NVIDIA P104-101

Benchmark Performance

The NVIDIA P104-101 is a specialized mining-oriented graphics card built on the Pascal architecture. Its benchmark data is unusual: the card holds a 50th percentile position among all GPUs tracked in the database, yet its average benchmark score is recorded as zero. This discrepancy stems from the card's design purpose, it was engineered for computational workloads rather than traditional rendering benchmarks, and the database shows no standardized test scores for it.

The silicon itself is the GP104 chip, manufactured on TSMC's 16 nm process with 7,200 million transistors packed into a 314 mm² die. That yields a transistor density of 22.9 million per square millimeter. The chip's raw compute specifications tell a clear story about its intended use. The P104-101 delivers 8.617 TFLOPS of FP32 compute, which places it in the upper tier of Pascal-based accelerators. Its FP16 throughput, however, is dramatically limited at 134.6 GFLOPS, operating at a 1:64 ratio relative to FP32. This is a deliberate design choice, mining workloads of that era relied almost exclusively on FP32 and integer math, so the card's hardware simply does not allocate resources to half-precision compute.

The card's clock behavior further reinforces its mining pedigree. The base clock sits at 1506 MHz, boosting to 1683 MHz under load. These are respectable frequencies for the architecture, but the card lacks any game-mode clock adjustment, as the `game` field is null. The memory clock runs at 2002 MHz, translating to 8 Gbps effective data rate. The pixel rate is 107.7 GPixel/s, and the texture rate reaches 269.3 GTexel/s. These figures are competitive with gaming-oriented Pascal cards, but the P104-101's lack of display outputs makes such comparisons largely academic.

What the data shows is a card that would have excelled at parallel FP32 workloads, hash calculations, matrix operations, and similar compute tasks, while being wholly unsuitable for interactive graphics. The absence of any benchmark entries in the the benchmark database suggests that the database's testing methodology either could not run standard graphics benchmarks on this hardware or that the card was never submitted for such testing. The percentile rank of 50 indicates that when compared across all GPUs in the database, this card sits exactly at the median, but that median position is computed from a zero score, which skews the interpretation.

Ray Tracing and Feature Set

The P104-101 predates dedicated ray tracing hardware. The the benchmark database lists no RT cores and no tensor cores for this GPU. This is consistent with its Pascal architecture, which was released before NVIDIA introduced its RTX lineup. Ray tracing on Pascal hardware is theoretically possible via compute shaders, but the absence of dedicated acceleration units means any such workload would run at a severe performance penalty compared to later architectures.

The API support, however, is broader than one might expect for a mining card. The P104-101 supports DirectX 12 with feature level 12_1, OpenGL 4.6, and Vulkan 1.4. The DirectX 12_1 feature level includes support for conservative rasterization and rasterizer-ordered views, though these features are irrelevant for a card with no display outputs. The Vulkan 1.4 support is notable, it represents a modern API revision that allows for low-level hardware access, which could theoretically be used for compute workloads beyond mining. OpenGL 4.6 support similarly provides a mature compute shader interface.

The card's physical configuration reveals its specialized nature. It is dual-slot, measuring 267 mm (10.5 inches) in length. It requires a single 8-pin power connector, and NVIDIA recommends a 200 W power supply for a system containing this card. The bus interface is PCIe 1.0 x4, a severely limited connection compared to the PCIe 3.0 x16 that was standard for gaming GPUs of this era. This bandwidth constraint would not significantly hinder mining workloads, which mostly operate within the card's local memory, but it would cripple any data-transfer-heavy compute task.

There are no display outputs whatsoever. The `displayOutputs` field is explicitly "No outputs." This means the card cannot be used for any visual output, making it strictly a compute accelerator. The feature set, therefore, is entirely oriented toward raw number crunching with no consideration for graphics presentation or interactive rendering.

