GPU Comparison

The GeForce 920M and GeForce GT 640 are both end-of-life Kepler parts from NVIDIA, but they target very different platforms and eras. The GT 640 launched in June 2012 as a desktop card, while the 920M arrived in March 2015 for notebooks. Despite the three-year gap, their average benchmark scores are remarkably close: the 920M averages 3287, while the GT 640 averages 3210. This places the 920M 2.4% ahead of the GT 640 in overall average score, according to the nearestRivals data. Both sit at the 20th percentile of all GPUs, indicating they are entry-level performers in the modern landscape, though their individual strengths and weaknesses tell a more nuanced story.

Head-to-Head Benchmarks

The head-to-head data includes two compute benchmarks: Geekbench OpenCL and Geekbench Vulkan. In OpenCL, the GT 640 edges out the 920M by a razor-thin margin, scoring 3736 versus 3725, a delta of just -0.3% from the 920M’s perspective. This is effectively a statistical tie, meaning the two GPUs deliver nearly identical raw compute throughput in this API. The 920M’s FP32 rating of 732.7 GFLOPS is actually slightly higher than the GT 640’s 692.7 GFLOPS, which helps explain why the OpenCL gap is so small despite the GT 640’s wider memory bus.

The Vulkan test is a different story entirely. Here, the GT 640 wins decisively, scoring 3809 against the 920M’s 2849. That is a 25.2% advantage for the GT 640, a substantial margin that highlights a fundamental difference in how these two GPUs handle modern graphics APIs. The 920M’s Vulkan score is its weakest benchmark result, while the GT 640’s Vulkan score is its strongest, beating even its own OpenCL result. This suggests the GT 640’s driver support or hardware design is better suited to Vulkan’s lower-level overhead, whereas the 920M appears to struggle in that environment.

In the broader context of their nearest rivals, these scores place both GPUs in a tight cluster. The 920M’s closest competitor is the GeForce GT 730M, which averages 3316, just 0.9% higher. The Intel HD Graphics 530 sits at 3332, 1.4% above the 920M, while the Intel HD Graphics P4600 leads that group at 3389, 3% higher. The GT 640, meanwhile, is bracketed by the Intel Arc Pro B60 at 3182 (0.9% below) and the Quadro P1000 at 3163 (1.5% below). The fact that both GPUs are within a few percentage points of each other and their immediate rivals underscores how little separates them in aggregate performance, even though individual workloads can swing heavily in one direction or the other.

Architecture Differences

Both GPUs are built on TSMC’s 28 nm process, but they use different physical chips. The 920M uses the GK208B chip, a smaller die at 87 mm² with 1,020 million transistors, giving it a transistor density of 11.7 million per mm². The GT 640 uses the GK107 chip, which is larger at 118 mm² and packs 1,270 million transistors, resulting in a slightly lower density of 10.8 million per mm². The 920M is classified as Kepler 2.0 architecture, while the GT 640 is original Kepler, though both belong to NVIDIA’s GeForce 900M and GeForce 600 generations respectively.

The compute core counts are identical: both have 384 shading units and 32 texture mapping units. The key architectural divergence lies in the render output units and memory subsystem. The 920M has 8 ROPs, half the GT 640’s 16 ROPs. This difference is reflected in their pixel rates: the 920M achieves 7.632 GPixel/s, slightly ahead of the GT 640’s 7.216 GPixel/s, thanks to its higher clock speed. The 920M runs at a fixed 954 MHz base and boost, while the GT 640’s base and boost clocks are not listed in the data, but its lower pixel rate implies a lower effective clock.

Memory configuration is where the gap widens. Both have 2 GB of DDR3, but the 920M uses a 64-bit bus with a bandwidth of 14.40 GB/s, while the GT 640 uses a 128-bit bus with 28.51 GB/s, nearly double the bandwidth. The memory clocks differ as well: the 920M runs at 900 MHz (1800 Mbps effective), while the GT 640 runs at 891 MHz (1782 Mbps effective). The GT 640’s wider bus compensates for its marginally lower clock, delivering significantly more memory throughput, which is critical for texture-heavy workloads. This explains why the GT 640 wins the Vulkan benchmark so decisively; its memory subsystem is far more capable.

Power and physical design also diverge sharply. The 920M has a 33 W TDP and is an integrated graphics processor (IGP) with no power connectors, designed for portability. The GT 640 has a 65 W TDP, is a single-slot card, also has no power connectors, but requires a 250 W suggested PSU. The GT 640 is a desktop card measuring 145 mm (5.7 inches) in length, with display outputs of 1x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. The 920M has no fixed display outputs, as they depend on the portable device it is embedded in. The bus interface also differs: the 920M uses PCIe 3.0 x8, while the GT 640 uses PCIe 3.0 x16, giving the desktop card more bandwidth to the host system.

