NVIDIA GeForce GTX 650 Ti Boost vs NVIDIA GeForce GTX 660 Comparison

Head-to-Head Benchmarks

The benchmark data shows a clear split between these two Kepler cards. In Geekbench Metal, the GTX 650 Ti Boost takes the win with a score of 4858, beating the GTX 660's 4305 by 11.4%. That is a notable margin for a card that sits lower in the product stack, and it suggests the 650 Ti Boost has an edge in Metal-accelerated workloads.

However, the GTX 660 strikes back decisively in the other two recorded tests. In Geekbench OpenCL, the GTX 660 scores 11347, which is 22% higher than the 650 Ti Boost's 9302. That is the largest delta in the entire head-to-head comparison. In Geekbench Vulkan, the GTX 660 again leads with 11415 versus 10041, a 13.7% advantage.

So the overall tally favors the GTX 660 with two wins out of three. The average benchmark score tells the same story: the GTX 660 averages 9022, while the 650 Ti Boost averages 8067. That puts the GTX 660 roughly 12% ahead in the aggregate, a meaningful gap for anyone choosing between these two.

Looking at the broader database context, the GTX 660 sits at the 45th percentile of all GPUs, while the 650 Ti Boost sits at the 42nd. The GTX 660's nearest rivals include the NVIDIA TITAN V CEO Edition (score 9037, delta -0.2%), the GeForce GTX 560 (9058, -0.4%), the AMD Radeon 550X (8918, +1.2%), and the AMD Radeon Pro WX 5100 (8863, +1.8%). Those deltas are all within 2%, meaning the GTX 660's average score is essentially at parity with a group of very different cards, from a workstation GPU to a much older enthusiast card.

The 650 Ti Boost's nearest rivals are the NVIDIA GRID K2 (8080, -0.2%), the GeForce GTX 650 Ti (8053, +0.2%), the GeForce GTX 880M (8040, +0.3%), and the Quadro P5000 (8039, +0.3%). Again, all within 0.3%, which indicates the 650 Ti Boost's average score is tightly clustered. The practical takeaway is that the GTX 660 occupies a slightly higher performance tier in the database, and the head-to-head results confirm that the OpenCL and Vulkan wins are what push it there.

Architecture Differences

Both cards are built on the same GK106 chip, use the Kepler architecture, and come from the GeForce 600 generation. They share the same 28 nm process node, fabricated by TSMC, with an identical transistor count of 2,540 million and a die size of 221 mm². The transistor density is also identical at 11.5M / mm². So the fundamental silicon is the same, but the configurations differ.

The GTX 660 is the more fully enabled part. It has 960 shading units, 80 texture mapping units, and 24 ROPs. The GTX 650 Ti Boost, in contrast, has 768 shading units and 64 TMUs, but it keeps the same 24 ROPs. That means the GTX 660 has 25% more shading units and 25% more TMUs, which directly explains its higher compute and texture throughput. The pixel rate for the GTX 660 is 20.64 GPixel/s versus 16.51 GPixel/s for the 650 Ti Boost, and the texture rate is 82.56 GTexel/s versus 66.05 GTexel/s. The FP32 compute is 1.981 TFLOPS for the GTX 660 and 1.585 TFLOPS for the 650 Ti Boost.

Clock speeds are identical: both run at 980 MHz base and 1032 MHz boost, with memory at 1502 MHz (6 Gbps effective). Memory configuration is also the same: 2 GB of GDDR5 on a 192-bit bus, yielding 144.2 GB/s of bandwidth. So the memory subsystem is a non-factor in the comparison; the difference comes entirely from the SM count and the resulting shading and texture throughput.

Both cards support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. Neither has ray tracing or tensor cores. Both are dual-slot cards with a single 6-pin power connector and a suggested PSU of 300 W. The TDP differs slightly: 140 W for the GTX 660 versus 134 W for the 650 Ti Boost. Display outputs are identical at 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. Both use PCIe 3.0 x16 and have the same dimensions of 241 mm (9.5 inches). The release dates differ, with the GTX 660 launching on 2012-09-05 and the 650 Ti Boost on 2013-03-25, but the architecture and feature set are otherwise unchanged.

FAQ

Q: Which card has the higher average benchmark score?

A: The GTX 660 has an average score of 9022, while the GTX 650 Ti Boost averages 8067. The GTX 660 also ranks at the 45th percentile of all GPUs, compared to the 42nd for the 650 Ti Boost.

Q: Does the GTX 650 Ti Boost win any benchmark?

A: Yes, it wins Geekbench Metal with a score of 4858, beating the GTX 660's 4305 by 11.4%. That is its only head-to-head win, but it is a substantial one.

Q: Are the memory specifications different?

A: No. Both cards have 2 GB of GDDR5, a 192-bit bus, and 144.2 GB/s of bandwidth. The memory clock is also identical at 1502 MHz (6 Gbps effective).

