NVIDIA GeForce GTX 680 vs NVIDIA GeForce GTX 950 Comparison

The Verdict

The data presents a clear hierarchy: the NVIDIA GeForce GTX 680 outperforms the GTX 950 across every benchmark in the comparison set, but the margins are surprisingly modest for a card from an older generation. The GTX 680 averages a 7.8% lead across the three Geekbench tests, with its largest advantage (10.1%) coming in the Metal API test. For a user prioritizing raw compute performance, the GTX 680 is the stronger choice, particularly in Metal-based workloads where it achieves a score of 7897 versus the GTX 950's 7172. However, the GTX 950 counters with significant efficiency advantages: its 90 W TDP is less than half of the GTX 680's 195 W, and it requires only a single 6-pin power connector and a 250 W suggested PSU, compared to the GTX 680's dual 6-pin connectors and 450 W suggested PSU. The GTX 950 also brings a newer feature set with DirectX 12 (12_1) support and Vulkan 1.4, while the GTX 680 is limited to DirectX 12 (11_0) and Vulkan 1.2.175. Users with strict power or thermal constraints, or those needing the latest API certifications, should lean toward the GTX 950 despite its compute deficit. Users whose workloads favor compute throughput and who have adequate power delivery should pick the GTX 680.

Architecture Differences

The two cards represent distinct architectural generations from NVIDIA, both fabricated on the same 28 nm TSMC process. The GTX 680 uses the GK104 chip based on Kepler architecture, while the GTX 950 uses the GM206 chip based on Maxwell 2.0. The GK104 packs 3,540 million transistors into a 294 mm² die, yielding a transistor density of 12.0 million per square millimeter. The GM206 contains fewer transistors at 2,940 million but on a smaller 228 mm² die, achieving a higher density of 12.9 million per square millimeter.

The compute resources differ substantially. The GTX 680 carries 1,536 shading units, 128 texture mapping units, and 32 ROPs. The GTX 950 has exactly half the shading units at 768, along with 48 texture mapping units and 32 ROPs. This disparity shows directly in raw throughput figures: the GTX 680 reaches 3.250 TFLOPS FP32 and 135.4 GTexel/s texture rate, while the GTX 950 manages 1.825 TFLOPS and 57.02 GTexel/s. Interestingly, the pixel rate favors the newer card slightly, with the GTX 950 hitting 38.02 GPixel/s versus 33.86 GPixel/s for the GTX 680, likely due to its higher boost clock.

Clock speeds tell a similar story. The GTX 680 has a base clock of 1006 MHz and a boost of 1058 MHz. The GTX 950 runs at 1024 MHz base and 1188 MHz boost, giving it a higher peak frequency. Memory configurations also diverge: the GTX 680 uses a 256-bit bus with 2 GB GDDR5 at 6 Gbps effective, delivering 192.3 GB/s bandwidth. The GTX 950 uses a narrower 128-bit bus with 2 GB GDDR5 at 6.6 Gbps effective, resulting in 105.8 GB/s bandwidth. Both cards support PCIe 3.0 x16, but the GTX 950 offers HDMI 2.0 and three DisplayPort 1.2 outputs, while the GTX 680 provides HDMI 1.4a and a single DisplayPort 1.2. The GTX 950 also lists Vulkan 1.4 support versus the GTX 680's Vulkan 1.2.175, and DirectX 12 (12_1) versus DirectX 12 (11_0).

Where Each One Wins

The GTX 680 wins in every benchmark category recorded, but the magnitude of its victories varies by workload. Its strongest showing is in the Metal benchmark, where it leads by 10.1%. This suggests that the Kepler architecture's larger shading unit count provides a meaningful advantage in Metal-optimized compute tasks. The OpenCL test shows a 7.9% lead, and the Vulkan test shows a 5.4% lead. These results indicate the GTX 680's advantage narrows as the API becomes more modern and efficient, but it still holds the edge.

The GTX 950 wins where the benchmark numbers don't tell the full story. Its 90 W TDP makes it suitable for systems with limited power budgets, and its single 6-pin connector and 250 W suggested PSU requirement simplify installation in compact builds. The GTX 950 also measures just 202 mm (8 inches) in length, while the GTX 680 extends to 256 mm (10.1 inches), making the newer card easier to fit in smaller cases. Additionally, the GTX 950's display outputs include HDMI 2.0, which supports higher refresh rates and resolutions on modern displays, and its Vulkan 1.4 API support ensures compatibility with the latest graphics software. The GTX 680's 499 USD launch MSRP versus the GTX 950's 159 USD launch MSRP reflects their market positioning, though the GTX 680's end-of-life status and the GTX 950's end-of-life status mean both are legacy products now.

