Intel UHD Graphics P630 vs NVIDIA GeForce GTX 765M Comparison

The NVIDIA GeForce GTX 765M and Intel UHD Graphics P630 represent two fundamentally different approaches to mobile graphics: a dedicated discrete GPU from 2013 versus an integrated processor graphics solution from 2020. The benchmark data shows the GTX 765M winning both shared head-to-head tests, with a 40.4% margin in OpenCL and a 19.3% margin in Vulkan, yet the two chips end up remarkably close in overall average score — 5501 for the NVIDIA part versus 5370 for the Intel part. This narrow overall gap, combined with significant per-test differences, makes the choice between them highly dependent on the specific workload.

Where Each One Wins

The NVIDIA GeForce GTX 765M is the clear winner in raw compute throughput. In the Geekbench OpenCL test, it scores 7176 against the Intel UHD Graphics P630’s 5111, a 40.4% advantage. This is the largest single delta between the two parts and reflects the GTX 765M’s dedicated memory bandwidth of 64.13 GB/s and its 768 shading units. The Vulkan test narrows the gap but still favors NVIDIA: 6714 versus 5628, a 19.3% margin. For any workload that leverages general-purpose GPU compute or modern graphics APIs, the GTX 765M is decisively ahead.

The Intel UHD Graphics P630’s wins are more subtle and lie outside the head-to-head benchmark suite. It supports DirectX 12 (12_1) and Vulkan 1.3, whereas the GTX 765M is limited to DirectX 12 (11_0) and Vulkan 1.2.175. This means the Intel part can run newer feature-level graphics applications without compatibility layers. The P630 also has a much higher boost clock ratio — its 1200 MHz boost is 3.4 times its 350 MHz base, versus the GTX 765M’s 863 MHz boost against a 797 MHz base — suggesting better thermal scaling in bursty, short-duration workloads. However, the P630 does not win a single shared benchmark in the data; its advantage is purely in API support and integration efficiency, not in measured performance.

Architecture Differences

The two GPUs are built on vastly different foundations. The NVIDIA GeForce GTX 765M uses the GK106 chip on the Kepler architecture, fabricated on a 28 nm process at TSMC. It packs 2,540 million transistors into a 221 mm² die, yielding a transistor density of 11.5M per mm². The Intel UHD Graphics P630 uses the Comet Lake GT2 chip on the Generation 9.5 architecture, built on Intel’s 14 nm+++ process. Intel does not disclose transistor count or die size for this part, but the process node alone — 14 nm+++ versus 28 nm — indicates a significant generational gap in manufacturing technology.

The compute resources differ dramatically. The GTX 765M has 768 shading units, 64 texture mapping units, and 16 raster output units. The P630 has 192 shading units, 24 TMUs, and only 3 ROPs. This 4:1 ratio in shaders and 21:1 in ROPs explains why the NVIDIA part dominates in fill-rate-bound and shader-heavy tasks. Pixel rate is 13.81 GPixel/s for the GTX 765M versus 3.600 GPixel/s for the P630, and texture rate is 55.23 GTexel/s versus 28.80 GTexel/s. FP32 compute is 1,325.6 GFLOPS versus 460.8 GFLOPS — a 2.88x advantage for NVIDIA. Interestingly, the P630 does support FP16 at 921.6 GFLOPS with a 2:1 ratio, something the GTX 765M lacks entirely.

Memory architecture also diverges sharply. The GTX 765M has 2 GB of dedicated GDDR5 on a 128-bit bus, delivering 64.13 GB/s of bandwidth. The P630 uses system shared memory, with a bus width and bandwidth listed as "System Dependent." This means the Intel part’s performance scales with the host system’s memory configuration, while the NVIDIA part has predictable, fixed bandwidth. The GTX 765M draws 75 W under TDP, while the P630 is rated at just 15 W, reflecting the discrete versus integrated design philosophy.

The Verdict

From the data, the NVIDIA GeForce GTX 765M is the superior performer in every shared benchmark. It wins OpenCL by 40.4% and Vulkan by 19.3%, and its average benchmark score of 5501 places it at the 32nd percentile of all GPUs, slightly ahead of the P630’s 5370 and 31st percentile. The GTX 765M’s nearest rivals — the NVIDIA GeForce MX130 at 5508 (-0.1% delta), the AMD Radeon R7 M440 at 5483 (+0.3%), and the NVIDIA Quadro M4000 at 5467 (+0.6%) — all cluster within a fraction of a percent, indicating the GTX 765M sits in a stable, well-populated performance tier. The P630’s rivals are similarly close, with the AMD Radeon R7 M365X at 5416 (-0.8%) and the NVIDIA GeForce 840M at 5322 (+0.9%).

