NVIDIA GeForce GTX 1080 vs NVIDIA GeForce GTX 670 Comparison

The NVIDIA GeForce GTX 670 and the NVIDIA GeForce GTX 1080 represent two distinct eras of GPU design, separated by four years of architectural evolution. The data from the benchmark suite shows a decisive performance hierarchy, with the GTX 1080 winning all three head-to-head comparisons. The most lopsided result appears in the Geekbench OpenCL test, where the GTX 1080 scores 51,204 against the GTX 670’s 15,341, a 70% deficit for the older card. The Metal test tells a similar story, with the GTX 1080 posting 23,824 versus 7,424, a 68.8% gap. The Vulkan benchmark is comparatively closer, but the GTX 1080 still leads by 48.8%, scoring 30,398 against 15,553.

These deltas are not marginal; they represent a generational leap in compute throughput. The GTX 1080’s average benchmark score across all tests is 11,960, which places it at the 51st percentile of all GPUs. The GTX 670’s average of 12,773 is actually higher than the GTX 1080’s average, yet it sits at the 52nd percentile. This apparent contradiction is a function of the benchmark pool: the GTX 670’s average is buoyed by its three Geekbench results, while the GTX 1080’s average is dragged down by its Passmark DirectX 9, 10, and 12 scores, which are 211, 93, and 55 respectively. The GTX 670 has no Passmark data, so the comparison is not perfectly apples-to-apples, but the head-to-head Geekbench results leave no doubt about raw capability.

Head-to-Head Benchmarks

The Geekbench OpenCL test is the clearest indicator of compute potential. The GTX 1080’s 51,204 score is more than triple the GTX 670’s 15,341. This 70% delta reflects not just more shading units, but also the efficiency of the Pascal architecture in handling parallel workloads. For any OpenCL-accelerated task—be it video encoding, physics simulation, or data processing—the GTX 1080 is in a different performance class. The GTX 670’s Kepler architecture was competent in its era, but its 1,344 shading units and 2.634 TFLOPS of FP32 throughput are dwarfed by the GTX 1080’s 2,560 shading units and 8.873 TFLOPS.

The Geekbench Metal test shows a similar, though slightly less extreme, margin. The GTX 1080 scores 23,824, which is 68.8% ahead of the GTX 670’s 7,424. Metal is Apple’s low-level graphics API, and the results indicate that the GTX 1080 can feed modern graphics pipelines with significantly more geometry and pixel data. The GTX 670’s pixel rate of 27.44 GPixel/s and texture rate of 109.8 GTexel/s are less than a third of the GTX 1080’s 110.9 GPixel/s and 277.3 GTexel/s. In practical terms, the GTX 1080 can handle higher resolutions and more complex shaders without bottlenecking the CPU.

The Geekbench Vulkan result is the most interesting because the gap narrows to 48.8%. The GTX 1080 scores 30,398, while the GTX 670 manages 15,553. Vulkan is a low-overhead API that can sometimes favor older architectures if they have efficient driver implementations. The GTX 670’s Vulkan support is listed as 1.2.175, which is functional, but the GTX 1080’s 1.4 support is more complete. Even in this best-case scenario for the GTX 670, the GTX 1080 is still nearly 50% faster. The data suggests that while the older card can run Vulkan titles, it will do so at lower frame rates and with more frequent frame-time spikes.

The Verdict

The benchmark data is unambiguous: the GTX 1080 is the superior GPU for every measured workload. It wins all three head-to-head tests, with deltas ranging from 48.8% to 70%. The GTX 670’s sole statistical advantage is its average benchmark score of 12,773 versus 11,960, but this is misleading because the two cards were not subjected to the same test suite. The GTX 670 has no Passmark results, which is where the GTX 1080’s average suffers. A buyer choosing between these two cards on performance alone would select the GTX 1080 without hesitation.

The GTX 670’s percentile ranking of 52 is one point higher than the GTX 1080’s 51, but this is a rounding artifact of the different benchmark distributions. In the head-to-head tests, the GTX 670 is not competitive. The GTX 1080’s nearest rivals include the GTX 1660 (2.4% slower) and the RX 7800 XT (2.9% slower), which puts it in a performance tier that is still relevant for modern gaming. The GTX 670’s nearest rivals are the GTX 590, the Radeon Pro 455, and the RX 7600M XT, with deltas of -0.4% to 0.5%. This means the GTX 670 is essentially tied with those cards, but all of them are at the 52nd percentile, which is mid-pack.

