GPU Comparison

Head-to-Head Benchmarks

The benchmark data shows a clear overall victory for the NVIDIA P106-090, which wins both head-to-head tests against the GeForce GTX 950. The most decisive result comes in the Geekbench OpenCL test, where the P106-090 scores 21,304 against the GTX 950's 15,662. That is a 36% lead, a substantial margin that reflects the mining card's advantage in raw compute throughput. The gap is narrower in the Geekbench Vulkan test, but the P106-090 still comes out ahead with 18,596 points versus 16,734 points for the GTX 950, a delta of 11.1%.

Looking at the broader average benchmark score, the P106-090 posts an average of 13,470 across all tests, while the GTX 950 averages 13,189. That difference of roughly 281 points places the P106-090 in the 54th percentile of all GPUs, one point ahead of the GTX 950's 53rd percentile. The near-rival comparison reinforces the P106-090's position: its average score sits within 0.5% of the AMD Radeon Pro 555 and within 0.3% of the NVIDIA GeForce GTX 570. The GTX 950, meanwhile, lands within 0.9% of the NVIDIA Tesla M2090 and 0.8% of the GTX 480.

The P106-090's OpenCL advantage is especially notable because both cards share the same shading unit count of 768. Yet the P106-090's higher clock speeds, a 1,354 MHz base and 1,531 MHz boost versus 1,024 MHz and 1,188 MHz for the GTX 950, translate into a 28.9% faster FP32 throughput at 2.352 TFLOPS versus 1.825 TFLOPS. That compute headroom is the primary driver of its 36% OpenCL win. The Vulkan result, while still a win for the P106-090, is closer because API overhead and memory access patterns can mitigate raw compute differences. Still, the data shows the P106-090 winning every measured test, with no single benchmark where the GTX 950 pulls ahead.

Where Each One Wins

The P106-090 is the clear winner in compute-oriented workloads. Its OpenCL score of 21,304 demonstrates a commanding lead in general-purpose GPU compute, which matters for tasks like rendering, physics simulation, and data processing. The Vulkan score of 18,596 also favors the P106-090, indicating stronger performance in modern graphics APIs that leverage low-level hardware access. The P106-090's 3 GB of GDDR5 memory, paired with a 192-bit bus and 192.2 GB/s of bandwidth, gives it a memory bandwidth advantage of 81.7% over the GTX 950's 105.8 GB/s. That bandwidth headroom helps in texture-heavy scenes and larger working sets.

The GTX 950, despite losing both head-to-head tests, has its own areas of strength. It is the only card of the two with display outputs, 1x DVI, 1x HDMI 2.0, and 3x DisplayPort 1.2, making it the only viable option for actually driving a monitor. The P106-090 has no outputs at all, which is expected for a mining-focused product. The GTX 950 also has a more conventional bus interface: PCIe 3.0 x16 versus the P106-090's PCIe 1.0 x1. That means the GTX 950 can slot into standard desktop motherboards without bandwidth bottlenecks, while the P106-090's x1 interface severely limits data transfer to and from the host system. For gaming, the GTX 950's 2 GB of VRAM is smaller, but its 1,188 MHz boost clock and 57.02 GTexel/s texture rate still provide competent 1080p performance for its era. The GTX 950 also supports the same DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4 APIs as the P106-090, so software compatibility is identical.

The GTX 950's predecessor is the GeForce 700 series, and its successor is the GeForce 10 series, placing it in a well-trodden product lineage. The P106-090, by contrast, belongs to the "Mining GPUs" generation, with no predecessor or successor listed. That distinction matters: the P106-090 is a specialized tool, while the GTX 950 is a general-purpose graphics card.

Architecture Differences

The P106-090 is built on NVIDIA's Pascal architecture, fabricated on a 16 nm process at TSMC. The GTX 950 uses the older Maxwell 2.0 architecture on a 28 nm process, also at TSMC. The process node shrink is significant: 16 nm versus 28 nm allows the P106-090 to pack 4,400 million transistors into a 200 mm² die, yielding a transistor density of 22.0 million transistors per mm². The GTX 950's GM206 chip contains 2,940 million transistors on a larger 228 mm² die, resulting in a much lower density of 12.9 million per mm². The Pascal chip is physically smaller but electrically denser, which explains how it achieves higher clock speeds and efficiency despite a lower TDP.

