NVIDIA GeForce GTX 1070 Ti vs NVIDIA GeForce GTX 670 Comparison

Head-to-Head Benchmarks

The recorded data shows a decisive generational sweep for the NVIDIA GeForce GTX 1070 Ti across all three shared benchmark tests. In Geekbench Metal, the GTX 1070 Ti posts a score of 22,180 against the GTX 670's 7,424, a delta of 198.8%. This is not a marginal improvement; it is a near-tripling of raw compute throughput in Apple's Metal API. The gap widens further in Geekbench OpenCL, where the newer card scores 49,219 versus 15,341, translating to a 220.8% advantage. For Vulkan workloads, the result is even more lopsided: 61,006 for the GTX 1070 Ti versus 15,553 for the GTX 670, a 292.2% lead. Every head-to-head measurement favors the Pascal-based card, with no benchmark in the database where the GTX 670 manages a win.

The magnitude of these deltas is worth emphasizing. A 292.2% improvement in Vulkan indicates that the GTX 1070 Ti is not merely faster in a linear sense; it is operating in a different performance class. The GTX 670's best showing is in Metal, where it still trails by nearly 200%. Even the GTX 1070 Ti's weakest relative result, the 198.8% Metal gap, represents more than double the older card's output. These are not close contests. The data suggests architectural obsolescence rather than incremental progress, as the GTX 670's Kepler design simply cannot keep pace with Pascal's efficiency and throughput advantages.

Looking at average benchmark scores across all recorded tests, the GTX 1070 Ti averages 14,277, while the GTX 670 averages 12,773. That 1,504-point gap equates to roughly 11.8% overall, but this figure masks the true disparity because the GTX 670 only has three recorded benchmarks, all of which are in the same Geekbench suite. The GTX 1070 Ti's average includes additional tests like 3DMark Steel Nomad (1,471), PassMark DirectX 10 (90), DirectX 11 (107), DirectX 12 (51), DirectX 9 (207), G2D (876), G3D (14,674), and GPU Compute (7,165). When comparing only the overlapping Geekbench tests, the GTX 1070 Ti's superiority is overwhelming.

Where Each One Wins

Based on the benchmark results, the GTX 1070 Ti wins in every tested category. For compute-heavy applications using OpenCL or Vulkan, the gap is massive, with the newer card delivering over three times the performance in Vulkan. This makes it the clear choice for modern workloads that leverage these APIs, such as GPU-accelerated rendering, machine learning inference, or game engines built on Vulkan. The Metal results, while less extreme, still show a 198.8% advantage, positioning the GTX 1070 Ti as the superior option for macOS or iOS development environments that rely on Apple's graphics framework.

The GTX 670 has no benchmark wins in the database. Its only recorded scores are the three Geekbench entries, and it loses each one by a significant margin. However, the data does not imply the GTX 670 is useless; it simply shows that in every measured scenario, the GTX 1070 Ti outperforms it. For users constrained to older DirectX 9 or DirectX 10 titles, the GTX 670 could still function, but the database does not contain passmark scores for that card to confirm this. The GTX 1070 Ti's PassMark DirectX 9 score of 207 and DirectX 10 score of 90 suggest it handles legacy APIs capably, but without equivalent GTX 670 numbers, a direct comparison is impossible.

The percentile rankings place the GTX 1070 Ti at the 56th percentile among all GPUs, while the GTX 670 sits at the 52nd percentile. This four-point difference indicates that the GTX 1070 Ti is positioned slightly above average in the overall GPU landscape, whereas the GTX 670 barely clears the midpoint. For users seeking a card that remains competitive in modern benchmarks, the GTX 1070 Ti's higher percentile and superior average score make it the rational choice. The GTX 670's nearest rivals include the NVIDIA GeForce GTX 590 (average 12,830, delta -0.4%) and AMD Radeon Pro 455 (12,831, -0.4%), showing it clusters with older high-end parts, while the GTX 1070 Ti's rivals include the Intel Iris Xe MAX Graphics (14,315, -0.3%) and AMD Radeon Vega 11 (14,352, -0.5%), placing it among more modern integrated and entry-level discrete solutions.

FAQ

Q: Which card has the higher average benchmark score?

A: The NVIDIA GeForce GTX 1070 Ti averages 14,277 across all recorded benchmarks, while the NVIDIA GeForce GTX 670 averages 12,773. The newer card leads by 1,504 points.

Q: How much faster is the GTX 1070 Ti in Vulkan compared to the GTX 670?

A: In Geekbench Vulkan, the GTX 1070 Ti scores 61,006 versus 15,553 for the GTX 670, a 292.2% advantage. This is the largest performance gap between the two cards in any shared test.

Q: Does the GTX 670 win any benchmark against the GTX 1070 Ti?

A: No. The database records three head-to-head benchmarks (Geekbench Metal, OpenCL, and Vulkan), and the GTX 1070 Ti wins all three. The GTX 670 has zero wins.

Q: What is the GTX 1070 Ti's percentile ranking among all GPUs?

A: The GTX 1070 Ti sits at the 56th percentile, meaning it outperforms 56% of all GPUs in the database. The GTX 670 is at the 52nd percentile.

Q: Are both cards the same physical size?

A: No. The GTX 1070 Ti is 267 mm long, 112 mm high, and 40 mm wide. The GTX 670 is 241 mm long, 111 mm high, and 38 mm wide. Both are dual-slot designs, but the GTX 1070 Ti is slightly larger in every dimension.

Q: Which card supports a newer version of Vulkan?

