NVIDIA Tesla P6 Mobile

NVIDIA Tesla P6 Mobile is a professional mobile graphics solution built on the 16 nm Pascal architecture from TSMC. It uses the GP104 chip, which contains 7,200 million transistors on a 314 mm² die. The card operates at a base clock of 1012 MHz and a boost clock of 1506 MHz, delivering a peak FP32 throughput of 6.169 TFLOPS. With a 90 W TDP and an MXM Module slot width, this is an end-of-life product released on 2017-03-23, positioned between the Tesla Maxwell and Tesla Volta generations in the professional lineup.

How It Compares

The Tesla P6 Mobile holds a 50th percentile ranking among all GPUs, indicating it sits exactly at the midpoint of the performance distribution. This means half of all tracked graphics cards deliver higher benchmark scores, while the other half deliver lower scores. In practical terms, this places the P6 Mobile in a solidly mid-range position for its era, neither a flagship nor an entry-level part. The data shows a balanced overall standing that reflects its professional orientation rather than gaming-focused design.

When compared to its predecessor in the Tesla Maxwell generation, the P6 Mobile benefits from the architectural shift to Pascal. The move from Maxwell to Pascal brought improvements in clock efficiency and feature support, which the benchmark percentile reflects. The P6 Mobile's 6.169 TFLOPS FP32 rate and 192.3 GB/s memory bandwidth are characteristic of the Pascal generation's focus on compute density and memory throughput. This generational step provides a meaningful performance uplift over older Tesla Maxwell parts, though the specific delta percentage is not available in the data.

Against its successor, the Tesla Volta generation, the P6 Mobile occupies an earlier point in NVIDIA's professional roadmap. The Pascal architecture lacks the dedicated tensor cores and RT cores that Volta introduced, meaning the P6 Mobile cannot accelerate AI inference or ray tracing workloads in hardware. The benchmark results indicate that the P6 Mobile's 50th percentile standing would be lower relative to Volta parts, which benefit from newer compute features. However, the exact performance gap is not quantified in the nearestRivals data.

The P6 Mobile's 16 GB GDDR5 memory capacity is notable for a mobile professional card of its time. This large frame buffer, combined with a 256-bit bus and 192.3 GB/s bandwidth, allows the card to handle sizable datasets and high-resolution textures without constant memory swapping. The 90 W TDP suggests a power-efficient design that fits within the thermal constraints of a portable workstation chassis, trading some raw performance for mobility.

Memory Subsystem

The Tesla P6 Mobile is equipped with 16 GB of GDDR5 memory, which is a substantial capacity for a mobile professional GPU. The memory operates at 1502 MHz, translating to 6 Gbps effective data rate per pin. The 256-bit memory interface provides a total bandwidth of 192.3 GB/s, which is the maximum rate at which data can be transferred between the GPU cores and the frame buffer.

This memory configuration has direct implications for high-resolution workloads. At 4K and beyond, the 16 GB capacity allows for large textures, complex scene geometry, and multiple render targets to reside entirely in VRAM. The 192.3 GB/s bandwidth ensures that the 2048 shading units have a steady stream of data to process, reducing stalls that would otherwise occur with a narrower interface. For compute tasks that involve large matrices or deep neural networks, the 16 GB pool is a practical advantage, enabling larger batch sizes without spilling to system memory.

The pixel rate of 96.38 GPixel/s and texture rate of 192.8 GTexel/s are consistent with the memory bandwidth available. These rates indicate the card can fill framebuffers and sample textures at speeds that match its FP32 compute throughput. In practice, this means the P6 Mobile can drive high-resolution displays and handle multi-monitor configurations, though the display outputs are marked as "Portable Device Dependent," meaning the actual connector layout varies by the host laptop or mobile workstation.

Ray Tracing and Feature Set

The Tesla P6 Mobile does not include dedicated ray tracing cores or tensor cores, as these hardware units were introduced in later generations. The architecture is Pascal, which predates the RTX lineup, so ray tracing workloads must be handled through compute shaders or OpenGL extensions rather than dedicated hardware acceleration. This limits the card's suitability for real-time ray tracing in professional visualization tools that rely on RT core acceleration.

In terms of API support, the P6 Mobile supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. The DirectX 12_1 feature level indicates support for conservative rasterization and rasterizer-ordered views, which are relevant for advanced rendering techniques. Vulkan 1.4 support provides low-overhead access to the GPU, beneficial for compute-heavy applications that need fine-grained control over command buffers and memory management. OpenGL 4.6 covers legacy professional applications that have not yet migrated to Vulkan or DirectX 12.