FAQ

Q: Which card has a higher average benchmark score?

A: The GTX 680 has an average benchmark score of 14150, placing it in the 55th percentile of all GPUs. The GTX 950 scores 13189 on average, sitting in the 53rd percentile.

Q: How does the GTX 950's power consumption compare to the GTX 680?

A: The GTX 950 has a TDP of 90 W, while the GTX 680 has a TDP of 195 W. The GTX 950 also requires a suggested PSU of 250 W and a single 6-pin connector, whereas the GTX 680 needs a 450 W suggested PSU and two 6-pin connectors.

Q: Are there any benchmark tests where the GTX 950 beats the GTX 680?

A: No. In the three head-to-head benchmark tests (Geekbench Metal, OpenCL, and Vulkan), the GTX 680 wins all three. The GTX 950 records zero wins in the comparison set.

Q: What is the memory bandwidth difference between these two cards?

A: The GTX 680 provides 192.3 GB/s of bandwidth via a 256-bit bus, while the GTX 950 offers 105.8 GB/s over a 128-bit bus. Both have 2 GB of GDDR5 memory, but the GTX 950 runs at a higher effective speed of 6.6 Gbps versus 6 Gbps.

Q: Which card has better API support for modern graphics?

A: The GTX 950 supports DirectX 12 (12_1) and Vulkan 1.4, while the GTX 680 supports DirectX 12 (11_0) and Vulkan 1.2.175. Both cards support OpenGL 4.6.

Q: How do the nearest rivals compare to the GTX 680?

A: The GTX 680's closest competitors by average score are the NVIDIA Tesla K10 (0.9% lower), the NVIDIA GeForce GTX 1070 Ti (0.9% higher), the Intel Iris Xe MAX Graphics (1.2% higher), and the AMD Radeon Vega 11 (1.4% higher).

Head-to-Head Benchmarks

The Geekbench Metal test delivers the largest margin in this comparison. The GTX 680 scores 7897, while the GTX 950 scores 7172, producing a 10.1% delta in favor of the older card. This result is notable because Metal is a relatively modern API, and the GTX 680's Kepler architecture, despite its age, still manages to carve out a double-digit performance lead. The GTX 950's higher boost clock of 1188 MHz cannot compensate for the GTX 680's 1536 shading units versus 768, and the extra texture units on the GTX 680 (128 versus 48) likely contribute to its advantage in geometry-bound workloads.

In the OpenCL test, the GTX 680 scores 16906 against the GTX 950's 15662, a 7.9% gap. OpenCL is a general-purpose compute API, and here the GTX 680's 3.250 TFLOPS FP32 throughput versus the GTX 950's 1.825 TFLOPS plays a decisive role. The GTX 680's higher memory bandwidth (192.3 GB/s versus 105.8 GB/s) also helps feed its larger compute array, reducing the likelihood of bandwidth bottlenecks in memory-intensive kernels. The GTX 950's 57.02 GTexel/s texture rate is less than half of the GTX 680's 135.4 GTexel/s, which further limits its compute potential in texture-heavy operations.

The Vulkan test shows the closest margin, with the GTX 680 scoring 17646 and the GTX 950 scoring 16734, a 5.4% difference. Vulkan is a low-overhead API designed to maximize efficiency on modern hardware, and the GTX 950's Maxwell 2.0 architecture benefits from architectural refinements that narrow the gap. The GTX 950's higher pixel rate (38.02 GPixel/s versus 33.86 GPixel/s) suggests that its 32 ROPs are better utilized at higher clocks, but the GTX 680's superior raw shading power still prevails. This pattern—where the GTX 680 wins by a smaller margin in the newest API—indicates that the GTX 950's efficiency gains partially offset its compute deficit, but not enough to topple the GTX 680 in any recorded test.

Overall, the GTX 680 demonstrates a consistent, if modest, performance advantage across all three benchmark categories, with the average benchmark score of 14150 reinforcing its position above the GTX 950's 13189. The GTX 680's nearest rivals bracket it closely—the Tesla K10 is just 0.9% behind, and the GTX 1070 Ti is 0.9% ahead—while the GTX 950 sits in a similar competitive band, with the GTX 480 0.8% ahead and the Tesla M2090 0.9% behind. Neither card dominates its peer group, but the GTX 680 holds a higher percentile rank (55 versus 53) and a higher average score, making it the stronger performer in pure benchmark terms.