A: The GTX 1070 Ti supports Vulkan 1.4, while the GTX 670 supports Vulkan 1.2.175. Both cards support DirectX 12, but the GTX 1070 Ti implements version 12_1, whereas the GTX 670 implements version 11_0.

Specification Differences

The two cards differ substantially in nearly every specification. The GTX 1070 Ti uses the GP104 chip on a 16 nm process, while the GTX 670 uses GK104 on a 28 nm process. Transistor counts reflect this: the GTX 1070 Ti packs 7,200 million transistors on a 314 mm² die, giving a density of 22.9M per mm². The GTX 670 has 3,540 million transistors on a 294 mm² die, at 12.0M per mm². Clock speeds are higher on the newer card, with a base of 1607 MHz and boost of 1683 MHz, compared to 915 MHz base and 980 MHz boost on the GTX 670. Memory speeds also differ: the GTX 1070 Ti runs at 2002 MHz (8 Gbps effective), while the GTX 670 runs at 1502 MHz (6 Gbps effective).

Memory capacity and bandwidth favor the GTX 1070 Ti. It ships with 8 GB of GDDR5 on a 256-bit bus, delivering 256.3 GB/s. The GTX 670 has 2 GB of GDDR5 on the same 256-bit bus, but only achieves 192.3 GB/s due to slower memory clocks. The compute resources are dramatically different: the GTX 1070 Ti has 2,432 shading units, 152 texture mapping units, and 64 ROPs. The GTX 670 has 1,344 shading units, 112 TMUs, and 32 ROPs. Pixel and texture rates follow suit: the GTX 1070 Ti outputs 107.7 GPixel/s and 255.8 GTexel/s, while the GTX 670 manages 27.44 GPixel/s and 109.8 GTexel/s. Floating-point performance is 8.186 TFLOPS for the GTX 1070 Ti versus 2.634 TFLOPS for the GTX 670 in FP32. The GTX 1070 Ti also supports FP16 at 127.9 GFLOPS (1:64 ratio), while the GTX 670 has no recorded FP16 capability.

Power delivery differs as well. The GTX 1070 Ti has a TDP of 180 W and uses a single 8-pin power connector. The GTX 670 has a TDP of 170 W and uses two 6-pin connectors. Both cards recommend a 450 W power supply and use a PCIe 3.0 x16 interface. Display outputs vary: the GTX 1070 Ti offers 1x DVI, 1x HDMI 2.0, and 3x DisplayPort 1.4a, while the GTX 670 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. API support differs in DirectX version (12_1 versus 11_0) and Vulkan version (1.4 versus 1.2.175), though both support OpenGL 4.6. Physical dimensions show the GTX 1070 Ti is longer (267 mm versus 241 mm) and slightly wider (40 mm versus 38 mm), with equal height (112 mm versus 111 mm). The GTX 1070 Ti was released on 2017-11-01, while the GTX 670 launched on 2012-05-09. Both cards have a launch MSRP of 399 USD.

Architecture Differences

The architectural split between Pascal and Kepler explains the performance chasm. The GTX 1070 Ti is built on NVIDIA's Pascal architecture, which introduced significant improvements in clock scaling and power efficiency compared to the older Kepler design found in the GTX 670. Pascal's 16 nm TSMC process allows for much higher transistor density (22.9M per mm² versus 12.0M per mm²) and enables the higher clock speeds observed in the benchmarks. Kepler, by contrast, was designed for the 28 nm node and relied on higher core counts to achieve performance, which limited its frequency headroom.

The GP104 chip in the GTX 1070 Ti features a more modern execution pipeline with improved scheduling and memory compression techniques. This is evident in the FP32 throughput, where the GTX 1070 Ti delivers 8.186 TFLOPS, over three times the GTX 670's 2.634 TFLOPS. The shading unit count difference (2,432 versus 1,344) is substantial, but the per-clock efficiency gains in Pascal mean the newer card extracts more work from each shader. The GTX 1070 Ti also benefits from a larger L2 cache and better memory compression, contributing to its 256.3 GB/s effective bandwidth despite using the same 256-bit bus width as the GTX 670.

The GTX 670's Kepler architecture introduced features like GPU Boost and improved geometry processing for its time, but it lacks the asynchronous compute and enhanced multi-projector capabilities that Pascal brought. DirectX 12 support is present on both cards, but the GTX 1070 Ti implements feature level 12_1, including conservative rasterization and rasterizer-ordered views, while the GTX 670 is limited to 11_0. Vulkan support also differs: the GTX 1070 Ti supports version 1.4, which includes newer extensions for ray tracing and mesh shaders (though the card has no dedicated RT cores), while the GTX 670 supports only 1.2.175. The GTX 1070 Ti's FP16 capability at 127.9 GFLOPS with a 1:64 ratio indicates it can handle some mixed-precision workloads, a feature absent from the GTX 670 entirely.

Process node improvements are the root cause of most differences. The 16 nm node allows the GTX 1070 Ti to fit 7,200 million transistors into a die only 20 mm² larger than the GTX 670's 294 mm², despite having over twice the transistor count. This density gain translates directly to the performance deltas seen in the benchmarks. The GTX 670's 170 W TDP versus the GTX 1070 Ti's 180 W shows that the newer card delivers roughly three times the compute throughput while consuming only 10 W more power, a signal of Pascal's efficiency. The power connector layout also reflects this: the GTX 1070 Ti requires a single 8-pin connector, while the GTX 670 needs two 6-pins, suggesting the newer card draws power more efficiently.