Where Each One Wins

The GT 640 is the clear winner in graphics API performance, particularly in Vulkan where it leads by 25.2%. This makes it the better choice for any application or game that leverages Vulkan’s modern, low-overhead rendering path. Its doubled memory bandwidth (28.51 GB/s versus 14.40 GB/s) and double the ROPs (16 versus 8) give it a structural advantage in fill-rate-bound scenarios, even though its raw FP32 is slightly lower. The GT 640 also has a higher texture rate at 28.86 GTexel/s versus the 920M’s 30.53 GTexel/s, but the 920M wins that specific metric due to its higher clock; the GT 640 still takes the overall compute wins in the head-to-head.

The 920M, despite losing both head-to-head benchmarks, has its own niches. Its OpenCL score of 3725 is within 0.3% of the GT 640, meaning for pure compute tasks that use OpenCL, the two are interchangeable. The 920M’s higher FP32 (732.7 GFLOPS) and higher pixel rate (7.632 GPixel/s) suggest it can hold its own in specific shader-bound workloads that don’t stress memory bandwidth. Additionally, its 33 W TDP makes it far more suitable for thin-and-light laptops, where the GT 640’s 65 W desktop requirement would be impractical. The 920M’s integration also means it has no need for external power connectors, and its PCIe 3.0 x8 interface, while narrower, is adequate for its memory bandwidth.

For users deciding between these two, the choice hinges on platform and API. If the workload is Vulkan-based, the GT 640 is overwhelmingly superior, with a benchmark delta of 25.2%. If the workload is OpenCL-based or the system is a portable device, the 920M is nearly equivalent in performance while consuming roughly half the power. The GT 640’s display outputs and single-slot form factor make it a drop-in desktop solution, whereas the 920M is fixed inside a laptop with no upgrade path. The GT 640 also has the advantage of a launch MSRP of 99 USD, which can be stated as its historical launch price, though the 920M’s pricing is not listed in the data.

FAQ

Q: Which GPU is faster in Vulkan performance?

A: The GeForce GT 640 is significantly faster, scoring 3809 in Geekbench Vulkan versus the 920M’s 2849, a 25.2% advantage for the GT 640.

Q: How do their OpenCL scores compare?

A: They are nearly identical. The GT 640 scores 3736, while the 920M scores 3725, a difference of only 0.3% in favor of the GT 640, making them effectively tied in this API.

Q: Do both GPUs have the same number of shading units?

A: Yes, both have 384 shading units and 32 texture mapping units. However, the GT 640 has 16 ROPs while the 920M has only 8 ROPs.

Q: What is the memory bandwidth difference?

A: The GT 640 has 28.51 GB/s of bandwidth over a 128-bit bus, while the 920M has 14.40 GB/s over a 64-bit bus. The GT 640’s bandwidth is nearly double.

Q: Are both GPUs the same architecture?

A: No. The 920M uses Kepler 2.0 with the GK208B chip, while the GT 640 uses the original Kepler with the GK107 chip. Both are on a 28 nm process from TSMC.

Q: What is the power consumption difference?

A: The 920M has a 33 W TDP and is an integrated part, while the GT 640 has a 65 W TDP and is a single-slot desktop card requiring a 250 W suggested PSU.

Specification Differences

| Specification | GeForce 920M | GeForce GT 640 |

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

| Chip | GK208B | GK107 |

| Architecture | Kepler 2.0 | Kepler |

| Generation | GeForce 900M | GeForce 600 |

| Transistors | 1,020 million | 1,270 million |

| Die Size | 87 mm² | 118 mm² |

| Transistor Density | 11.7M / mm² | 10.8M / mm² |

| Base Clock | 954 MHz | Not listed |

| Boost Clock | 954 MHz | Not listed |

| Memory Clock | 900 MHz, 1800 Mbps effective | 891 MHz, 1782 Mbps effective |

| Memory Bus Width | 64 bit | 128 bit |

| Memory Bandwidth | 14.40 GB/s | 28.51 GB/s |

| ROPs | 8 | 16 |

| Pixel Rate | 7.632 GPixel/s | 7.216 GPixel/s |

| FP32 | 732.7 GFLOPS | 692.7 GFLOPS |

| TDP | 33 W | 65 W |

| Slot Width | IGP | Single-slot |

| Suggested PSU | Not listed | 250 W |

| Bus Interface | PCIe 3.0 x8 | PCIe 3.0 x16 |

| Display Outputs | Portable Device Dependent | 1x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 |

| Length | Not listed | 145 mm, 5.7 inches |

| Release Date | 2015-03-12 | 2012-06-04 |

| Predecessor | GeForce 800M | GeForce 500 |

| Successor | GeForce 10 Mobile | GeForce 700 |

| Launch MSRP | Not listed | 99 USD |

| Geekbench OpenCL | 3725 | 3736 |

| Geekbench Vulkan | 2849 | 3809 |

| Avg Benchmark Score | 3287 | 3210 |