Where Each One Wins

The GTX 1080 wins every category where data exists. For compute-heavy tasks like OpenCL and Vulkan workloads, it is the clear choice. The GTX 1080’s 8 GB of GDDR5X memory provides 320.3 GB/s of bandwidth, compared to the GTX 670’s 2 GB of GDDR5 at 192.3 GB/s. This 8 GB capacity is critical for modern textures and high-resolution assets; the GTX 670’s 2 GB will cause texture thrashing in any game with high-detail settings. The GTX 1080 also has 64 ROPs versus 32, which doubles its pixel fill rate and helps with anti-aliasing and post-processing effects.

The GTX 670 does have one theoretical advantage: power consumption. Its TDP is 170 W, which is 10 W lower than the GTX 1080’s 180 W. Both cards require a 450 W power supply, but the GTX 670 uses two 6-pin connectors while the GTX 1080 needs only one 8-pin. This makes the GTX 1080 easier to cable-manage, but the GTX 670 is marginally more power-efficient. However, this is a negligible difference in real-world usage, and the GTX 1080’s performance per watt is far superior given its 3.4x higher FP32 throughput.

Specification Differences

The most significant specification gap is in memory. The GTX 1080 offers 8 GB of GDDR5X, while the GTX 670 is limited to 2 GB of GDDR5. Both use a 256-bit bus, but the GTX 1080’s faster memory clock (10 Gbps effective versus 6 Gbps) yields a bandwidth of 320.3 GB/s, nearly double the GTX 670’s 192.3 GB/s. The shading unit count also differs drastically: 2,560 versus 1,344. The GTX 1080 has 160 texture mapping units (TMUs) to the GTX 670’s 112, and 64 ROPs to 32.

Clock speeds show the architectural efficiency gain. The GTX 670’s base clock is 915 MHz with a boost of 980 MHz, while the GTX 1080 runs at 1,607 MHz base and 1,733 MHz boost. This is a 75% higher base clock, which is only possible because of the 16 nm process node versus 28 nm. The GTX 1080’s die size is slightly larger at 314 mm² versus 294 mm², but it packs 7,200 million transistors against 3,540 million, resulting in a transistor density of 22.9M / mm² versus 12.0M / mm². The GTX 1080 also supports DirectX 12 (12_1), while the GTX 670 is limited to DirectX 12 (11_0). Display outputs differ as well: the GTX 1080 has 1x DVI, 1x HDMI 2.0, and 3x DisplayPort 1.4a, while the GTX 670 has 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

Architecture Differences

The GTX 670 is built on the Kepler architecture, featuring the GK104 chip. Kepler was designed for high clock speeds and efficient power use, but it lacks the asynchronous compute and improved scheduling of later architectures. The GTX 1080 uses the Pascal architecture with the GP104 chip. Pascal introduced a unified memory architecture with better L2 cache handling and more efficient compute preemption. The GTX 1080’s FP16 performance is listed at 138.6 GFLOPS (1:64), which is a token capability, but its FP32 is 8.873 TFLOPS, indicating that Pascal’s primary focus was on single-precision compute.

The process node is the biggest architectural divider: 28 nm for Kepler versus 16 nm for Pascal. This allowed the GTX 1080 to nearly double the transistor count while maintaining a similar TDP (180 W versus 170 W). The GTX 670 is 241 mm long and 38 mm thick, while the GTX 1080 is 267 mm long and 40 mm thick, making the newer card slightly larger. Both are dual-slot designs. The GTX 670’s API support tops out at Vulkan 1.2.175, while the GTX 1080 reaches Vulkan 1.4. The GTX 1080’s DirectX 12 (12_1) feature level enables advanced rendering techniques like conservative rasterization and rasterizer ordered views, which the GTX 670 cannot use. For a modern workload, the GTX 1080 is not just faster; it is architecturally capable of tasks that the GTX 670 cannot perform at all.