NVIDIA GeForce GTX 670M vs NVIDIA GeForce GTX 765M Comparison

FAQ

Q: Which GPU has the higher average benchmark score in the database?

A: The NVIDIA GeForce GTX 765M records an average benchmark score of 5501, while the NVIDIA GeForce GTX 670M records 6513. However, the GTX 670M holds a higher percentile rank at 38 compared to the GTX 765M's 32.

Q: How do the two GPUs compare in the Geekbench OpenCL test?

A: The GTX 765M wins the head-to-head OpenCL benchmark with a score of 7176 versus the GTX 670M's 6513, a delta of -9.2% from the perspective of the GTX 670M.

Q: What are the architectural generations of these two mobile GPUs?

A: The GTX 670M belongs to the GeForce 600M generation and uses the Fermi 2.0 architecture with the GF114 chip. The GTX 765M belongs to the GeForce 700M generation and uses the Kepler architecture with the GK106 chip.

Q: Which GPU has the higher transistor count and what are their process nodes?

A: The GTX 765M has 2,540 million transistors on a 28 nm process, while the GTX 670M has 1,950 million transistors on a 40 nm process. The GTX 765M also achieves a higher transistor density of 11.5M per mm² versus 5.9M per mm² for the GTX 670M.

Q: Do both GPUs support the same DirectX and OpenGL versions?

A: Yes, both support DirectX 12 (11_0) and OpenGL 4.6. However, the GTX 765M also supports Vulkan 1.2.175, while the GTX 670M has no Vulkan support listed in the database.

Q: Which GPU has more shading units and what is the memory configuration difference?

A: The GTX 765M has 768 shading units versus 336 for the GTX 670M. The GTX 670M uses 1536 MB of GDDR5 on a 192-bit bus with 72.00 GB/s bandwidth, while the GTX 765M uses 2 GB of GDDR5 on a 128-bit bus with 64.13 GB/s bandwidth.

Architecture Differences

The two GPUs represent distinct architectural generations from NVIDIA. The GTX 670M is built on the Fermi 2.0 architecture with the GF114 chip, manufactured on a 40 nm process at TSMC. The GTX 765M moves to the Kepler architecture with the GK106 chip, manufactured on a 28 nm process, also at TSMC. This process shrink is significant: the GTX 670M has a die size of 332 mm², while the GTX 765M packs more transistors into a smaller 221 mm² die.

The transistor counts reflect the generational leap. The GTX 670M contains 1,950 million transistors, while the GTX 765M contains 2,540 million. Transistor density jumps from 5.9M per mm² on the older chip to 11.5M per mm² on the newer one. This density improvement is a direct result of the smaller manufacturing node, allowing the Kepler chip to fit roughly 30% more transistors into a die that is about 33% smaller by area.

The shading unit count differs dramatically: the GTX 765M has 768 shading units, more than double the GTX 670M's 336. Texture mapping units also increase from 56 on the GTX 670M to 64 on the GTX 765M. However, the render output units move in the opposite direction: the GTX 670M has 24 ROPs, while the GTX 765M has only 16. This combination means the newer GPU relies on a wider shading array but a narrower final pixel output stage.

Memory architecture also differs substantially. The GTX 670M uses a 192-bit memory bus with 1536 MB of GDDR5, yielding 72.00 GB/s of bandwidth. The GTX 765M uses a narrower 128-bit bus but pairs it with 2 GB of GDDR5, resulting in 64.13 GB/s. The GTX 765M's memory clock is higher at 1002 MHz (4 Gbps effective) versus 750 MHz (3 Gbps effective) for the GTX 670M, but the narrower bus limits the overall bandwidth.

Clock behavior also differs. The GTX 670M has no listed base or boost clock in the database, while the GTX 765M has a base clock of 797 MHz and a boost clock of 863 MHz. The GTX 765M also supports Vulkan 1.2.175, whereas the GTX 670M has no Vulkan support recorded. Both GPUs share the same 75 W TDP, MXM module slot width, no power connectors, and MXM-B (3.0) bus interface.

