NVIDIA P104-100

The NVIDIA P104-100 is a Pascal-architecture mining GPU built on TSMC's 16 nm process, featuring the GP104 chip with 7,200 million transistors on a 314 mm² die. It sits at the 76th percentile of all GPUs in the database, with an average benchmark score of 32,747, placing it in a competitive mid-to-upper tier for its generation. The card is end-of-life, launched on December 11, 2017, and its data reflects a product designed specifically for compute workloads rather than traditional graphics output.

Benchmark Performance

The P104-100’s average benchmark score of 32,747 positions it nearly exactly at parity with its closest rivals, though the distribution of specific test results reveals a more nuanced picture. In 3DMark Steel Nomad DX12, the card scores 1,413, a modest result that reflects its mining-oriented design and lack of display outputs. This score is dwarfed by its OpenCL and Vulkan performance, where the card achieves 51,663 and 45,165 respectively, indicating that raw compute throughput is its strength rather than rasterized gaming performance. The FP32 throughput of 6.655 TFLOPS and texture rate of 208.0 GTexel/s corroborate this: the card is built for parallel processing, not frame rendering.

Compared to the NVIDIA T600 Mobile, which averages 32,849, the P104-100 trails by just 0.3%. This is effectively a statistical tie, but the delta is meaningful in context: the T600 Mobile is a professional laptop GPU with lower power constraints, while the P104-100 is a desktop mining card with a 200 W suggested PSU. The benchmark data shows that despite different intended markets, their compute performance aligns closely, with the P104-100’s advantage in raw shader count (1,920 shading units) offset by the T600 Mobile’s more efficient architecture for certain workloads.

Against the AMD Radeon RX 7800 XT, the P104-100 leads by 0.4% (32,747 vs 32,619). This is a surprising result given the RX 7800 XT is a much newer gaming card, but the average benchmark score masks the divergent workload characteristics. The P104-100’s Vulkan score of 45,165 demonstrates strong API-level compute, while the RX 7800 XT’s gaming-oriented design likely excels in DirectX titles. The 0.4% delta is within noise, yet it underscores that the P104-100’s compute-centric architecture remains competitive in synthetic benchmarks.

The AMD FirePro S9300 X2 trails by 0.6%, and the AMD FirePro S10000 by 1.1%. These deltas are small, but they show the P104-100 holding a slight edge over professional AMD offerings from the same era. The pixel rate of 110.9 GPixel/s and 64 ROPs provide a theoretical ceiling that these rivals may match or exceed in specific tests, but the average data indicates the P104-100 is marginally ahead. Across all four rivals, the performance spread is just 1.4%, meaning the P104-100 occupies a tightly contested performance band where architectural differences, not raw scores, determine suitability.

Who Should Consider It

The benchmark data indicates the P104-100 is not suited for conventional gaming or display-based workloads, as it has no display outputs. Instead, its performance profile targets compute-heavy tasks like cryptocurrency mining, data processing, or scientific calculations. The 3DMark Steel Nomad score of 1,413 is low enough to confirm that gaming at any resolution would be impractical; this is not a card for 1080p or 4K frame rates. The OpenCL score of 51,663, however, suggests strong parallel compute capability that could handle tasks like hash calculations or machine learning inference, though the 4 GB memory capacity limits dataset sizes.

For users running compute workloads at high resolutions or with large datasets, the memory subsystem becomes the limiting factor. The 320.3 GB/s bandwidth and 256-bit bus provide ample throughput for mid-sized workloads, but the 4 GB GDDR5X capacity means textures or matrices exceeding that footprint will spill into system memory, halving effective performance. The Vulkan score of 45,165 indicates that API-optimized compute kernels can leverage the card effectively, but only if they fit within the memory budget. Resolution-specific recommendations are moot for a card with no outputs; instead, consider it for single-pass compute tasks where the 6.655 TFLOPS FP32 rate and 208.0 GTexel/s texture throughput can be fully utilized without memory pressure.

Power and Cooling

The P104-100 carries no official TDP figure in the data, but the suggested PSU rating of 200 W provides a clear power envelope. This is a modest requirement for a dual-slot card with a 267 mm length (10.5 inches), indicating that the GP104 chip is power-efficient at the given clocks. The base clock of 1607 MHz and boost clock of 1733 MHz are the highest in its performance band among the listed rivals, yet the 200 W PSU recommendation suggests the card draws less power than its performance might imply. A single 8-pin power connector is required, which is standard for this class of card and compatible with most modern PSUs without adapters.

Cooling is handled by a dual-slot design, which provides adequate surface area for heat dissipation given the 200 W envelope. The 16 nm process node contributes to thermal efficiency, and the card’s mining-oriented design likely prioritizes sustained load operation over acoustic comfort. The PCIe 1.0 x4 bus interface is an unusual choice, as it severely limits host-to-device data transfer rates compared to PCIe 3.0 or 4.0 x16 slots. This means that while compute performance is strong, any workload requiring frequent data uploads or downloads to the host system will be bottlenecked by the interface. For mining or batch compute where data is loaded once and processed in-place, this is acceptable; for interactive workloads, it is a significant drawback.

