NVIDIA GeForce GTX 670 vs NVIDIA GeForce GTX 680 Comparison

FAQ

Q: Which GPU is faster in the database benchmarks, the GTX 680 or the GTX 670?

A: The NVIDIA GeForce GTX 680 wins all three recorded head-to-head tests. It leads by 6.4% in Geekbench Metal, by 10.2% in Geekbench OpenCL, and by 13.5% in Geekbench Vulkan.

Q: What is the average benchmark score for each card?

A: The GTX 680 has an average benchmark score of 14150, while the GTX 670 scores 12773. That places the GTX 680 in the 55th percentile of all GPUs in the database, compared to the 52nd percentile for the GTX 670.

Q: Do both cards use the same GPU chip and manufacturing process?

A: Yes, both are built on the GK104 chip with the Kepler architecture, fabricated by TSMC on a 28 nm process. Both have 3,540 million transistors and a die size of 294 mm².

Q: How much memory does each card have, and what is the memory bandwidth?

A: Both cards come with 2 GB of GDDR5 memory on a 256-bit bus, delivering 192.3 GB/s of bandwidth. The effective memory clock is 6 Gbps for both.

Q: Is there a difference in power consumption between the two?

A: Yes, the GTX 680 has a TDP of 195 W, while the GTX 670 is rated at 170 W. Both use dual-slot cooling and require two 6-pin power connectors, with a suggested power supply of 450 W.

Q: Which card has a higher launch MSRP?

A: The GTX 680 had a launch MSRP of 499 USD, while the GTX 670 had a launch MSRP of 399 USD.

Architecture Differences

The two cards share the same fundamental architecture: both use the GK104 chip, the Kepler design, and the 28 nm process from TSMC. They also have identical transistor counts at 3,540 million and the same die size of 294 mm², which means the transistor density is exactly 12.0M per mm² for both. The memory subsystem is also identical, with 2 GB of GDDR5, a 256-bit bus, 192.3 GB/s bandwidth, and 6 Gbps effective memory speed.

The key architectural difference lies in the compute resources. The GTX 680 packs 1,536 shading units, 128 texture mapping units, and 32 ROPs. The GTX 670 scales down to 1,344 shading units, 112 TMUs, and retains 32 ROPs. This reduction in shader and texture hardware is the primary reason for the performance gap. Clock speeds also differ: the GTX 680 runs at 1006 MHz base and 1058 MHz boost, while the GTX 670 operates at 915 MHz base and 980 MHz boost.

These differences translate into measurable throughput figures. The GTX 680 achieves a pixel rate of 33.86 GPixel/s and a texture rate of 135.4 GTexel/s, while the GTX 670 delivers 27.44 GPixel/s and 109.8 GTexel/s. In terms of floating-point compute, the GTX 680 reaches 3.250 TFLOPS of FP32 performance, whereas the GTX 670 tops out at 2.634 TFLOPS. The GTX 680 also carries a higher TDP at 195 W versus 170 W for the GTX 670, reflecting the extra hardware and higher clocks.

Both cards support the same API feature set, including DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. They have identical display outputs (2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2) and the same PCIe 3.0 x16 bus interface. Physical dimensions differ slightly in length: the GTX 680 measures 256 mm (10.1 inches), while the GTX 670 is shorter at 241 mm (9.5 inches). Height and width are the same for both at 111 mm (4.4 inches) and 38 mm (1.5 inches), respectively. The release dates are also different, with the GTX 680 launching on 2012-03-21 and the GTX 670 following on 2012-05-09.

Where Each One Wins

The GTX 680 is the clear winner in every recorded benchmark. It takes all three head-to-head tests, which means the GTX 670 does not have a single win in the database. The GTX 680's advantage is most pronounced in compute-heavy workloads, as reflected by the 13.5% lead in Geekbench Vulkan and the 10.2% lead in Geekbench OpenCL. These are significant margins that suggest the extra shading units and higher clocks deliver tangible benefits in parallel compute tasks.