Q: What is the difference in shading units?

A: The GTX 660 has 960 shading units, while the GTX 650 Ti Boost has 768. That is a 25% advantage for the GTX 660, which contributes to its higher texture rate (82.56 GTexel/s versus 66.05 GTexel/s) and FP32 compute (1.981 TFLOPS versus 1.585 TFLOPS).

Q: Do these cards require different power supplies?

A: Both have the same suggested PSU rating of 300 W and use a single 6-pin power connector. The GTX 660 has a slightly higher TDP at 140 W, compared to 134 W for the 650 Ti Boost.

Q: Which card is closer to its nearest rivals in the database?

A: The GTX 650 Ti Boost is extremely close, with all four nearest rivals within 0.3% of its average score. The GTX 660's nearest rivals are within 1.8%, but the spread is larger, from a 0.2% margin to a 1.8% margin.

Specification Differences

The only differences between these two cards are in the compute configuration, power draw, and release timing. The GTX 660 has 960 shading units versus 768, 80 TMUs versus 64, and the same 24 ROPs. Its pixel rate is 20.64 GPixel/s versus 16.51 GPixel/s, and its texture rate is 82.56 GTexel/s versus 66.05 GTexel/s. FP32 compute is 1.981 TFLOPS for the GTX 660 and 1.585 TFLOPS for the 650 Ti Boost. The TDP is 140 W for the GTX 660 and 134 W for the 650 Ti Boost. The GTX 660 launched on 2012-09-05, while the 650 Ti Boost launched on 2013-03-25. The launch MSRP for the GTX 660 was 229 USD, and the 650 Ti Boost was 169 USD.

Everything else is identical: chip (GK106), architecture (Kepler), process node (28 nm), transistor count (2,540 million), die size (221 mm²), base and boost clocks (980 MHz / 1032 MHz), memory configuration (2 GB GDDR5, 192-bit, 144.2 GB/s), slot width (dual-slot), power connector (1x 6-pin), suggested PSU (300 W), bus interface (PCIe 3.0 x16), display outputs (2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2), API support (DirectX 12 (11_0), OpenGL 4.6, Vulkan 1.2.175), dimensions (241 mm / 9.5 inches), and production status (end-of-life).

Where Each One Wins

The GTX 660 wins in compute-heavy and modern API workloads. Its 22% lead in Geekbench OpenCL is the largest margin in the comparison, and its 13.7% lead in Vulkan shows it handles lower-level, multi-threaded rendering better. The extra 192 shading units and 16 TMUs give it a clear throughput advantage that shows up in raw compute tests. If the workload scales with shader count, the GTX 660 is the obvious pick. It also has a higher average score (9022 versus 8067) and a better percentile rank (45th versus 42nd), so in the database's aggregate view, it is the stronger card.

The GTX 650 Ti Boost wins in Geekbench Metal, and it does so by a meaningful 11.4%. That is not a marginal edge; it is a decisive one in that specific API. For anyone targeting Metal-accelerated applications or games on macOS, the 650 Ti Boost is the better choice according to the recorded data. It also has a lower TDP at 134 W versus 140 W, so it draws slightly less power, though the difference is small. Its nearest rivals are all within 0.3% of its average score, which suggests its performance profile is very stable relative to other cards in its tier.

The use-case split is therefore clean: pick the GTX 660 for OpenCL and Vulkan workloads, pick the 650 Ti Boost for Metal. For general-purpose use, the GTX 660's two out of three wins and higher average score make it the safer recommendation.

The Verdict

The data points to the GTX 660 as the superior card overall. It wins two of the three head-to-head benchmarks, has a higher average score (9022 versus 8067), and ranks higher in the percentile distribution (45th versus 42nd). The OpenCL margin of 22% is the single largest difference in the comparison, and the Vulkan margin of 13.7% reinforces that the GTX 660's extra shading units translate into real performance gains in compute and modern API scenarios. It also launched earlier and at a higher MSRP, which is consistent with its position as the more capable part.

The GTX 650 Ti Boost should only be chosen if the primary workload is Metal-based, where it leads by 11.4%. Its lower TDP (134 W versus 140 W) is a minor advantage, but it does not offset the GTX 660's broad superiority. The 650 Ti Boost's average score places it in a tight cluster with its nearest rivals, all within 0.3%, so it is a solid card in its own tier, but it simply does not match the GTX 660's compute throughput.

For a builder deciding between these two today, the GTX 660 is the default pick unless the specific application is Metal-heavy. The GTX 650 Ti Boost is a niche choice, and the benchmark results justify that positioning. The GTX 660's 25% more shading units and 25% more TMUs are the structural reasons for its wins, and the recorded scores confirm that those hardware differences matter in practice.