AMD Radeon Pro 580X vs NVIDIA P104-100 Comparison

The Verdict

The data separates these two cards into very different roles. The AMD Radeon Pro 580X is the broader, more flexible option, landing in the 82nd percentile of all GPUs in the database, while the NVIDIA P104-100 sits slightly lower at the 77th percentile. However, the raw benchmark comparisons show the NVIDIA card winning both head-to-head tests, including a decisive 30.4% lead in Geekbench OpenCL. The Radeon Pro 580X is the pick for anyone needing a working display output and modern API support, as it carries two HDMI 2.0b ports and a DirectX 12 (12_0) feature level. The P104-100 has no display outputs at all, making it strictly a compute-oriented board. For a builder targeting raw compute throughput in OpenCL or Vulkan workloads without needing a monitor attached, the P104-100 delivers more performance per the recorded scores. For a general-purpose workstation card that can drive a screen, the Radeon Pro 580X is the practical choice, despite losing the head-to-head compute tests.

Where Each One Wins

The NVIDIA P104-100 wins every recorded head-to-head benchmark. In Geekbench OpenCL, it scores 52,368 against the AMD's 36,426, a margin of 30.4%. In Geekbench Vulkan, it posts 45,165 versus 40,115, an 11.2% advantage. The AMD card does not win a single direct comparison, but it offers capabilities the NVIDIA card lacks entirely: two HDMI 2.0b outputs for display connectivity. The Radeon Pro 580X also has a higher transistor density at 24.6 million transistors per square millimeter versus 22.9 million for NVIDIA, and it supports a 1:1 FP16 ratio, meaning its half-precision throughput matches its FP32 rate at 5.530 TFLOPS. The P104-100, by contrast, has a heavily reduced FP16 rate of 104.0 GFLOPS, a 1:64 ratio, so it is not suited for workloads that rely on half-precision math. The AMD card also carries 8 GB of GDDR5 memory versus 4 GB of GDDR5X on the NVIDIA, which matters for larger datasets that fit in VRAM. The NVIDIA card counters with a much higher memory bandwidth of 320.3 GB/s versus 218.9 GB/s, and a faster pixel rate of 110.9 GPixel/s against 38.40 GPixel/s.

Architecture Differences

The two cards come from different foundries and process nodes. The AMD Radeon Pro 580X uses the Ellesmere chip on a 14 nm process from GlobalFoundries, built on GCN 4.0 architecture. It packs 5,700 million transistors into a 232 mm² die. The NVIDIA P104-100 uses the GP104 chip on a 16 nm process from TSMC, on the older Pascal architecture, with 7,200 million transistors on a larger 314 mm² die. Despite having fewer transistors, the AMD die achieves a higher density at 24.6 million transistors per square millimeter. The AMD card has more shading units at 2,304 versus 1,920, more texture mapping units at 1,920 versus 120, but far fewer ROPs at 32 versus 64. Clock speeds favor NVIDIA: a base of 1607 MHz and boost of 1733 MHz against AMD's 1100 MHz base and 1200 MHz boost. Memory type differs as well, with AMD using GDDR5 at 6.8 Gbps effective and NVIDIA using GDDR5X at 10 Gbps effective. The AMD card has a TDP of 185 W, while the NVIDIA card has no TDP listed, but its suggested PSU is 200 W. The AMD card uses an Apple MPX bus interface and is an IGP slot design, while the NVIDIA card uses PCIe 1.0 x4 and is a dual-slot card with a single 8-pin power connector. API support differs: AMD offers DirectX 12 (12_0), Vulkan 1.3, and OpenGL 4.6; NVIDIA offers DirectX 12 (12_1), Vulkan 1.4, and OpenGL 4.6. The NVIDIA card has a higher feature level for DirectX 12 and a newer Vulkan version. The AMD card was released on 2019-03-17, about 15 months after the NVIDIA card, which launched on 2017-12-11.

FAQ

Q: Which card is faster in Geekbench OpenCL?

A: The NVIDIA P104-100 scores 52,368 versus the AMD Radeon Pro 580X's 36,426, a 30.4% advantage.

Q: Can I connect a monitor to the NVIDIA P104-100?

A: No, the P104-100 has no display outputs. The AMD Radeon Pro 580X has 2x HDMI 2.0b outputs.

Q: Which card has more memory?