The Verdict

The data points to a clear but nuanced conclusion. For raw compute throughput and modern feature support, the GTX 765M is the stronger choice. Its OpenCL score of 7176 beats the GTX 670M's 6513 by 9.2%. The Kepler architecture brings a much larger shading array, higher texture rate, and Vulkan support, which the Fermi-based GTX 670M lacks entirely.

However, the GTX 670M holds a higher overall percentile rank (38 versus 32) and a higher average benchmark score (6513 versus 5501). This apparent contradiction stems from the benchmarking methodology: the GTX 670M has only one recorded OpenCL benchmark, while the GTX 765M has three recorded tests including Metal and Vulkan results. The GTX 765M's average is dragged down by its Metal score of 2612, which is far below its OpenCL and Vulkan results.

For users who prioritize the OpenCL workload, the GTX 765M is the clear winner. The 9.2% lead in that specific test is meaningful, and the additional Vulkan capability makes it more future-proof for modern applications. The GTX 765M also delivers a higher pixel rate (13.81 GPixel/s versus 8.372 GPixel/s) and texture rate (55.23 GTexel/s versus 33.49 GTexel/s), which translates to better fill-rate performance in graphics-heavy scenarios.

For users who care about the aggregate benchmark profile, the GTX 670M presents a more consistent picture. Its single recorded score of 6513 places it at the 38th percentile among all GPUs, while the GTX 765M sits at the 32nd percentile. The GTX 670M also offers higher memory bandwidth (72.00 GB/s versus 64.13 GB/s), which can matter in bandwidth-sensitive workloads despite the older architecture.

The GTX 765M is the better pick for modern graphics workloads, Vulkan-based applications, and scenarios where shading throughput matters most. The GTX 670M remains competitive for memory-bandwidth-bound tasks and holds a better overall percentile standing in the database. Both are end-of-life products, so the decision hinges on workload specificity rather than future upgradability.

Specification Differences

The two GPUs differ across nearly every core specification. The GTX 670M uses the GF114 chip on a 40 nm process, while the GTX 765M uses the GK106 chip on a 28 nm process. Transistor counts are 1,950 million versus 2,540 million, and die sizes are 332 mm² versus 221 mm². Transistor density improves from 5.9M per mm² to 11.5M per mm².

Shading units jump from 336 to 768, TMUs from 56 to 64, but ROPs drop from 24 to 16. The GTX 670M has no listed base or boost clocks, while the GTX 765M has a base clock of 797 MHz and a boost clock of 863 MHz. Memory clocks differ as well: 750 MHz (3 Gbps effective) for the GTX 670M versus 1002 MHz (4 Gbps effective) for the GTX 765M.

Memory configuration changes from 1536 MB on a 192-bit bus to 2 GB on a 128-bit bus. Bandwidth decreases from 72.00 GB/s to 64.13 GB/s. Pixel rate increases from 8.372 GPixel/s to 13.81 GPixel/s, and texture rate increases from 33.49 GTexel/s to 55.23 GTexel/s. FP32 performance rises from 803.7 GFLOPS to 1,325.6 GFLOPS.

The GTX 670M belongs to the GeForce 600M generation with Fermi 2.0 architecture, while the GTX 765M belongs to the GeForce 700M generation with Kepler architecture. Release dates differ: March 2012 for the GTX 670M and May 2013 for the GTX 765M. The predecessor and successor chains also differ: the GTX 670M follows GeForce 500M and precedes GeForce 700M, while the GTX 765M follows GeForce 600M and precedes GeForce 800M.

API support is identical for DirectX (12 (11_0)) and OpenGL (4.6), but the GTX 765M adds Vulkan 1.2.175 support while the GTX 670M has none. Both GPUs share the same TDP of 75 W, MXM module slot width, no power connectors, MXM-B (3.0) bus interface, and portable-device-dependent display outputs.