Memory Subsystem

The P104-101 comes equipped with 4 GB of GDDR5 memory on a 256-bit bus. Memory bandwidth is rated at 256.3 GB/s, which is a substantial figure for the era. The memory clock runs at 2002 MHz, yielding an 8 Gbps effective data rate across the bus.

This memory configuration is notable for a mining card. The 4 GB capacity was adequate for the Dagger-Hashimoto algorithm (Ethereum's PoW) of that period, which required roughly 2-3 GB of VRAM. The 256-bit bus width is also well-suited to memory-intensive workloads, as the wide bus allows for high sustained bandwidth. The 256.3 GB/s bandwidth figure is competitive with contemporary gaming cards, though it is importantly mining algorithms of the time were often more sensitive to memory latency and capacity than raw bandwidth.

For high-resolution rendering, this memory subsystem would be insufficient by modern standards, 4 GB is below the 8 GB baseline that current AAA games expect at 1080p. But the card was never intended for rendering. The memory configuration appears optimized for the specific access patterns of cryptocurrency mining, which involve repeated reads and writes to a working set that fits within the VRAM. The 256-bit bus reduces the number of memory transactions needed to move data between the GPU and VRAM, which improves energy efficiency per hash.

The lack of any display outputs means this memory is exclusively devoted to compute. There is no frame buffer allocation, no desktop compositing overhead, and no video decode buffer requirements. Every byte of the 4 GB is available for computation. This explains why the card could operate with a relatively modest 4 GB capacity, it had no graphics-related memory demands competing with its compute tasks.

FAQ

Q: Why does the P104-101 have no display outputs?

A: The card is specifically designed for mining workloads, not graphics rendering. The the benchmark database lists "No outputs" for display connections, indicating that all hardware resources are dedicated to computation.

Q: What is the FP16 compute capability of this card?

A: The P104-101 delivers 134.6 GFLOPS of FP16 performance, which operates at a 1:64 ratio relative to its FP32 throughput of 8.617 TFLOPS. This extremely low half-precision performance indicates the card was not designed for AI or machine learning workloads.

Q: Can this card run DirectX 12 applications?

A: Yes, the card supports DirectX 12 with feature level 12_1. However, since it has no display outputs, it cannot render graphics to a screen. The API support is present but functionally irrelevant for gaming purposes.

Q: What power supply is recommended for this card?

A: The suggested PSU rating is 200 W. The card uses a single 8-pin power connector and occupies a dual-slot form factor.

Q: How much memory does the P104-101 have and what type is it?

A: The card has 4 GB of GDDR5 memory on a 256-bit bus, providing 256.3 GB/s of bandwidth. The memory operates at 2002 MHz (8 Gbps effective).

Q: What is the production status of this card?

A: The P104-101 is marked as "End-of-life" in the database. It was released on January 4, 2018, and has since been discontinued.

How It Compares

The the benchmark database provides no nearest rivals for the P104-101, and its benchmark scores are recorded as zero. This makes direct quantitative comparison impossible. What can be said is that the card sits at the 50th percentile among all GPUs in the database, which is a statistical position derived from its zero benchmark score. This suggests that the database's performance ranking treats the card as having no measurable graphics performance, which is consistent with its mining-only design.

Without rival data, the comparison must focus on architectural positioning. The P104-101 is a Pascal-generation GPU built on TSMC's 16 nm process, sharing the GP104 die with gaming cards of that era. However, its feature set diverges significantly: no RT cores, no tensor cores, no display outputs, and a PCIe 1.0 x4 interface. These omissions place it firmly outside the gaming and professional visualization markets.

The card's 8.617 TFLOPS FP32 throughput would have been competitive with mid-range gaming GPUs of its time, but the lack of rendering hardware means it cannot be evaluated on the same terms. Its 4 GB GDDR5 memory with 256.3 GB/s bandwidth is modest by today's standards but was appropriate for the mining algorithms of its release period. The end-of-life status and absence of benchmark entries in the database indicate that this card's relevance has fully passed, leaving it as a historical footnote in the mining GPU era rather than a participant in current performance comparisons.