The core configurations differ in one key area: both cards have 768 shading units and 48 TMUs, but the P106-090 has 48 ROPs versus the GTX 950's 32 ROPs. That 50% increase in ROP count directly contributes to the P106-090's pixel rate of 73.49 GPixel/s, nearly double the GTX 950's 38.02 GPixel/s. The texture rate also favors the P106-090 at 73.49 GTexel/s versus 57.02 GTexel/s, driven by its higher boost clock.

FP16 compute is another differentiator: the P106-090 lists 36.74 GFLOPS at a 1:64 ratio, while the GTX 950 has no FP16 data listed. That suggests the Pascal card has at least some half-precision capability, however limited, whereas the Maxwell part does not expose any. Both cards share the same API support, DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4, so there is no architectural difference in software feature levels, but the underlying hardware implementations are generations apart.

Specification Differences

The two cards diverge on nearly every measurable specification. The P106-090 uses the GP106 chip, while the GTX 950 uses GM206. Process node: 16 nm versus 28 nm. Transistor count: 4,400 million versus 2,940 million. Die size: 200 mm² versus 228 mm². Transistor density: 22.0M/mm² versus 12.9M/mm². Base clock: 1,354 MHz versus 1,024 MHz. Boost clock: 1,531 MHz versus 1,188 MHz. Memory clock: 2,002 MHz (8 Gbps effective) versus 1,653 MHz (6.6 Gbps effective). Memory size: 3 GB versus 2 GB. Memory bus width: 192-bit versus 128-bit. Memory bandwidth: 192.2 GB/s versus 105.8 GB/s. ROPs: 48 versus 32. Pixel rate: 73.49 GPixel/s versus 38.02 GPixel/s. Texture rate: 73.49 GTexel/s versus 57.02 GTexel/s. FP32: 2.352 TFLOPS versus 1.825 TFLOPS. TDP: 75 W versus 90 W. Bus interface: PCIe 1.0 x1 versus PCIe 3.0 x16. Display outputs: none versus 1x DVI, 1x HDMI 2.0, 3x DisplayPort 1.2. Dimensions: 250 mm (9.8 inches) versus 202 mm (8 inches). Release date: 2017-07-30 versus 2015-08-19. Generation: Mining GPUs versus GeForce 900.

The two cards are identical in several areas: 768 shading units, 48 TMUs, 1x 6-pin power connector, 250 W suggested PSU, dual-slot width, DirectX 12 (12_1), OpenGL 4.6, Vulkan 1.4, and end-of-life production status. The GTX 950 has a launch MSRP of 159 USD; the P106-090 has no listed MSRP.

FAQ

Q: Which card has a higher average benchmark score?

A: The NVIDIA P106-090 averages 13,470 across all tests, compared to 13,189 for the GeForce GTX 950. The P106-090 sits in the 54th percentile of all GPUs, while the GTX 950 is in the 53rd.

Q: How much faster is the P106-090 in OpenCL?

A: The P106-090 scores 21,304 in Geekbench OpenCL versus 15,662 for the GTX 950, a 36% advantage. This is the largest performance gap between the two cards in any tested workload.

Q: Does the GTX 950 support display outputs?

A: Yes. The GTX 950 includes 1x DVI, 1x HDMI 2.0, and 3x DisplayPort 1.2 outputs. The P106-090 has no display outputs, making it unsuitable for direct monitor connection.

Q: What is the memory bandwidth difference?

A: The P106-090 offers 192.2 GB/s over a 192-bit bus with 3 GB of GDDR5. The GTX 950 provides 105.8 GB/s over a 128-bit bus with 2 GB of GDDR5. The P106-090 has 81.7% more bandwidth.

Q: Which card has a lower TDP?

A: The P106-090 has a 75 W TDP, while the GTX 950 is rated at 90 W. Both require a 1x 6-pin power connector and a 250 W suggested PSU.

Q: Do both cards support the same APIs?

A: Yes. Both support DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4, despite the P106-090 using the older PCIe 1.0 x1 bus interface versus the GTX 950's PCIe 3.0 x16.