NVIDIA GeForce GTX 650 Ti Boost

The NVIDIA GeForce GTX 650 Ti Boost is a Kepler-generation graphics card built on the 28 nm process at TSMC, featuring the GK106 chip with 2,540 million transistors on a 221 mm² die. It ships with 2 GB of GDDR5 memory on a 192-bit bus, delivering 144.2 GB/s of bandwidth, and its benchmark data places it at the 45th percentile of all GPUs, with an average score of 9,659 across Geekbench OpenCL and Vulkan tests. This positions it as a decidedly mid-range part from the GeForce 600 era, now end-of-life, but its performance profile still offers a useful baseline for understanding legacy GPU scaling.

How It Compares

Against the NVIDIA GeForce GTX 960M, the GTX 650 Ti Boost trails by a razor-thin margin of -0.1% in average benchmark score (9,659 vs 9,670). That delta is within noise, meaning the two cards are effectively interchangeable in raw compute throughput, despite the 960M being a mobile part from a later generation. The data suggests desktop Kepler and mobile Maxwell at this tier landed at nearly identical performance levels.

The AMD Radeon Pro WX 2100 edges out the GTX 650 Ti Boost by -0.2% (9,675 vs 9,659), again a negligible difference. This is notable because the WX 2100 is a professional workstation card, yet its OpenCL/Vulkan scores align closely with this consumer gaming GPU. The implication is that for compute workloads, the GTX 650 Ti Boost can hold its own against entry-level pro hardware, though driver optimizations may favor one over the other in specific tasks.

The NVIDIA Tesla M10 scores 9,634, which is 0.3% lower than the GTX 650 Ti Boost. The Tesla M10 is a datacenter-oriented card, so this near-parity is surprising on the surface, but both share similar memory bandwidth constraints (the M10 also uses GDDR5), and the benchmark results indicate that raw FP32 throughput is comparable. The GTX 650 Ti Boost’s slight lead suggests its 1.585 TFLOPS of FP32 performance is well-balanced for its class.

Finally, the NVIDIA Tesla C2070 posts a score of 9,716, putting the GTX 650 Ti Boost 0.6% behind. The C2070 is an older Fermi-based compute card, yet its higher average score reflects a wider memory bus and higher bandwidth in its architecture. Still, a 0.6% gap is trivial, and for gaming-oriented tasks, the GTX 650 Ti Boost’s newer Kepler features likely provide a better experience despite the slight compute deficit.

Memory Subsystem

The GTX 650 Ti Boost comes equipped with 2 GB of GDDR5 memory, which is less than the 4 GB or 8 GB capacities that became standard in later generations, but it was ample for its 2013 launch context. The 192-bit memory bus is narrower than high-end cards of its era, yet the 144.2 GB/s bandwidth is sufficient to feed the 768 shading units at 1080p. Benchmark results show no memory-bound bottlenecks in the OpenCL and Vulkan tests, as the card’s scores are nearly identical to rivals with similar bandwidth.

For high resolutions, the data implies a clear limitation. At 1440p or above, the 2 GB VRAM capacity becomes a constraint, especially in texture-heavy modern titles that were released after this card’s end-of-life status. The 144.2 GB/s bandwidth also trails newer GPUs by a wide margin, so while the card can handle 1080p with medium settings, pushing to 4K would likely exceed both the frame buffer and the memory throughput, resulting in stuttering or texture pop-in. The 6 Gbps effective memory clock is fixed, and there is no indication of overclocking headroom in the benchmarks.

Ray Tracing and Feature Set

This card predates ray tracing hardware entirely, as its Kepler architecture lacks dedicated RT cores and tensor cores. The API support includes DirectX 12 (11_0), which means it can run DX12 titles but only at the feature level 11_0, not the full DX12 feature set. OpenGL 4.6 and Vulkan 1.2.175 are also listed, providing some modern API compatibility, but the absence of hardware-accelerated ray tracing means any such effects would be software-based, which is impractical given the 1.585 TFLOPS FP32 throughput.

