NVIDIA GeForce GTX 660 vs NVIDIA GeForce GTX 870M Comparison

The Verdict

The benchmark data consistently favors the NVIDIA GeForce GTX 870M across both recorded tests. The 870M wins 2 out of 2 head-to-head comparisons, with its largest advantage appearing in the Geekbench Metal workload, where it leads by 69.3% (7288 versus 4305). In the OpenCL test, the margin narrows to 11.3% (12630 versus 11347), but the result remains decisive. The 870M also holds a higher average benchmark score of 9959 compared to 9022 for the GTX 660, which corresponds to a 48th percentile ranking versus 45th in the database. The GTX 660 shows no recorded test where it outperforms the 870M, so the verdict is straightforward: the 870M is the stronger choice for users prioritizing compute workloads and general GPU performance. The GTX 660 remains relevant only for those who specifically need a desktop card with a PCIe 3.0 x16 interface, a dual-slot cooler, or a 300 W suggested PSU, but the data does not support choosing it for raw benchmark performance.

Architecture Differences

Both GPUs are built on NVIDIA's Kepler architecture and both use a 28 nm process node from TSMC, but they diverge significantly in their underlying chips. The GeForce GTX 870M uses the GK104 chip, while the GeForce GTX 660 uses the GK106. This difference in die selection has direct consequences for resource counts. The 870M integrates 3,540 million transistors on a 294 mm² die, resulting in a transistor density of 12.0M per mm². The GTX 660, by contrast, packs 2,540 million transistors onto a 221 mm² die, with a density of 11.5M per mm². The 870M therefore carries roughly 39.4% more transistors and a 33.0% larger die area.

The shading resources also differ meaningfully. The 870M features 1,344 shading units and 112 texture mapping units, while the GTX 660 has 960 shading units and 80 texture mapping units. Both cards share 24 ROPs, so pixel output capability is identical in terms of ROP count. However, the higher clock speeds of the GTX 660 partially offset its lower resource counts. The GTX 660 runs at a base clock of 980 MHz and a boost clock of 1032 MHz, whereas the 870M operates at 941 MHz base and 967 MHz boost. Despite the clock disadvantage, the 870M's larger shader count yields higher theoretical throughput: 2.599 TFLOPS FP32 versus 1.981 TFLOPS for the GTX 660. The texture rate follows the same pattern, with the 870M at 108.3 GTexel/s versus 82.56 GTexel/s for the GTX 660. The pixel rate favors the 870M as well, at 27.08 GPixel/s versus 20.64 GPixel/s, driven by its higher clock and identical ROP count.

Memory architecture differs in capacity and bandwidth. The 870M comes with 3 GB of GDDR5, while the GTX 660 has 2 GB. Both use a 192-bit bus, but the GTX 660 posts a higher memory clock of 1502 MHz (6 Gbps effective) versus 1250 MHz (5 Gbps effective) on the 870M. Consequently, the GTX 660 achieves 144.2 GB/s of bandwidth compared to 120.0 GB/s for the 870M. This is the one area where the GTX 660 holds a clear advantage in the recorded specifications.

The cards also differ in their power and physical design. The 870M is rated at 100 W TDP and ships as an MXM Module with no power connectors, reflecting its mobile-orientation. The GTX 660 draws 140 W, uses a dual-slot cooler, requires a single 6-pin power connector, and lists a 300 W suggested PSU. The bus interface also differs: the 870M uses MXM-B (3.0), while the GTX 660 uses PCIe 3.0 x16. Display outputs are portable-device dependent on the 870M, whereas the GTX 660 provides 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. Both cards support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175.

Head-to-Head Benchmarks

The most striking result in the head-to-head data is the Geekbench Metal score. The 870M reaches 7288, while the GTX 660 manages only 4305. That is a 69.3% advantage for the 870M, the largest delta recorded between the two cards. This substantial margin suggests that the 870M's additional shading units and higher transistor count translate effectively into Metal compute performance, despite its lower memory bandwidth.