The FP16 throughput is 96.38 GFLOPS, which is a 1:64 ratio relative to FP32 performance. This indicates that half-precision compute is heavily de-emphasized in this architecture. For workloads that rely on FP16 acceleration, such as certain machine learning inference tasks, the P6 Mobile would perform poorly compared to GPUs with native FP16 support. The 2048 shading units are the primary compute resource, and their FP32 performance is the main metric for general-purpose compute.

FAQ

Q: What is the memory bandwidth of the Tesla P6 Mobile?

A: The memory bandwidth is 192.3 GB/s, derived from 16 GB of GDDR5 memory on a 256-bit bus running at 6 Gbps effective.

Q: Does the Tesla P6 Mobile support hardware ray tracing?

A: No, the P6 Mobile has no RT cores. It relies on the Pascal architecture's compute shaders for any ray tracing effects, without dedicated acceleration.

Q: What is the FP32 compute throughput?

A: The FP32 throughput is 6.169 TFLOPS, based on 2048 shading units at a boost clock of 1506 MHz.

Q: What is the production status of this GPU?

A: The production status is end-of-life, meaning NVIDIA has discontinued manufacturing this part.

Q: Which APIs are supported by the Tesla P6 Mobile?

A: The card supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4.

Q: What is the transistor count and die size?

A: The GP104 chip contains 7,200 million transistors on a 314 mm² die, manufactured on a 16 nm process at TSMC.

Benchmark Performance

The Tesla P6 Mobile's benchmark performance is defined by its 50th percentile ranking among all GPUs, with an average benchmark score of zero in the dataset. This zero score indicates that no standardized benchmark results have been recorded for this specific card, making the percentile a derived estimate based on its hardware specifications relative to other GPUs. The 50th percentile placement is significant because it splits the performance distribution exactly in half, suggesting the card offers a balanced blend of compute and memory capabilities.

In the absence of direct benchmark scores, the raw specifications provide the basis for performance analysis. The FP32 rate of 6.169 TFLOPS is a strong figure for a 90 W mobile part, indicating efficient compute per watt. The pixel fill rate of 96.38 GPixel/s and texture fill rate of 192.8 GTexel/s are consistent with a mid-range professional GPU. These rates suggest the card can handle 1440p and 4K rendering tasks at reasonable frame rates in professional applications, though not at the level of high-end desktop parts.

The memory subsystem is a key differentiator. With 16 GB of GDDR5 and 192.3 GB/s bandwidth, the P6 Mobile outperforms many mobile gaming GPUs of its generation in memory capacity, which is crucial for large dataset compute workloads. The 256-bit bus width ensures that the bandwidth is evenly distributed across the memory channels, reducing latency spikes during heavy access patterns. This makes the card well-suited for scientific visualization, finite element analysis, and other memory-intensive professional tasks.

Comparing to rivals, the lack of nearestRivals data means the 50th percentile is the primary positional reference. This percentile implies that the P6 Mobile sits at the median of all GPUs ever tracked by the benchmark database, which includes both consumer and professional parts. For a mobile professional card, this positioning is respectable, as it must balance thermal and power constraints against compute capability. The 90 W TDP is a limiting factor, preventing the card from reaching the performance levels of 250 W desktop counterparts, but it enables deployment in thin-and-light mobile workstations.

The FP16 performance of 96.38 GFLOPS, at a 1:64 ratio to FP32, is a notable weakness. Modern compute frameworks increasingly rely on FP16 for deep learning training and inference, and the P6 Mobile's poor FP16 throughput would cause significant slowdowns in such workloads. This contrasts sharply with its FP32 capabilities, which are adequate for traditional HPC tasks. The memory bandwidth, while sufficient for FP32, might become a bottleneck for FP16 operations that require high data throughput with reduced precision.

The pixel rate of 96.38 GPixel/s and texture rate of 192.8 GTexel/s are directly proportional to the clock speed and the number of ROPs and TMUs. With 64 ROPs and 128 TMUs, the card can sustain high fill rates at the boost clock of 1506 MHz. These figures are competitive for a mobile part, enabling smooth rendering of complex scenes with multiple texture layers. The 16 GB memory capacity further supports this by allowing large texture atlases to be stored on-chip.

Overall, the data indicates that the Tesla P6 Mobile is a capable mid-range professional GPU, excelling in memory capacity and FP32 compute per watt, but lacking in FP16 performance and hardware acceleration for ray tracing or tensor operations. Its 50th percentile standing confirms it as a balanced performer, suitable for a range of professional applications that do not require the latest AI features. The end-of-life status suggests it has been superseded by newer parts, but its specifications remain relevant for legacy systems and specific compute tasks where 16 GB VRAM is a requirement.