Head-to-Head Benchmarks

The database contains one direct head-to-head benchmark between these two GPUs: Geekbench OpenCL. In this test, the GTX 765M scores 7176 against the GTX 670M's 6513. The delta is -9.2% from the GTX 670M's perspective, meaning the GTX 765M holds a 9.2% advantage in this workload.

This OpenCL result aligns with the broader compute specifications. The GTX 765M's FP32 throughput of 1,325.6 GFLOPS is substantially higher than the GTX 670M's 803.7 GFLOPS, a 64.9% difference. The shading unit count of 768 versus 336 explains much of this gap, as OpenCL workloads typically scale with shading throughput.

The GTX 765M also records two additional benchmarks that the GTX 670M does not have: Geekbench Metal with a score of 2612 and Geekbench Vulkan with a score of 6714. The Vulkan score is particularly notable, as it approaches the OpenCL result, suggesting consistent compute performance across different APIs. The Metal score is considerably lower, likely reflecting the different optimization paths for that API on this hardware.

The GTX 670M's single benchmark score of 6513 in OpenCL is respectable for its generation, placing it at the 38th percentile among all GPUs. Its nearest rivals in the database include the NVIDIA GeForce GT 555M at 6493 (0.3% behind), the NVIDIA Quadro M5000M at 6481 (0.5% behind), and the AMD Radeon Vega 10 Mobile at 6476 (0.6% behind). The Intel UHD Graphics P750 sits slightly ahead at 6554 (-0.6% delta from the GTX 670M's perspective).

The GTX 765M's nearest rivals cluster around its average score of 5501. The NVIDIA GeForce MX130 is nearly identical at 5508 (-0.1% delta), the AMD Radeon R7 M440 trails at 5483 (0.3% ahead of the GTX 765M), and the NVIDIA Quadro M4000 sits at 5467 (0.6% ahead). The AMD FirePro M4000 leads slightly at 5537 (-0.7% delta from the GTX 765M's perspective).

Where Each One Wins

The GTX 765M wins in compute-heavy scenarios. Its OpenCL score of 7176 outperforms the GTX 670M's 6513 by 9.2%. The higher shading unit count (768 versus 336) and FP32 throughput (1,325.6 GFLOPS versus 803.7 GFLOPS) make it the stronger choice for general-purpose GPU compute tasks. The texture rate of 55.23 GTexel/s versus 33.49 GTexel/s also favors the GTX 765M in texture-bound rendering workloads.

The GTX 765M also wins on modern API support. It is the only one of the two with Vulkan 1.2.175 support, and its Vulkan benchmark score of 6714 confirms that the hardware can deliver strong performance through that API. The Metal score of 2612 provides an additional compatibility path for Apple ecosystem applications, though the performance is notably lower than its other results.

The GTX 670M wins in memory bandwidth. Its 72.00 GB/s bandwidth exceeds the GTX 765M's 64.13 GB/s, and the wider 192-bit bus provides more headroom for bandwidth-sensitive workloads. The higher percentile rank (38 versus 32) and higher average benchmark score (6513 versus 5501) also favor the GTX 670M in the database's overall ranking.

The GTX 670M wins in pixel throughput relative to its ROP count. While the GTX 765M has a higher absolute pixel rate (13.81 GPixel/s versus 8.372 GPixel/s), the GTX 670M achieves its rate with only 24 ROPs, suggesting efficient ROP utilization. The GTX 670M's nearest rivals are all within 0.6% of its score, indicating a tightly competitive field, while the GTX 765M's rivals span a slightly wider range.

For users running OpenCL applications on older Fermi-compatible codebases, the GTX 670M remains viable due to its competitive single-test score and better memory bandwidth. For users running Vulkan applications or needing maximum shading throughput, the GTX 765M is the clear choice. The data ultimately shows two GPUs from different architectural eras, with the newer Kepler chip delivering superior compute performance and the older Fermi chip retaining advantages in bandwidth and overall percentile standing.