The Geekbench OpenCL test shows a tighter contest. The 870M scores 12630 against 11347 for the GTX 660, a lead of 11.3%. This narrower gap indicates that the GTX 660's higher memory bandwidth (144.2 GB/s versus 120.0 GB/s) and higher boost clock (1032 MHz versus 967 MHz) help it close some of the distance in OpenCL workloads. However, the 870M still comes out ahead, driven by its 1,344 shading units versus 960, its 112 TMUs versus 80, and its higher FP32 throughput of 2.599 TFLOPS versus 1.981 TFLOPS.

The average benchmark scores in the database reinforce the head-to-head results. The 870M averages 9959 across all recorded tests, placing it at the 48th percentile among all GPUs. The GTX 660 averages 9022, placing it at the 45th percentile. The relative positioning is consistent with the head-to-head deltas: the 870M sits slightly above its rival in the overall distribution.

It is worth remembering the GTX 660 has a recorded Geekbench Vulkan score of 11415, a test that the 870M does not have in the database. This means the GTX 660 can be evaluated in Vulkan, but no direct comparison is possible from the recorded data. For the two tests where both cards are measured, the 870M wins both.

Specification Differences

The following fields differ between the two cards, based solely on the recorded data:

• Chip: GK104 (870M) versus GK106 (GTX 660)
• Generation: GeForce 800M versus GeForce 600
• Transistors: 3,540 million versus 2,540 million
• Die size: 294 mm² versus 221 mm²
• Transistor density: 12.0M / mm² versus 11.5M / mm²
• Base clock: 941 MHz versus 980 MHz
• Boost clock: 967 MHz versus 1032 MHz
• Memory clock: 1250 MHz (5 Gbps effective) versus 1502 MHz (6 Gbps effective)
• Memory size: 3 GB versus 2 GB
• Memory bandwidth: 120.0 GB/s versus 144.2 GB/s
• Shading units: 1344 versus 960
• TMUs: 112 versus 80
• Pixel rate: 27.08 GPixel/s versus 20.64 GPixel/s
• Texture rate: 108.3 GTexel/s versus 82.56 GTexel/s
• FP32: 2.599 TFLOPS versus 1.981 TFLOPS
• TDP: 100 W versus 140 W
• Slot width: MXM Module versus Dual-slot
• Power connectors: None versus 1x 6-pin
• Suggested PSU: not listed versus 300 W
• Bus interface: MXM-B (3.0) versus PCIe 3.0 x16
• Display outputs: Portable Device Dependent versus 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2
• Dimensions: no length recorded versus 241 mm (9.5 inches)
• Release date: 2014-03-11 versus 2012-09-05
• Predecessor: GeForce 700M versus GeForce 500
• Successor: GeForce 900M versus GeForce 700
• Launch MSRP: not listed versus 229 USD

Fields that are identical include architecture (Kepler), process node (28 nm), foundry (TSMC), memory type (GDDR5), bus width (192 bit), ROPs (24), API support (DirectX 12 (11_0), OpenGL 4.6, Vulkan 1.2.175), and production status (End-of-life).

FAQ

Q: Which GPU wins the Geekbench Metal test?

A: The NVIDIA GeForce GTX 870M wins with a score of 7288, compared to 4305 for the GTX 660, a 69.3% advantage.

Q: What is the difference in average benchmark scores?

A: The 870M averages 9959, while the GTX 660 averages 9022. The 870M ranks at the 48th percentile of all GPUs, and the GTX 660 ranks at the 45th percentile.

Q: Does the GTX 660 have any advantage in memory bandwidth?

A: Yes. The GTX 660 achieves 144.2 GB/s of bandwidth due to its higher memory clock of 1502 MHz (6 Gbps effective). The 870M provides 120.0 GB/s from a 1250 MHz (5 Gbps effective) clock.

Q: How do the shading unit counts differ?

A: The 870M has 1,344 shading units and 112 TMUs. The GTX 660 has 960 shading units and 80 TMUs. Both have 24 ROPs.

Q: What is the TDP of each card?

A: The 870M is rated at 100 W, while the GTX 660 is rated at 140 W. The GTX 660 also requires a 6-pin power connector and a 300 W suggested PSU, while the 870M has no power connectors.

Q: Which card has a higher FP32 throughput?

A: The 870M reaches 2.599 TFLOPS, while the GTX 660 reaches 1.981 TFLOPS. This reflects the 870M's larger shader count despite its lower clock speeds.