For users who need maximum compute performance in a laptop, the GTX 765M is the clear choice. Its 40.4% OpenCL lead means CUDA-accelerated or OpenCL-based applications will run substantially faster. For users who prioritize modern API support, the P630’s DirectX 12 (12_1) and Vulkan 1.3 compatibility gives it a software advantage that the GTX 765M cannot match, despite its lower raw scores. The P630’s 15 W TDP also makes it far more suitable for thin-and-light systems where power draw and heat are critical constraints.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA GeForce GTX 765M has an average benchmark score of 5501, compared to the Intel UHD Graphics P630’s 5370. The GTX 765M also holds a higher percentile rank at 32 versus 31.

Q: How much faster is the GTX 765M in OpenCL?

A: The GTX 765M scores 7176 in Geekbench OpenCL, which is 40.4% higher than the P630’s 5111. This is the largest performance gap between the two in any shared test.

Q: Does the Intel UHD Graphics P630 support newer graphics APIs?

A: Yes. The P630 supports DirectX 12 (12_1) and Vulkan 1.3, while the GTX 765M is limited to DirectX 12 (11_0) and Vulkan 1.2.175. This gives the Intel part better forward compatibility.

Q: What is the memory configuration difference?

A: The GTX 765M has 2 GB of dedicated GDDR5 memory on a 128-bit bus with 64.13 GB/s bandwidth. The P630 uses system shared memory, with bandwidth and bus width listed as "System Dependent."

Q: How do the power requirements compare?

A: The GTX 765M has a TDP of 75 W and uses an MXM module slot, while the P630 is an IGP with a 15 W TDP and connects via Ring Bus. This makes the Intel part far more power-efficient.

Q: Which GPU has more shading units?

A: The GTX 765M has 768 shading units, exactly four times the P630’s 192. The GTX 765M also has 64 TMUs versus 24, and 16 ROPs versus 3.

Head-to-Head Benchmarks

The two shared benchmarks tell a story of consistent NVIDIA dominance, but with very different margins. In Geekbench OpenCL, the GTX 765M posts 7176 against the P630’s 5111, a 40.4% lead. This massive delta reflects the NVIDIA part’s dedicated GDDR5 memory — 64.13 GB/s of fixed bandwidth versus the P630’s system-dependent shared memory — and its 768 shading units. The P630’s 460.8 GFLOPS FP32 compute is simply overwhelmed by the GTX 765M’s 1,325.6 GFLOPS. For OpenCL workloads like physics simulation, video encoding, or data-parallel compute, the GTX 765M is in a different class.

The Vulkan test narrows the gap considerably but still favors NVIDIA. The GTX 765M scores 6714, a 19.3% advantage over the P630’s 5628. This closer margin suggests that the P630’s newer architecture — Generation 9.5 on 14 nm+++ — handles Vulkan’s driver overhead more efficiently than the older Kepler design. The P630’s 1200 MHz boost clock, which is 3.4x its 350 MHz base, may also help in bursty Vulkan workloads. However, the GTX 765M’s superior pixel rate of 13.81 GPixel/s versus 3.600 GPixel/s and texture rate of 55.23 GTexel/s versus 28.80 GTexel/s still carry the day in render-heavy scenes.

Notably, the GTX 765M has a third benchmark score — 2612 in Geekbench Metal — which the P630 does not share. This gives the NVIDIA part broader API coverage, particularly for Apple ecosystem applications. The P630, meanwhile, has an FP16 capability of 921.6 GFLOPS (2:1), which the GTX 765M lacks entirely, suggesting the Intel part could be more efficient in certain AI or machine-learning inference tasks that use half-precision arithmetic. Yet without a shared benchmark to confirm this, the data can only support the GTX 765M’s superiority in the tests that exist. Overall, the GTX 765M wins both head-to-head tests, taking 2 wins to 0, and its 40.4% OpenCL margin is the single most decisive data point in this comparison.