GPU Comparison

The NVIDIA GeForce 920M and the NVIDIA GeForce GT 635M are both end-of-life mobile graphics solutions from different architectural eras, with the 920M representing the newer Kepler 2.0 design and the GT 635M hailing from the older Fermi generation. Based on the available benchmark data, these two GPUs are remarkably close in overall performance, with the GT 635M holding a slight edge in the single available head-to-head comparison.

Head-to-Head Benchmarks

The only direct benchmark comparison in the data is the Geekbench OpenCL test, where the two GPUs produce nearly identical results. The NVIDIA GeForce GT 635M scores 3,740 points, while the NVIDIA GeForce 920M trails with 3,725 points. This difference of just 15 points translates to a delta of -0.4% for the 920M, making this effectively a statistical tie in raw compute performance. The GT 635M wins this head-to-head matchup, but by a margin so thin that it would be imperceptible in real-world usage.

Looking at average benchmark scores across all tests, the GT 635M maintains its slight lead. The GT 635M averages 3,740 points, placing it in the 22nd percentile of all GPUs, while the 920M averages 3,287 points, placing it in the 20th percentile. This 453-point gap in average scores is more pronounced than the single OpenCL test suggests, though the 920M also has a Geekbench Vulkan score of 2,849 points that the GT 635M cannot match, as the older Fermi architecture lacks Vulkan support entirely.

The GT 635M's nearest rivals provide context for its positioning. It sits just 0.6% ahead of the NVIDIA Quadro 3000M (3,718 points) and the NVIDIA GeForce GT 740M (3,717 points), while trailing the Intel UHD Graphics 710 (3,792 points) by -1.4%. The 920M, meanwhile, is sandwiched between the NVIDIA GeForce GT 640 (3,210 points) and the Intel HD Graphics 530 (3,332 points), sitting 2.4% above the former and -1.4% below the latter. These rival comparisons show both GPUs occupying a similar performance tier, with neither able to establish a decisive advantage.

Where Each One Wins

The GT 635M wins the only direct benchmark comparison, but its advantage is minimal. In the Geekbench OpenCL test, it outperforms the 920M by just 0.4%, a margin that falls well within typical run-to-run variance. The GT 635M also holds a higher average benchmark score (3,740 vs 3,287) and a higher overall percentile ranking (22nd vs 20th), suggesting slightly better consistency across multiple test scenarios.

The 920M's clear victory comes in API support. It offers Vulkan 1.2.175 support and achieves a Geekbench Vulkan score of 2,849 points, while the GT 635M has no Vulkan capability listed. This gives the 920M a distinct advantage in any application or game that leverages the Vulkan API, potentially offsetting its slight deficit in raw OpenCL compute. For users running modern titles that utilize Vulkan, the 920M would likely deliver a smoother experience despite its lower average benchmark score.

The 920M also wins on efficiency metrics. It has a lower TDP of 33 W compared to the GT 635M's 35 W, and it achieves this while using a smaller die (87 mm² vs 116 mm²) with more transistors (1,020 million vs 585 million). This suggests the Kepler 2.0 architecture is substantially more power-efficient than the older Fermi design, delivering comparable performance with less power draw and better transistor density.

Architecture Differences

The architectural divide between these two GPUs is significant. The 920M uses the GK208B chip built on Kepler 2.0 architecture, manufactured on a 28 nm process at TSMC. The GT 635M uses the GF108 chip based on the older Fermi architecture, manufactured on a 40 nm process, also at TSMC. This process node difference explains much of the efficiency gap, with the 28 nm node allowing for nearly double the transistor density (11.7M per mm² vs 5.0M per mm²).

The 920M packs 1,020 million transistors into an 87 mm² die, while the GT 635M fits just 585 million transistors onto a larger 116 mm² die. Despite having fewer transistors, the GT 635M's Fermi architecture is less efficient, which is why it consumes more power (35 W vs 33 W) despite delivering only marginally better performance.

Core configurations differ dramatically. The 920M features 384 shading units, 32 texture mapping units (TMUs), and 8 raster operations pipelines (ROPs). The GT 635M has just 96 shading units, 16 TMUs, and 4 ROPs. This 4x advantage in shading units and 2x advantage in TMUs and ROPs means the 920M has substantially more raw compute resources, yet the GT 635M still manages to match it in OpenCL performance, a testament to the architectural efficiency improvements in Kepler 2.0.