A: The AMD Radeon Pro 580X has 8 GB of GDDR5, while the NVIDIA P104-100 has 4 GB of GDDR5X.

Q: How does memory bandwidth compare?

A: The NVIDIA P104-100 has 320.3 GB/s of bandwidth, which is significantly higher than the AMD Radeon Pro 580X's 218.9 GB/s.

Q: Which card has better FP16 performance?

A: The AMD Radeon Pro 580X has 5.530 TFLOPS FP16 at a 1:1 ratio, while the NVIDIA P104-100 has only 104.0 GFLOPS FP16 at a 1:64 ratio.

Q: What are the API differences?

A: The AMD card supports DirectX 12 (12_0) and Vulkan 1.3, while the NVIDIA card supports DirectX 12 (12_0) and Vulkan 1.4. Both support OpenGL 4.6.

Head-to-Head Benchmarks

The only two head-to-head tests in the database both go to the NVIDIA P104-100. The largest gap is in Geekbench OpenCL, where the NVIDIA card scores 52,368 against the AMD's 36,426, a delta of 30.4%. This is a substantial lead and indicates the NVIDIA card has a significant advantage in OpenCL compute workloads. In Geekbench Vulkan, the NVIDIA card wins again with 45,165 versus 40,115, a smaller but still clear 11.2% margin. These results place the P104-100 ahead in both compute APIs that were tested. The AMD card's closest rival in the database is the AMD Radeon Pro 575X, which scores 39,116, about 1% higher than the Radeon Pro 580X's average of 38,706. The NVIDIA card's nearest rival is the AMD Radeon Pro 570, scoring 33,207, which is 0.7% higher than the P104-100's average of 32,982. Notably, the NVIDIA card's average benchmark score is lower than its head-to-head scores would suggest, because it also includes a 3DMark Steel Nomad DX12 result of 1,413, which drags down the average. The AMD card has no 3DMark result in its benchmark list, only Geekbench Metal, OpenCL, and Vulkan scores. The Geekbench Metal score for AMD is 39,577, which is its highest recorded result. The P104-100 has no Metal score, as Metal is not supported on NVIDIA hardware in this context.

Specification Differences

The two cards differ across nearly every major specification field. The AMD Radeon Pro 580X uses a 14 nm process node from GlobalFoundries with the Ellesmere chip, while the NVIDIA P104-100 uses a 16 nm node from TSMC with the GP104 chip. Transistor counts differ at 5,700 million for AMD and 7,200 million for NVIDIA, with die sizes of 232 mm² and 314 mm² respectively. Clock speeds are higher on NVIDIA at 1100 MHz base and 1200 MHz boost versus AMD's 1100 MHz base and 1733 MHz boost for NVIDIA. Memory configurations diverge in size, type, and effective speed: AMD has 8 GB of GDDR5 at 6.8 Gbps, while NVIDIA has 4 GB of GDDR5X at 10 Gbps. Memory bandwidth favors NVIDIA at 320.3 GB/s versus AMD's 218.9 GB/s. Shading units are 2,304 on AMD and 1,920 on NVIDIA, with TMUs at 144 and 120, and ROPs at 32 and 64. Pixel rate is notably higher on NVIDIA at 110.9 GPixel/s versus 38.40 GPixel/s, and texture rate is higher on NVIDIA at 208.0 GTexel/s versus 172.8 GTexel/s. FP32 performance favors NVIDIA at 6.655 TFLOPS versus 5.530 TFLOPS, while FP16 heavily favors AMD at 5.530 TFLOPS versus 104.0 GFLOPS. Power specifications differ, with AMD listing a TDP of 185 W and NVIDIA listing no TDP but a suggested PSU of 200 W. Slot width differs by design: AMD is an IGP, NVIDIA is dual-slot. The NVIDIA card has a single 8-pin power connector, while AMD has none listed. Bus interfaces are different: AMD uses Apple MPX, NVIDIA uses PCIe 1.0 x4. Display outputs are exclusive to AMD with 2x HDMI 2.0b, while NVIDIA has no outputs. API versions differ in DirectX and Vulkan, with NVIDIA supporting DirectX 12 (12_1) and Vulkan 1.4, while AMD supports DirectX 12 (12_0) and Vulkan 1.3; both support OpenGL 4.6. Physical dimensions are only listed for NVIDIA at 267 mm or 10.5 inches in length.