GPU Comparison

The NVIDIA GeForce 920M and NVIDIA GeForce GTX 860M represent two very different tiers of mobile graphics from the same vendor, separated by a full generation of positioning and silicon. The data shows a clear hierarchy, but the details of that gap, where it is largest and what it means for real workloads, are worth examining closely.

Head-to-Head Benchmarks

The benchmark results are unambiguous. Across the two shared tests, the GTX 860M dominates the 920M completely. In Geekbench OpenCL, the GTX 860M scores 10,461 against the 920M’s 3,725. That is a delta of -64.4% from the perspective of the 920M, meaning the 920M trails by more than two-thirds. The margin only widens in Geekbench Vulkan: the GTX 860M posts 9,648 while the 920M manages 2,849, a -70.5% delta. The 920M wins zero head-to-head benchmarks; the GTX 860M wins both.

These are not marginal differences. The 920M’s best result in either test is less than 40% of the GTX 860M’s corresponding score. For context, the 920M’s average benchmark score of 3,287 places it in the 20th percentile of all GPUs, while the GTX 860M’s average of 2,967 puts it at the 19th percentile. Despite the raw score gap between the two parts, their percentile rankings are nearly identical, a quirk of the broader distribution that does not reflect the head-to-head reality. The GTX 860M’s nearest rivals include the GeForce GTX 750 Ti (deltaPct 0.5) and the GeForce RTX 4060 Ti 8 GB (deltaPct 1.9), which shows it sits in a higher-performance cluster than the 920M, whose nearest rivals are integrated and older discrete parts like the Intel HD Graphics 530 (deltaPct -1.4) and the GeForce GT 640 (deltaPct 2.4).

Architecture Differences

The architectural gap between these two GPUs is as wide as the benchmark delta. The 920M uses the GK208B chip, built on Kepler 2.0 architecture, while the GTX 860M uses the GK104 chip on the original Kepler architecture. Both are manufactured by TSMC on a 28 nm process, but that is where the similarities end. The GK208B packs 1,020 million transistors on an 87 mm² die, giving a transistor density of 11.7M per mm². The GK104 is a much larger design: 3,540 million transistors on 294 mm², with a density of 12.0M per mm². The GTX 860M has more than three times the transistor count and a die over three times larger.

The execution resources follow the same pattern. The 920M has 384 shading units, 32 texture mapping units, and 8 ROPs. The GTX 860M has 1,152 shading units, 96 TMUs, and 16 ROPs, exactly three times the shaders and TMUs, and double the ROPs. Clock speeds tell a different story: the 920M runs at a base and boost of 954 MHz, while the GTX 860M has a base of 797 MHz and a boost of 915 MHz. The 920M is clocked higher, but that advantage is dwarfed by the GTX 860M’s massive resource advantage. The resulting throughput figures are lopsided: the 920M delivers 732.7 GFLOPS of FP32, 30.53 GTexel/s of texture rate, and 7.632 GPixel/s of pixel rate. The GTX 860M delivers 2.108 TFLOPS, 87.84 GTexel/s, and 21.96 GPixel/s, roughly 2.9 times the compute, 2.9 times the texture fill, and 2.9 times the pixel throughput.

Memory is another major differentiator. The 920M has 2 GB of DDR3 on a 64-bit bus, yielding 14.40 GB/s of bandwidth. The GTX 860M also has 2 GB, but it is GDDR5 on a 128-bit bus, producing 80.00 GB/s, over 5.5 times the memory bandwidth. The GTX 860M’s memory clock is 1250 MHz with 5 Gbps effective, versus the 920M’s 900 MHz with 1800 Mbps effective. The GTX 860M also has a wider PCIe interface: PCIe 3.0 x16 versus PCIe 3.0 x8 on the 920M. Both support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, so API compatibility is equal. Power draw differs significantly: the 920M is rated at 33 W TDP, while the GTX 860M draws 75 W. Both are listed as IGP slot width with no power connectors.

