Intel UHD Graphics P750 vs NVIDIA GeForce GTX 680M Comparison

The NVIDIA GeForce GTX 680M and the Intel UHD Graphics P750 occupy opposite ends of the graphics hardware spectrum, yet they converge in the database’s performance percentiles. The GTX 680M sits at the 39th percentile, while the Intel part is at the 38th, placing them within a single percentage point of each other in overall standing. Despite this close overall ranking, the head-to-head benchmark data reveals a decisive gap in raw compute performance, driven by fundamentally different design philosophies, power envelopes, and target use cases. The following analysis breaks down the data to see where each part excels and where it falls short.

Head-to-Head Benchmarks

The only direct benchmark comparison available is Geekbench OpenCL, and it is not close. The NVIDIA GeForce GTX 680M scores 9230 points, while the Intel UHD Graphics P750 scores 6554 points. That is a delta of 40.8 percent, meaning the GTX 680M delivers roughly 41 percent higher performance in this compute-oriented test. This is a commanding margin, placing the NVIDIA part in a different performance tier for OpenCL workloads despite the two GPUs’ similar overall percentile rankings.

The magnitude of this win is contextualized by the nearest rival data. The GTX 680M’s average benchmark score is 7023, with its closest competitor, the NVIDIA T600, scoring 7035 (a delta of -0.2 percent). The Intel UHD Graphics P750’s average score is 6554, with the AMD Radeon HD 7730M at 6581 (a delta of -0.4 percent). The 40.8 percent gap in the head-to-head test is far larger than the differences seen between either card and its own nearest rivals. In practical terms, the GTX 680M’s OpenCL score is not merely ahead; it is in the neighborhood of the NVIDIA GeForce GTX 970, which scores 7157, while the Intel part lags behind even the modest AMD Radeon R5 M240, which scores 6975.

There are no reverse wins for the Intel part in the benchmark data. The wins tally stands at one for NVIDIA and zero for Intel. The Geekbench OpenCL result is the only data point, but it is a decisive one. For any workload that relies on OpenCL compute—such as certain video encoding filters, physics simulations, or GPGPU tasks—the GTX 680M is the clear choice. The Intel P750, by contrast, trails by a margin that no architectural feature can offset in this specific test.

Architecture Differences

The two GPUs are built on starkly different foundations. The NVIDIA GeForce GTX 680M uses the GK104 chip, based on the Kepler architecture, manufactured on a 28 nm process at TSMC. This chip packs 3,540 million transistors onto a die size of 294 mm², yielding a transistor density of 12.0M / mm². In contrast, the Intel UHD Graphics P750 is built on the Rocket Lake chip, using Intel’s Generation 12.1 architecture, fabricated on a 14 nm+++ process at Intel’s own fabs. Intel does not disclose transistor count or die size for this part, but the process node alone indicates a less dense design.

The compute resources differ dramatically. The GTX 680M features 1344 shading units, 112 texture mapping units (TMUs), and 32 raster output units (ROPs). The Intel P750 has 256 shading units, 64 TMUs, and 32 ROPs. While the ROP count is identical, the NVIDIA part has more than five times the shading units and nearly double the TMUs. This explains the raw throughput advantage: the GTX 680M achieves 2.038 TFLOPS of FP32 performance, while the Intel part manages only 665.6 GFLOPS. The Intel P750 does support FP16 at 1,331.2 GFLOPS with a 2:1 ratio, a feature the GTX 680M lacks entirely, but this does little to close the gap in general-purpose FP32 workloads.

Memory architecture is another fundamental split. The GTX 680M uses 4 GB of dedicated GDDR5 memory on a 256-bit bus, delivering 115.2 GB/s of bandwidth. The Intel P750 relies on System Shared memory, with its bandwidth listed as System Dependent. This means the Intel part’s performance is tied to the host system’s RAM speed and configuration, while the NVIDIA part has its own high-speed pool. In practice, dedicated memory eliminates contention with the CPU and provides consistent bandwidth, whereas shared memory can become a bottleneck.

Clock speeds also tell a story. The GTX 680M runs at a base clock of 719 MHz with a boost of 758 MHz, while the Intel P750 has a base of 350 MHz and a boost of 1300 MHz. The Intel part’s higher boost clock helps it compensate for fewer shading units, but the sheer difference in unit count remains insurmountable. The pixel rates reflect this: the GTX 680M hits 21.22 GPixel/s, while the Intel P750 reaches 41.60 GPixel/s. Interestingly, the Intel part has a higher pixel rate due to its higher clock and identical ROP count, yet its texture rate of 83.20 GTexel/s is slightly below the GTX 680M’s 84.90 GTexel/s.

Where Each One Wins

The data points to a clear split in use cases. The NVIDIA GeForce GTX 680M wins in compute-heavy scenarios, as evidenced by its 40.8 percent lead in Geekbench OpenCL. This makes it suitable for applications that offload parallel processing to the GPU, such as video transcoding with OpenCL filters, scientific computing kernels, or any workload that leverages the 2.038 TFLOPS FP32 throughput. Its dedicated 4 GB GDDR5 frame buffer also makes it a better fit for texture-heavy 3D rendering or gaming at moderate settings, where consistent memory bandwidth of 115.2 GB/s matters.

