GPU Comparison

# Head-to-Head Benchmarks

The only shared benchmark between these two mobile GPUs is Geekbench OpenCL, and the result is remarkably close. The NVIDIA GeForce 920M scores 3725, while the NVIDIA GeForce 825M scores 3694. That is a delta of -0.8% favoring the 920M, essentially a statistical tie in real-world terms. The 920M wins this head-to-head, but the margin is so narrow that it would be imperceptible in any actual workload.

Looking at the nearest rivals for each card provides more context. The 825M's closest competitor is the NVIDIA GeForce GT 740M, which scores 3717, the 825M trails by just 0.6%. The 825M also sits 0.6% behind the Quadro 3000M (3718) and 1.2% behind the GeForce GT 635M (3740). Conversely, the 825M leads the AMD Radeon HD 6770 by 1.2% (3649). These delta values reinforce that the 825M sits in a tightly packed performance cluster where a few points separate entire product tiers.

The 920M's rival set tells a similar story. It leads the NVIDIA GeForce GT 640 by 2.4% (3210) and the NVIDIA GeForce GT 730M by 0.9% (3316). However, it trails the Intel HD Graphics 530 by 1.4% (3332) and the Intel HD Graphics P4600 by 3% (3389). This is notable: the 920M's nearest rivals include integrated graphics solutions, indicating that its performance envelope overlaps with high-end iGPUs from the same era.

The percentile rankings confirm the modest positioning. The 825M sits at the 22nd percentile of all GPUs, while the 920M sits at the 20th percentile. Despite the 920M winning the head-to-head, its average benchmark score across both of its tested workloads (OpenCL and Vulkan) is 3287, which is lower than its 3725 OpenCL score alone. This drags its overall ranking below the 825M's, which has only one benchmark result. The Vulkan score of 2849 for the 920M pulls its aggregate down considerably.

In the single direct comparison, the 920M is faster. But the data also shows that both cards occupy nearly identical performance bands, with the 825M arguably having a slight edge in consistency when considering the full benchmark suite.

Architecture Differences

Both GPUs are built on the same fundamental architecture: Kepler 2.0. The silicon is nearly identical, with both using a 28 nm process at TSMC. Transistor counts match at 1,020 million, and die size is the same at 87 mm². Transistor density is consequently identical at 11.7M per mm². The chip designations differ, the 825M uses GK208, while the 920M uses GK208B, but the core configuration is unchanged.

The shading unit count is exactly the same at 384, as are the texture mapping units at 32 and the render output units at 8. Neither card features ray tracing cores or tensor cores, which is expected for this generation. The memory subsystem is also identical in type and bus width: both use DDR3 on a 64-bit interface, yielding 14.40 GB/s of bandwidth. Memory capacity is a differentiator, though. The 825M ships with 1024 MB, while the 920M doubles that to 2 GB. This capacity difference does not affect raw compute rates, but it can impact texture-heavy scenes and larger frame buffers.

Clock speeds show the most meaningful architectural divergence. The 825M has a base clock of 850 MHz and a boost clock of 941 MHz. The 920M, by contrast, runs at a fixed 954 MHz for both base and boost, there is no boost headroom because it is already at its maximum. This higher sustained clock gives the 920M a slight throughput advantage. Pixel rate on the 920M is 7.632 GPixel/s versus 7.528 GPixel/s on the 825M. Texture rates are similarly close: 30.53 GTexel/s for the 920M against 30.11 GTexel/s for the 825M. FP32 compute reaches 732.7 GFLOPS on the 920M, compared to 722.7 GFLOPS on the 825M.

Both cards share a 33 W TDP, IGP slot width, no power connectors, and a PCIe 3.0 x8 bus interface. Display outputs are listed as "Portable Device Dependent" for both, reflecting their mobile-oriented design. API support is identical across the board: DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. Production status for both is end-of-life, with the 825M released in January 2014 and the 920M in March 2015. The 825M's predecessor is the GeForce 700M series and its successor is the GeForce 900M series; the 920M's predecessor is the GeForce 800M series and its successor is the GeForce 10 Mobile lineup.

The only real architectural difference is the GK208B revision on the 920M, which enables the higher fixed clock. Everything else, from the memory bus to the core counts, is functionally identical.

FAQ

Q: Which GPU is faster in OpenCL?

