Intel UHD Graphics P630 vs NVIDIA GeForce GTX 670MX Comparison

The NVIDIA GeForce GTX 670MX and Intel UHD Graphics P630 represent two fundamentally different approaches to mobile graphics, separated by nearly eight years of silicon evolution. The GTX 670MX is a discrete Kepler part from 2012, built on TSMC's 28 nm process with a massive 3,540 million transistors spread across a 294 mm² die. The UHD P630 is an integrated graphics processor within Intel's Comet Lake GT2 chip, fabricated on Intel's 14 nm+++ process, and relies on system memory rather than dedicated VRAM. The data shows a 33rd percentile ranking for the NVIDIA part versus a 31st percentile for Intel, indicating both are positioned in the lower-to-mid range of the overall GPU landscape, but their architectural philosophies could not be more different.

Architecture Differences

The GTX 670MX is built on NVIDIA's Kepler architecture, the same design lineage that powered the desktop GTX 600 series. It employs 960 shading units, 80 texture mapping units, and 24 ROPs, all fed by a 192-bit memory bus connected to 3 GB of GDDR5 memory. This dedicated memory configuration delivers 67.20 GB/s of bandwidth, a figure that is fully independent of the host system's RAM. The chip's base and boost clocks are identical at 601 MHz, suggesting a conservative power envelope for a notebook part rated at 75 W TDP. Its 28 nm process yields a transistor density of 12.0M / mm², which was competitive for its era but pales against modern integration.

The Intel UHD P630, in contrast, is a Generation 9.5 architecture part, specifically the Comet Lake GT2 variant. It packs 192 shading units, 24 TMUs, and only 3 ROPs — a drastically simplified pipeline. Its memory subsystem is entirely "System Shared," meaning it borrows from the CPU's main memory pool, with bandwidth described as "System Dependent." The clock behavior is the opposite of the NVIDIA part: a 350 MHz base that boosts up to 1200 MHz, allowing the iGPU to ramp up when thermal and power budgets permit. The TDP is just 15 W, making it a fraction of the discrete card's power draw. Perhaps most tellingly, the Intel part supports DirectX 12 (12_1) and Vulkan 1.3, while the older NVIDIA chip only reaches DirectX 12 (11_0) and Vulkan 1.2.175, reflecting its earlier design era.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The NVIDIA GeForce GTX 670MX averages 5721 points across its benchmark suite, while the Intel UHD Graphics P630 averages 5370 points. This gives the NVIDIA part a 6.5% advantage in overall performance, though the gap is smaller than the architectural differences might suggest.

Q: How do the two compare in OpenCL performance specifically?

A: The GTX 670MX scores 6125 in Geekbench OpenCL, which is 19.8% higher than the UHD P630's 5111. This is the NVIDIA card's strongest win, likely attributable to its dedicated GDDR5 memory and higher raw shader count.

Q: Does the Intel UHD P630 win any benchmark?

A: Yes, the Intel part wins the Geekbench Vulkan test with a score of 5628 versus the GTX 670MX's 5316, a 5.5% advantage. This suggests the newer architecture handles modern API overhead more efficiently despite having far fewer shading units.

Q: What are the nearest rivals for each GPU?

A: The GTX 670MX sits within 0.2% of the Intel Iris Pro Graphics P6300 (5712) and 0.2% of the NVIDIA GeForce GTX 550 Ti (5731). The UHD P630 is 0.2% ahead of the AMD Radeon R7 M445 (5358) and 0.8% behind the AMD Radeon R7 M365X (5416).

Q: Which GPU has a higher pixel fill rate?

A: The NVIDIA GTX 670MX achieves 12.02 GPixel/s, whereas the Intel UHD P630 manages only 3.600 GPixel/s. The NVIDIA part is 3.34 times faster in this metric, reflecting its 24 ROPs versus the Intel's 3.

Q: What is the texture fill rate difference?

A: The GTX 670MX delivers 48.08 GTexel/s, while the UHD P630 reaches 28.80 GTexel/s. The NVIDIA card is 67% faster here, driven by its 80 TMUs versus the Intel's 24.

Where Each One Wins

The GTX 670MX clearly dominates in compute-heavy workloads that rely on raw shader throughput and dedicated memory bandwidth. Its 1,153.9 GFLOPS of FP32 performance is 2.5 times the UHD P630's 460.8 GFLOPS, making it the obvious choice for tasks like OpenCL-accelerated rendering, video encoding, or any legacy application that leverages GPGPU compute. The 3 GB of GDDR5 memory also gives it a massive advantage in workloads that need to hold large datasets locally, rather than constantly fetching from system RAM.