Memory subsystems also diverge. Both GPUs use 2 GB of DDR3 memory, but the GT 635M has a 128-bit memory bus with 28.80 GB/s bandwidth, while the 920M is constrained to a 64-bit bus delivering just 14.40 GB/s. This halved memory bandwidth could bottleneck the 920M in bandwidth-sensitive tasks, partially explaining why its larger compute core doesn't translate into a performance lead. Both run memory at 900 MHz with 1800 Mbps effective speed.

The 920M's rated pixel rate is 7.632 GPixel/s and texture rate is 30.53 GTexel/s, far exceeding the GT 635M's 1.900 GPixel/s and 7.600 GTexel/s. The 920M also delivers 732.7 GFLOPS of FP32 compute versus 182.4 GFLOPS for the GT 635M. These theoretical peak rates suggest the 920M should dominate, yet real-world OpenCL scores show parity, indicating the GT 635M's wider memory bus helps it extract more practical performance from its smaller compute core.

Interface differences include PCIe 3.0 x8 on the 920M versus PCIe 2.0 x16 on the GT 635M. Both support DirectX 12 (11_0) and OpenGL 4.6, but only the 920M lists Vulkan support (version 1.2.175). The GT 635M has no Vulkan capability, a notable limitation for modern software compatibility.

FAQ

Q: Which GPU is faster in OpenCL benchmarks?

A: The NVIDIA GeForce GT 635M edges out the 920M with a Geekbench OpenCL score of 3,740 versus 3,725, a delta of just -0.4% for the 920M. This makes the GT 635M the winner of the only direct head-to-head benchmark available.

Q: Does the 920M support Vulkan?

A: Yes, the NVIDIA GeForce 920M supports Vulkan 1.2.175 and achieves a Geekbench Vulkan score of 2,849. The GT 635M has no Vulkan support listed in its specifications, making the 920M the clear choice for Vulkan-based applications.

Q: How do their power requirements compare?

A: The 920M has a TDP of 33 W, while the GT 635M is rated at 35 W. Both are integrated-class GPUs (IGP) with no power connectors and portable-device-dependent display outputs, making them suitable for laptops.

Q: What are the memory bandwidth differences?

A: The GT 635M has a 128-bit memory bus delivering 28.80 GB/s, while the 920M uses a 64-bit bus with only 14.40 GB/s. Both have 2 GB of DDR3 memory running at 900 MHz (1800 Mbps effective).

Q: Which GPU has more shading units?

A: The 920M has 384 shading units, 32 TMUs, and 8 ROPs, compared to the GT 635M's 96 shading units, 16 TMUs, and 4 ROPs. Despite this 4x advantage, the GT 635M still matches the 920M in OpenCL performance.

Q: How do their overall benchmark percentiles compare?

A: The GT 635M sits in the 22nd percentile of all GPUs with an average score of 3,740, while the 920M sits in the 20th percentile with an average score of 3,287. This places both in the low-to-mid range of GPU performance.

The Verdict

The data presents a surprisingly close matchup between two GPUs from different generations. The GT 635M wins the only direct benchmark comparison, achieving 3,740 points in Geekbench OpenCL versus the 920M's 3,725 points, and holds a higher average benchmark score and percentile ranking. For users prioritizing raw compute performance in OpenCL-based workloads, the GT 635M is the marginally better choice, though the 0.4% delta means the difference is negligible in practice.

The 920M's case rests on features beyond raw compute. Its Vulkan 1.2.175 support, which the GT 635M entirely lacks, makes it the more future-proof option for modern gaming and applications that leverage this API. Its lower TDP of 33 W versus 35 W, combined with a more modern 28 nm process and higher transistor density, indicates better power efficiency. The 920M also offers theoretical performance advantages in pixel rate (7.632 GPixel/s), texture rate (30.53 GTexel/s), and FP32 compute (732.7 GFLOPS), suggesting it has more headroom that could be unlocked in optimized scenarios.

The GT 635M counters with double the memory bandwidth (28.80 GB/s vs 14.40 GB/s), which likely explains why its smaller compute core keeps pace in real-world tests. For bandwidth-sensitive workloads, the GT 635M's 128-bit memory bus provides a tangible advantage that the 920M cannot overcome without a wider interface.

Ultimately, the GT 635M wins on measured performance, but the 920M wins on modern features and efficiency. Users who prioritize Vulkan compatibility and lower power draw should select the 920M, while those who value the slight edge in OpenCL compute and higher memory bandwidth should opt for the GT 635M. Given the near-identical real-world performance, the 920M's Vulkan support and architectural modernity make it the more compelling choice for longevity, despite losing the direct head-to-head benchmark.