For the GTX 670, the situation is less favorable. It trails in all scenarios, but the gap narrows in Geekbench Metal, where the delta is just 6.4%. This suggests that in certain API-bound or driver-optimized workloads, the GTX 670 can stay relatively close to its bigger sibling. The GTX 670 also has a lower TDP, which could be an advantage in thermally constrained systems, and it is shorter in length, making it easier to fit in compact cases. However, from a pure performance standpoint, the GTX 680 wins outright.

The database's percentile rankings reinforce this split. The GTX 680 sits at the 55th percentile of all GPUs, while the GTX 670 sits at the 52nd percentile. That three-point gap places the GTX 680 in a slightly higher performance tier overall. For users seeking maximum compute throughput, the GTX 680 is the obvious choice. For those prioritizing lower power draw or a smaller physical footprint, the GTX 670 offers a viable alternative, but it cannot match the GTX 680's benchmark results.

Specification Differences

The two cards differ in several key specifications, all of which favor the GTX 680 except for power consumption and physical length.

• Base Clock: GTX 680 at 1006 MHz, GTX 670 at 915 MHz
• Boost Clock: GTX 680 at 1058 MHz, GTX 670 at 980 MHz
• Shading Units: 1536 on GTX 680, 1344 on GTX 670
• Texture Mapping Units: 128 on GTX 680, 112 on GTX 670
• Pixel Rate: 33.86 GPixel/s on GTX 680, 27.44 GPixel/s on GTX 670
• Texture Rate: 135.4 GTexel/s on GTX 680, 109.8 GTexel/s on GTX 670
• FP32 Compute: 3.250 TFLOPS on GTX 680, 2.634 TFLOPS on GTX 670
• TDP: 195 W on GTX 680, 170 W on GTX 670
• Length: 256 mm on GTX 680, 241 mm on GTX 670
• Launch MSRP: 499 USD on GTX 680, 399 USD on GTX 670

The memory configuration is identical, with 2 GB GDDR5, 256-bit bus, 192.3 GB/s bandwidth, and 6 Gbps effective speed. ROPs are also the same at 32. Both use the same GK104 chip, 28 nm process, 3,540 million transistors, and 294 mm² die size. The bus interface, display outputs, API support, power connectors, and suggested PSU are also shared between the two.

Head-to-Head Benchmarks

The database records three direct comparisons between these two cards, and the GTX 680 wins all of them. The largest margin comes in the Geekbench Vulkan test, where the GTX 680 scores 17646 against the GTX 670's 15553, a delta of 13.5%. This is a substantial gap that highlights the GTX 680's advantage in compute-heavy workloads that leverage the Vulkan API. The extra shading units and higher clock speeds clearly translate into measurable performance gains.

In Geekbench OpenCL, the GTX 680 again takes the lead with a score of 16906 versus 15341 for the GTX 670, representing a 10.2% advantage. This result aligns with the compute-focused nature of OpenCL, where the GTX 680's higher FP32 throughput of 3.250 TFLOPS versus 2.634 TFLOPS provides a clear edge. The GTX 670's lower texture rate of 109.8 GTexel/s also contributes to its deficit in this workload.

The closest contest is in Geekbench Metal, where the GTX 680 scores 7897 and the GTX 670 scores 7424, a 6.4% lead for the GTX 680. While this is still a decisive win, the smaller margin suggests that Metal workloads are less sensitive to the differences in shader count and clocks. Even so, the GTX 680 maintains its winning streak across all three APIs.

Looking at the broader context, the GTX 680's average benchmark score of 14150 places it just 0.9% above the NVIDIA Tesla K10 (14029) and 0.9% below the NVIDIA GeForce GTX 1070 Ti (14277) in the nearest rivals list. The GTX 670, with an average score of 12773, sits 0.4% below the NVIDIA GeForce GTX 590 (12830) and 0.5% above the AMD Radeon RX 7600M XT (12710). These relative positions confirm that the GTX 680 occupies a higher performance tier, and the head-to-head data shows it is consistently faster than the GTX 670 in every recorded workload.