A: The GeForce 920M scores 3725 in Geekbench OpenCL, beating the 825M's 3694 by 0.8%. This is the only direct head-to-head benchmark available.

Q: Do these GPUs have the same memory bandwidth?

A: Yes. Both use DDR3 on a 64-bit bus, resulting in 14.40 GB/s of bandwidth. The 920M does offer twice the memory capacity (2 GB versus 1 GB).

Q: Are the core counts identical?

A: Yes. Both have 384 shading units, 32 TMUs, and 8 ROPs. The 920M achieves its higher performance through a higher clock speed, not additional cores.

Q: How do these cards compare to their nearest rivals?

A: The 825M is 0.6% behind the GT 740M and Quadro 3000M, and 1.2% behind the GT 635M, but 1.2% ahead of the Radeon HD 6770. The 920M leads the GT 640 by 2.4% and the GT 730M by 0.9%, yet trails the Intel HD Graphics 530 by 1.4% and the HD Graphics P4600 by 3%.

Q: What is the performance percentile of each card?

A: The 825M sits at the 22nd percentile of all GPUs, while the 920M sits at the 20th percentile. These are very close positions in the low-mid range of the performance distribution.

Q: Do these cards support modern graphics APIs?

A: Both support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. Neither has ray tracing or tensor cores.

The Verdict

The benchmark data shows a clear but narrow winner: the GeForce 920M edges out the 825M in the only shared test. The 0.8% OpenCL advantage is real but nearly negligible. The 920M's higher fixed clock of 954 MHz versus the 825M's 850 MHz base and 941 MHz boost explains the margin. The 920M also offers double the memory capacity at 2 GB versus 1 GB, which could matter in memory-constrained scenarios even though bandwidth is unchanged.

However, the 920M's overall standing is slightly worse due to its Vulkan score. With an average benchmark score of 3287, the 920M ranks at the 20th percentile, while the 825M's single OpenCL score of 3694 places it at the 22nd percentile. The 920M's Vulkan result of 2849 drags its aggregate down, suggesting that its OpenCL strength does not translate to other workloads.

For users choosing between these two, the data points toward the 920M for raw OpenCL performance and memory capacity. The 825M, however, offers a more consistent profile with no below-3000 benchmark result. Both are end-of-life mobile parts with identical 33 W TDPs, so power consumption is not a differentiator.

The nearest rival data reinforces that neither card is positioned strongly in the broader GPU landscape. The 825M's closest competitors are all within 1.2% of its score, and the 920M's rivals span from a 3% deficit to a 2.4% lead. These are entry-level parts whose performance is tightly clustered with older discrete GPUs and contemporary integrated graphics.

The verdict: pick the 920M if you prioritize the slightly higher compute throughput and doubled memory capacity. Pick the 825M if you value consistent benchmark performance across workloads, as its average score is higher despite losing the head-to-head. In practical terms, the difference is unlikely to be noticeable in most applications.

Specification Differences

| Specification | NVIDIA GeForce 825M | NVIDIA GeForce 920M |

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

| Chip | GK208 | GK208B |

| Generation | GeForce 800M | GeForce 900M |

| Base Clock | 850 MHz | 954 MHz |

| Boost Clock | 941 MHz | 954 MHz |

| Memory Size | 1024 MB | 2 GB |

| Pixel Rate | 7.528 GPixel/s | 7.632 GPixel/s |

| Texture Rate | 30.11 GTexel/s | 30.53 GTexel/s |

| FP32 Compute | 722.7 GFLOPS | 732.7 GFLOPS |

| Release Date | 2014-01-26 | 2015-03-12 |

| Predecessor | GeForce 700M | GeForce 800M |

| Successor | GeForce 900M | GeForce 10 Mobile |

| OpenCL Benchmark | 3694 | 3725 |

| Vulkan Benchmark | N/A | 2849 |

| Average Benchmark Score | 3694 | 3287 |

| Percentile (All GPUs) | 22 | 20 |

All other specifications, including process node (28 nm), transistors (1,020 million), die size (87 mm²), shading units (384), TMUs (32), ROPs (8), memory type (DDR3), bus width (64 bit), bandwidth (14.40 GB/s), TDP (33 W), slot width (IGP), power connectors (None), bus interface (PCIe 3.0 x8), and API support (DirectX 12 (11_0), OpenGL 4.6, Vulkan 1.2.175), are identical between the two cards.