Where Each One Wins

The data does not leave much room for the 920M to claim a win in raw performance. In both Geekbench OpenCL and Vulkan, the GTX 860M is ahead by 64.4% and 70.5% respectively. The 920M’s only advantages are efficiency and clock speed. It has a lower TDP of 33 W versus 75 W, and its base clock of 954 MHz is 157 MHz higher than the GTX 860M’s base, and 39 MHz higher than its boost. That higher clock rate does not translate into any benchmark win, but it does suggest the 920M can sustain a higher frequency under load without the power budget of the larger chip.

The GTX 860M’s victories are comprehensive. It wins on memory bandwidth by a factor of 5.5, on FP32 compute by a factor of 2.9, on texture rate by a factor of 2.9, and on pixel rate by a factor of 2.9. It also has three times the shading units and TMUs, and double the ROPs. The GTX 860M’s additional benchmark tests, PassMark DirectX 9, 10, 11, and 12, plus G2D, G3D, and GPU compute, show a range of scores from 13 to 3081, but there are no comparable scores for the 920M, so those cannot be used for a direct comparison. What the head-to-head data shows is that the GTX 860M is the clear choice for any workload that stresses the GPU, whether compute-heavy OpenCL or graphics-oriented Vulkan.

The Verdict

The verdict is straightforward. The NVIDIA GeForce GTX 860M is the superior GPU in every measurable head-to-head metric. It delivers 10,461 in Geekbench OpenCL versus 3,725, and 9,648 in Geekbench Vulkan versus 2,849, deltas of -64.4% and -70.5% from the 920M’s perspective. The GTX 860M’s larger GK104 chip, with 3,540 million transistors, 1,152 shading units, and 80.00 GB/s of GDDR5 bandwidth, puts it in a completely different performance class. It is also the more recent design in terms of release, launching on 2014-03-09, versus the 920M’s 2015-03-12, though both are now end-of-life.

The 920M is not without a rationale. Its 33 W TDP and 954 MHz clock suggest it is built for power-constrained notebooks where the GTX 860M’s 75 W draw would be unacceptable. Its average benchmark score of 3,287 is higher than the GTX 860M’s 2,967, and its percentile rank of 20 is marginally above the GTX 860M’s 19. But those aggregate numbers are misleading in a direct comparison. The 920M’s nearest rivals are the GeForce GT 730M, Intel HD Graphics 530, GeForce GT 640, and Intel HD Graphics P4600, parts that are either older or integrated. The GTX 860M’s nearest rivals are the GeForce GTX 750 Ti, GeForce 820A, Quadro P600, and GeForce RTX 4060 Ti 8 GB, which are all discrete GPUs with meaningful performance. Anyone choosing between these two should pick the GTX 860M unless power draw is the absolute constraint.

FAQ

Q: Which GPU has a higher Geekbench OpenCL score?

A: The NVIDIA GeForce GTX 860M scores 10,461 in Geekbench OpenCL, while the NVIDIA GeForce 920M scores 3,725. The GTX 860M wins by a -64.4% delta from the 920M’s perspective.

Q: What is the difference in memory bandwidth?

A: The GTX 860M has 80.00 GB/s of bandwidth from 2 GB of GDDR5 on a 128-bit bus. The 920M has 14.40 GB/s from 2 GB of DDR3 on a 64-bit bus. The GTX 860M has over 5.5 times the bandwidth.

Q: Are both GPUs based on the same architecture?

A: No. The 920M uses Kepler 2.0 on the GK208B chip, while the GTX 860M uses the original Kepler on the GK104 chip. Both are 28 nm parts from TSMC, but the GK104 has 3,540 million transistors versus 1,020 million on the GK208B.

Q: How do their TDPs compare?

A: The 920M has a TDP of 33 W, while the GTX 860M has a TDP of 75 W. Both are listed as IGP slot width with no power connectors.

Q: Which GPU has more shading units?

A: The GTX 860M has 1,152 shading units, exactly three times the 384 shading units on the 920M. The GTX 860M also has 96 TMUs versus 32, and 16 ROPs versus 8.

Q: What is the release date difference?

A: The GTX 860M was released on 2014-03-09, and the 920M was released on 2015-03-12. Both are end-of-life, with the 920M succeeding the GeForce 800M and the GTX 860M succeeding the GeForce 700M.