX 470 sits in the 74th percentile. The percentile difference is modest, but the raw score gap in specific workloads is substantial. The data suggests that the P104-100 is not just marginally faster, but often dramatically faster, especially where compute throughput is the primary factor.

Where Each One Wins

The P104-100 wins in every recorded category, but the nature of its wins tells a story about its strengths. The biggest advantage comes in the OpenCL and 3DMark DX12 tests, where the delta exceeds 50%. These workloads benefit from high memory bandwidth and high FP32 throughput. The P104-100’s GDDR5X memory operates at 10 Gbps effective and delivers 320.3 GB/s of bandwidth, while the RX 470’s GDDR5 memory runs at 6.6 Gbps effective and provides 211.2 GB/s. That nearly 52% bandwidth advantage translates directly into the observed compute performance gains.

For gaming, the story is more nuanced. The RX 470 is a consumer card with display outputs, including 1x HDMI 2.0b and 3x DisplayPort 1.4a, so it is the only one of the two that can actually drive a monitor. The P104-100 has no display outputs, meaning it cannot be used for gaming on a screen without a separate graphics solution. The database records show that the P104-100 wins the 3DMark DX12 test by a wide margin, but that test does not account for the practical limitation of having no video outputs. If a user could somehow connect it, the P104-100 would be the faster gaming card based on the recorded score, but its mining heritage means it lacks the essential hardware for display output.

The RX 470 wins in the practical sense for any user who needs a standard desktop experience. Its PCIe 3.0 x16 interface is also more flexible than the P104-100’s PCIe 1.0 x4 interface, which can bottleneck data transfer in some systems. For compute workloads where the card does not need to render to a screen, the P104-100 is the clear winner. For a conventional build with a display, the RX 470 is the only viable option. The data also shows the RX 470 has a Metal benchmark score of 41690, which is its strongest recorded result, suggesting it has some utility in macOS environments, whereas the P104-100 has no such test in the database.

Architecture Differences

The NVIDIA P104-100 is built on the Pascal architecture using the GP104 chip, manufactured on a 16 nm process at TSMC. It contains 7,200 million transistors on a 314 mm² die, giving it a transistor density of 22.9M per mm². The AMD Radeon RX 470 uses the GCN 4.0 architecture, specifically the Ellesmere chip, built on a 14 nm process at GlobalFoundries. It has 5,700 million transistors on a 232 mm² die, leading to a higher density of 24.6M per mm². The RX 470’s denser packing partially compensates for its smaller chip, but the P104-100’s larger transistor count provides a raw computing advantage.

Memory is another point of divergence. The P104-100 uses GDDR5X memory with a 256-bit bus and 320.3 GB/s bandwidth. The RX 470 uses GDDR5 memory with the same 256-bit bus but only achieves 211.2 GB/s. Both have 4 GB of memory, but the P104-100’s faster memory type is a key factor in its benchmark dominance. The P104-100 has a base clock of 1607 MHz and a boost of 1733 MHz, while the RX 470 runs at 926 MHz base and 1206 MHz boost. The P104-100’s clocks are substantially higher, which boosts both pixel and texture fill rates.

The shading unit counts are close: the P104-100 has 1920 shading units and 120 texture units, while the RX 470 has 2048 shading units and 128 texture units. Despite having fewer shading units, the P104-100 achieves a higher FP32 compute rating of 6.655 TFLOPS compared to the RX 470’s 4.940 TFLOPS. This is because the P104-100’s higher clock speeds more than compensate for the lower unit count. The ROP count differs dramatically: the P104-100 has 64 ROPs, while the RX 470 has 32, leading to a pixel rate of 110.9 GPixel/s for the NVIDIA card versus 38.59 GPixel/s for the AMD card.

The FP16 compute is a curious difference. The P104-100 has an FP16 rate of 104.0 GFLOPS (1:64), meaning its FP16 throughput is heavily diluted compared to FP32. The RX 470 has an FP16 rate of 4.940 TFLOPS (1:1), equal to its FP32 rate. This means the RX 470 is far more efficient at half-precision math, which can be relevant for certain machine learning inference tasks, though neither card has dedicated tensor cores. The P104-100 also supports DirectX 12 (12_1) and Vulkan 1.4, while the RX 470 supports DirectX 12 (12_0) and Vulkan 1.3. The DX12 feature level difference is minor but present.

FAQ

Q: Which card has higher raw compute performance?

A: The NVIDIA P104-100 has a higher FP32 rating of 6.655 TFLOPS, compared to the AMD Radeon RX 470’s 4.940 TFLOPS. This matches the recorded benchmark scores where the P104-100 wins OpenCL by 56%.

Q: Why is the P104-100 so much faster in 3DMark DX12?

A: The 67.8% score difference in the 3DMark Steel Nomad DX12 test is likely due to the P104-100’s higher clock speeds, higher memory bandwidth (320.3 GB/s versus 211.2 GB/s), and double the ROP count (64 versus 32). The P104-100’s Pascal architecture also supports DirectX 12_1, while the RX 470 supports only 12_0.