The Intel UHD Graphics P750, despite losing the compute benchmark, has its own strengths. Its 15 W TDP is a fraction of the GTX 680M’s 100 W, making it far more power-efficient. It is an integrated graphics processor (IGP) that uses the Ring Bus interface, which means it requires no separate power connectors and fits into ultraportable or low-power designs. Its higher pixel rate of 41.60 GPixel/s suggests it can handle basic display output and 2D acceleration efficiently. For office productivity, video playback, or light casual gaming, the P750’s capabilities are adequate, and its System Shared memory architecture reduces system cost and complexity.

The benchmark data does not favor the Intel part in any direct comparison, but the absence of a win does not mean absence of utility. The P750’s nearest rivals include the NVIDIA GeForce GTX 670M, which scores 6513, and the NVIDIA GeForce GT 555M at 6493. The Intel part’s score of 6554 places it slightly above those older discrete parts, indicating that it can hold its own against entry-level discrete GPUs from a previous generation. The GTX 680M, meanwhile, sits near the NVIDIA T600 and GeForce GTX 970, suggesting it is closer to modern entry-level workstation performance.

Specification Differences

The specification sheet reveals several fields where the two GPUs diverge, and some where they align. The most obvious difference is the process node: 28 nm for NVIDIA versus 14 nm+++ for Intel. The transistor count is only listed for the NVIDIA part (3,540 million), with Intel’s left unspecified. Die size is likewise only known for NVIDIA at 294 mm². The shading unit count is a major differentiator: 1344 versus 256. TMUs differ at 112 versus 64. ROPs are the same at 32.

Clock speeds vary significantly, with the GTX 680M’s base at 719 MHz versus the P750’s 350 MHz, and boost clocks at 758 MHz versus 1300 MHz. Memory configuration is completely different: 4 GB GDDR5 on a 256-bit bus with 115.2 GB/s bandwidth versus System Shared memory with System Dependent bandwidth. The FP32 performance is 2.038 TFLOPS versus 665.6 GFLOPS, and only the Intel part lists FP16 at 1,331.2 GFLOPS. Pixel rate favors Intel at 41.60 GPixel/s versus 21.22 GPixel/s, while texture rate slightly favors NVIDIA at 84.90 GTexel/s versus 83.20 GTexel/s.

Power consumption is a stark differentiator: 100 W for the NVIDIA part versus 15 W for the Intel part. Form factor also differs: the GTX 680M is an MXM Module with a bus interface of MXM-B (3.0), while the P750 is an IGP with a Ring Bus interface. Display outputs are listed as Portable Device Dependent for NVIDIA and Motherboard Dependent for Intel. The API support shows NVIDIA at DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, while Intel supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. Both have a production status of End-of-life. The NVIDIA part has a release date of June 3, 2012, while Intel’s release date is not listed.

FAQ

Q: Which GPU has higher raw compute performance?

A: The NVIDIA GeForce GTX 680M scores 9230 in Geekbench OpenCL, which is 40.8 percent higher than the Intel UHD Graphics P750’s score of 6554.

Q: How do the shading unit counts compare?

A: The GTX 680M has 1344 shading units, while the Intel P750 has 256, a difference of more than five times in favor of NVIDIA.

Q: What is the memory configuration difference?

A: The GTX 680M uses 4 GB of dedicated GDDR5 memory on a 256-bit bus with 115.2 GB/s bandwidth, whereas the P750 uses System Shared memory with System Dependent bandwidth.

Q: Which GPU consumes less power?

A: The Intel UHD Graphics P750 has a TDP of 15 W, compared to the GTX 680M’s 100 W, making the Intel part significantly more power-efficient.

Q: Does the Intel GPU have any performance advantage?

A: The Intel P750 has a higher pixel rate at 41.60 GPixel/s versus 21.22 GPixel/s for the GTX 680M, and it supports FP16 at 1,331.2 GFLOPS, which the NVIDIA part does not.

Q: How do the two GPUs rank against all others?

A: The GTX 680M is at the 39th percentile, and the P750 is at the 38th percentile, placing them nearly identically in overall standing despite the large OpenCL score gap.

The Verdict

The data makes the choice straightforward for compute-oriented users. The NVIDIA GeForce GTX 680M wins the only head-to-head benchmark with a 40.8 percent margin, and its 2.038 TFLOPS FP32 performance is more than three times the Intel part’s 665.6 GFLOPS. This GPU is for anyone who needs dedicated graphics memory, high shading unit count, and sustained throughput in OpenCL workloads. Its 100 W TDP and MXM form factor indicate it belongs in a laptop or mobile workstation where performance is the priority over battery life.

The Intel UHD Graphics P750 is the pick for efficiency and simplicity. At 15 W, it sips power, and its IGP design means no extra slots or connectors are needed. Its higher pixel rate and FP16 support are notable, but its 6554 OpenCL score is 40.8 percent behind, and its System Shared memory ties it to the host system’s performance. For users whose primary tasks are basic display output, media playback, or light productivity, the P750 is adequate. For anyone who needs to run GPU-accelerated compute tasks, the GTX 680M is the only logical choice based on the benchmark data.