How It Compares

NVIDIA T600 Mobile: The P104-100 trails by 0.3% in average benchmark score, making them near-identical performers. The T600 Mobile is a laptop GPU with lower power draw and a different memory configuration, but the compute results are statistically indistinguishable. The P104-100’s advantage lies in its 1,920 shading units and higher clocks (1607 MHz base vs the T600 Mobile’s unspecified), while the T600 Mobile likely wins on efficiency and portability.

AMD Radeon RX 7800 XT: The P104-100 leads by 0.4%, a razor-thin margin that flatters the mining card. The RX 7800 XT is a modern gaming GPU with far superior rasterization capabilities, but the synthetic benchmark average does not capture gaming-specific features like ray tracing or DLSS. In pure compute tasks, the Pascal architecture’s mature driver stack and FP32 focus keep it competitive, though the RX 7800 XT’s larger memory and newer process node would likely dominate in real-world gaming.

AMD FirePro S9300 X2: The P104-100 is 0.6% ahead. This dual-GPU professional card was designed for compute, and its average score of 32,540 shows it slightly underperforms the single-GPU NVIDIA offering. The P104-100’s higher boost clock of 1733 MHz and 320.3 GB/s bandwidth likely compensate for the S9300 X2’s dual-chip complexity and driver overhead.

AMD FirePro S10000: The P104-100 leads by 1.1%, the largest delta among its rivals. This older dual-GPU card averages 32,388, and its age shows in the benchmark results. The P104-100’s unified 4 GB memory pool and 256-bit bus provide more coherent access patterns than the S10000’s split memory architecture, leading to the modest but consistent advantage.

Memory Subsystem

The P104-100 features 4 GB of GDDR5X memory on a 256-bit bus, delivering 320.3 GB/s of bandwidth. The memory clock is 1251 MHz, translating to 10 Gbps effective transfer rate. This configuration is well-balanced for the compute workloads the card targets: the 256-bit bus provides sufficient width for high-throughput operations, while GDDR5X offers superior bandwidth per pin compared to standard GDDR5. The 320.3 GB/s bandwidth is competitive with the nearest rivals, though the 4 GB capacity is modest by modern standards.

For high-resolution compute tasks, the 4 GB capacity is the primary constraint. The 320.3 GB/s bandwidth can move large data blocks efficiently, but if a workload requires more than 4 GB of working set, the card must either partition the data or fall back to system memory, which is bottlenecked by the PCIe 1.0 x4 interface. The pixel rate of 110.9 GPixel/s and texture rate of 208.0 GTexel/s are irrelevant without display outputs, but they indicate the memory subsystem can sustain heavy parallel loads. In practice, the P104-100 is best suited for workloads with small-to-medium memory footprints, such as cryptocurrency hash functions or array-based calculations, where the 320.3 GB/s bandwidth can be fully exploited without exceeding the 4 GB limit.

FAQ

Q: Is the NVIDIA P104-100 suitable for gaming?

A: No. The card has no display outputs, making it impossible to connect a monitor. Its 3DMark Steel Nomad score of 1,413 also indicates poor rasterized performance, and the PCIe 1.0 x4 interface would bottleneck game asset streaming.

Q: What is the average benchmark score of the P104-100?

A: The average benchmark score is 32,747, which places it at the 76th percentile of all GPUs in the database. It is 0.3% slower than the NVIDIA T600 Mobile, 0.4% faster than the AMD Radeon RX 7800 XT, 0.6% faster than the AMD FirePro S9300 X2, and 1.1% faster than the AMD FirePro S10000.

Q: How much power does the P104-100 require?

A: The suggested PSU rating is 200 W, and the card uses a single 8-pin power connector. No official TDP is listed in the data, but the 200 W PSU recommendation indicates a modest power draw relative to its compute performance.

Q: What memory configuration does the P104-100 use?

A: It has 4 GB of GDDR5X memory on a 256-bit bus, with a bandwidth of 320.3 GB/s. The memory clock is 1251 MHz, running at 10 Gbps effective. This configuration supports high-throughput compute but limits working sets to 4 GB.

Q: Can the P104-100 handle modern APIs like Vulkan and DirectX 12?

A: Yes. The card supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. Its Vulkan benchmark score is 45,165, and its OpenCL score is 51,663, indicating strong compute API performance despite the lack of display outputs.

Q: What is the card's physical footprint and interface?

A: The P104-100 is a dual-slot card measuring 267 mm (10.5 inches) in length. It uses a PCIe 1.0 x4 bus interface, which is a significant bottleneck for host-device data transfers but acceptable for compute tasks that load data once and process in bulk.