Q: Can I use the NVIDIA P104-100 for gaming on a monitor?

A: No. The P104-100 has no display outputs, so it cannot connect to a monitor. The RX 470 has 1x HDMI 2.0b and 3x DisplayPort 1.4a outputs, making it the only one of the two that is usable with a standard display.

Q: Which card has a lower power requirement?

A: The AMD Radeon RX 470 has a TDP of 120 W and requires a single 6-pin power connector, with a suggested PSU of 300 W. The P104-100 has no listed TDP but requires a single 8-pin connector and a suggested PSU of 200 W.

Q: What is the release date difference?

A: The AMD Radeon RX 470 was released on 2016-08-03, while the NVIDIA P104-100 was released on 2017-12-11. The RX 470 is from the Arctic Islands (RX 400) generation, and the P104-100 is from the Mining GPUs generation.

Q: Which card has a higher average benchmark score?

A: The P104-100 has an average benchmark score of 32982, which is higher than the RX 470’s 28996. The P104-100 sits at the 77th percentile of all GPUs, while the RX 470 is at the 74th percentile.

Specification Differences

The two cards differ in many key specifications. The P104-100 uses a 16 nm TSMC process, while the RX 470 uses a 14 nm GlobalFoundries process. The P104-100 has 7,200 million transistors on a 314 mm² die, whereas the RX 470 has 5,700 million on a 232 mm² die. The transistor density is 22.9M per mm² for the NVIDIA card and 24.6M per mm² for AMD.

Clock speeds are notably different: the P104-100 has a base clock of 1607 MHz and a boost of 1733 MHz, while the RX 470 has a base of 926 MHz and a boost of 1206 MHz. The memory type is GDDR5X for the P104-100, running at 10 Gbps effective, and GDDR5 for the RX 470, at 6.6 Gbps effective. The memory bandwidth is 320.3 GB/s for the P104-100 and 211.2 GB/s for the RX 470, both on a 256-bit bus with 4 GB capacity.

The shading units differ: 1920 for the P104-100 versus 2048 for the RX 470. The TMU count is 120 versus 128, but the ROP count is 64 versus 32. The pixel rate is 110.9 GPixel/s versus 38.59 GPixel/s, and the texture rate is 208.0 GTexel/s versus 154.4 GTexel/s. The FP32 throughput is 6.655 TFLOPS versus 4.940 TFLOPS. The FP16 throughput is 104.0 GFLOPS for the P104-100 and 4.940 TFLOPS for the RX 470.

The power connector is an 8-pin for the P104-100 and a 6-pin for the RX 470. The suggested PSU power is 200 W for the NVIDIA card and 300 W for the AMD card. The bus interface is PCIe 1.0 x4 for the P104-100 and PCIe 3.0 x16 for the RX 470. The display outputs are none for the P104-100 and 1x HDMI 2.0b plus 3x DisplayPort 1.4a for the RX 470. The physical dimensions are 267 mm length for the P104-100, while the RX 470 is 240 mm in length, 95 mm in height, and 35 mm in width. Both are dual-slot cards. The release date is 2017-12-11 for the P104-100 and 2016-08-03 for the RX 470.

The Verdict

The data points to a clear performance hierarchy: the NVIDIA P104-100 is a substantially faster card in every recorded benchmark, with a 67.8% lead in 3DMark DX12, a 56% lead in OpenCL, and a 13.2% lead in Vulkan. Its higher clocks, memory bandwidth, and ROP count give it a decisive edge in raw compute and rasterization. However, this performance comes at a usability cost. The P104-100 has no display outputs, so it is only suitable for mining or dedicated compute servers where a display is not needed. Its PCIe 1.0 x4 interface is also a bottleneck for modern motherboards, limiting its effectiveness in general systems.

The AMD Radeon RX 470 is the only choice for a user who needs a functional graphics card with display connectivity. It supports DirectX 12_0, Vulkan 1.3, and has a Metal score in the database, though it lacks the raw power of the P104-100. Its lower TDP of 120 W and single 6-pin connector make it easier to power in a typical desktop, and its PCIe 3.0 x16 interface is standard. The RX 470 also has a higher FP16 throughput, which could be useful for specific half-precision compute tasks, though the P104-100’s FP32 advantage is more relevant for most workloads.

For a user building a system for rendering, scientific computing, or any task that does not require a display, the P104-100 is the clear choice based on the recorded scores. Its average benchmark score of 32982 versus 28996 for the RX 470 is significant, and it places higher in the overall percentile ranking. For a user who needs a general-purpose graphics card for gaming or desktop use, the RX 470 is the only option, and its lower performance must be accepted as a trade-off. The data does not suggest that the RX 470 is a better card in any measured performance metric, but its feature set makes it the only one of the two that is practical for end-user systems. The launch MSRP of 179 USD for the RX 470 indicates its original market position, but the P104-100 had no MSRP listed, reflecting its non-retail mining focus.