The Intel UHD P630 wins in scenarios where API compatibility and modern feature support matter more than brute force. Its Vulkan 1.3 support and DirectX 12 (12_1) compliance mean it can run newer games and applications that leverage these APIs more efficiently. The 5.5% Vulkan win over the GTX 670MX, despite having only 20% of the shading units, indicates that architectural efficiency and driver optimization for modern APIs can overcome raw hardware deficits. Furthermore, its 15 W TDP and integrated nature make it the clear winner for battery-constrained environments where the discrete card's 75 W draw is simply impractical.

Head-to-Head Benchmarks

The two available benchmarks paint a picture of a split decision. In Geekbench OpenCL, the GTX 670MX posts 6125 against the UHD P630's 5111, a 19.8% margin that aligns with the raw compute gap. This result makes sense given the NVIDIA part's 960 shaders operating at a sustained 601 MHz with dedicated GDDR5 bandwidth of 67.20 GB/s. The Intel iGPU, with only 192 shaders and shared memory, simply cannot match this sustained throughput in a compute-heavy API.

Vulkan tells a different story. Here, the UHD P630 scores 5628, edging out the GTX 670MX's 5316 by 5.5%. This is a remarkable outcome when considering that the Intel part has 5 times fewer shading units and operates at a lower base clock. The likely explanation lies in the API support: the Intel chip natively supports Vulkan 1.3, while the NVIDIA part only reaches 1.2.175. Additionally, the Intel's boost clock of 1200 MHz, which is 599 MHz higher than the NVIDIA's fixed 601 MHz, suggests that the iGPU can aggressively turbo under lighter loads, potentially closing the gap in latency-sensitive workloads.

Notably, the GTX 670MX's nearest rival, the Intel Iris Pro P6300, scores 5712, which is nearly identical to the GTX 670MX's 5721 average. The UHD P630's closest competitor, the AMD Radeon R7 M445, scores 5358, just 0.2% below it. These rivalries show that both cards are clustered tightly within their respective performance tiers, with no single GPU holding a dominant edge in their immediate class.

The Verdict

For users who prioritize raw compute performance, the NVIDIA GeForce GTX 670MX is the data-backed choice. Its 19.8% OpenCL advantage and 2.5x higher FP32 throughput make it unequivocally stronger for compute tasks. The 3 GB of dedicated GDDR5 memory also provides a stability and predictability that shared-memory iGPUs cannot offer, since the Intel part's bandwidth is "System Dependent" and varies with the host CPU and RAM configuration.

However, for modern gaming and API-forward workloads, the Intel UHD Graphics P630 is the more future-proof option. Its Vulkan 1.3 and DirectX 12 (12_1) support, combined with a 5.5% Vulkan benchmark win, indicate that it handles contemporary graphics APIs more efficiently. The 15 W TDP also means it can operate in thin-and-light systems where a 75 W discrete GPU is physically impossible to integrate. The data suggests that if your primary use case is playing modern titles that leverage Vulkan, the Intel part's efficiency may yield a better experience than the older NVIDIA architecture.

Ultimately, the choice hinges on workload: compute-heavy legacy applications favor the GTX 670MX, while modern API-based gaming and power-constrained usage favor the UHD P630.

Specification Differences

| Specification | NVIDIA GeForce GTX 670MX | Intel UHD Graphics P630 |

|---|---|---|

| Architecture | Kepler | Generation 9.5 |

| Process Node | 28 nm | 14 nm+++ |

| Foundry | TSMC | Intel |

| Transistors | 3,540 million | Not specified |

| Die Size | 294 mm² | Not specified |

| Base Clock | 601 MHz | 350 MHz |

| Boost Clock | 601 MHz | 1200 MHz |

| Memory Size | 3 GB | System Shared |

| Memory Type | GDDR5 | System Shared |

| Memory Bus Width | 192 bit | System Shared |

| Memory Bandwidth | 67.20 GB/s | System Dependent |

| Shading Units | 960 | 192 |

| Texture Mapping Units | 80 | 24 |

| Raster Output Units | 24 | 3 |

| Pixel Rate | 12.02 GPixel/s | 3.600 GPixel/s |

| Texture Rate | 48.08 GTexel/s | 28.80 GTexel/s |

| FP32 Performance | 1,153.9 GFLOPS | 460.8 GFLOPS |

| TDP | 75 W | 15 W |

| Bus Interface | PCIe 3.0 x16 | Ring Bus |

| DirectX Support | 12 (11_0) | 12 (12_1) |

| Vulkan Support | 1.2.175 | 1.3 |

| Release Date | 2012-09-30 | 2020-05-12 |