The GTX 650 Ti Boost does support standard Kepler features like GPU Boost (the 1032 MHz boost clock vs 980 MHz base), but there are no tensor cores for DLSS or AI-based upscaling. For gamers, this means relying on traditional rendering techniques. The pixel rate of 16.51 GPixel/s and texture rate of 66.05 GTexel/s are modest figures, indicating that the card’s fill rate is adequate for older DirectX 11 games but will struggle with modern deferred rendering pipelines that demand higher pixel throughput.

Who Should Consider It

Given its 45th percentile ranking and an average benchmark score of 9,659, the GTX 650 Ti Boost is best suited for esports titles and older AAA games at 1080p with low-to-medium settings. The 2 GB VRAM and 144.2 GB/s bandwidth are enough for games from the 2013-2016 era, but not for recent releases that routinely require 6 GB or more. Benchmark results show it is on par with the GTX 960M and Radeon Pro WX 2100, so users upgrading from integrated graphics would see a substantial improvement, but those expecting high refresh rates at 1440p will be disappointed.

At 1080p, the card can achieve playable frame rates in competitive shooters like CS:GO or older titles, but its 16.51 GPixel/s pixel rate will cap performance in games with heavy post-processing. For productivity, the Vulkan score of 10,039 suggests decent compute capability for basic rendering or video encoding, though the lack of tensor and RT cores excludes modern AI workloads. This is a card for retro builds or budget secondary systems, not for current-generation gaming.

Benchmark Performance

The Geekbench OpenCL score of 9,278 and Vulkan score of 10,039 yield an average of 9,659, but the variance between the two tests is revealing. The Vulkan score is 8.2% higher than OpenCL (10,039 vs 9,278), indicating that the card’s Kepler architecture benefits from lower-overhead APIs. This is a meaningful insight: in Vulkan-based games or compute tasks, the GTX 650 Ti Boost punches above its weight, while in older OpenCL workloads it falls slightly behind.

Relative to rivals, the deltaPct values are all within ±0.6%, so there is no dominant winner. The GTX 650 Ti Boost is 0.3% faster than the Tesla M10 (9,659 vs 9,634), which is the only rival it beats outright. Against the GTX 960M (-0.1%), Radeon Pro WX 2100 (-0.2%), and Tesla C2070 (-0.6%), it loses by margins that are statistically insignificant. This suggests that at this performance tier, architectural differences matter less than driver optimization and thermal headroom.

The 45th percentile ranking means 55% of all GPUs in the database are faster, which is expected for a 2013 mid-range card. However, the fact that it trades blows with a 2015 mobile GPU (GTX 960M) and a 2017 workstation card (WX 2100) indicates that desktop Kepler aged gracefully in raw compute terms. The FP32 throughput of 1.585 TFLOPS is the key metric here, and it aligns closely with the scores, reinforcing that this card’s performance is compute-bound rather than memory-bound.

Power and Cooling

The GTX 650 Ti Boost has a TDP of 134 W, which is modest by modern standards but notable for its dual-slot cooler design. The card requires a single 6-pin PCIe power connector, and the suggested PSU rating is 300 W, making it compatible with older or lower-wattage power supplies. The 241 mm (9.5 inches) length is compact enough for most mid-tower cases, though dual-slot spacing is required.

The 28 nm process and 134 W TDP mean the card runs relatively cool under load, but the dual-slot cooler suggests it was designed to handle sustained boost clocks at 1032 MHz. There is no data on acoustic levels or thermal throttling, but the 66.05 GTexel/s texture rate indicates that the card can push geometry without excessive power draw. For a system with a 300 W PSU, this is a safe addition, but pairing it with a high-end CPU could exceed the recommended power budget.

FAQ

Q: How does the GTX 650 Ti Boost compare to the GTX 960M in synthetic benchmarks?

A: The GTX 650 Ti Boost scores 9,659 on average, which is 0.1% lower than the GTX 960M’s 9,670. The difference is negligible, so performance is effectively identical in OpenCL and Vulkan workloads.

Q: What is the memory bandwidth, and does it limit 4K gaming?

A: The card has 144.2 GB/s of bandwidth on a 192-bit bus. This is insufficient for 4K gaming, as the 2 GB VRAM and bandwidth would bottleneck modern titles, leading to significant frame drops.

Q: Does the card support hardware ray tracing?

A: No, the Kepler architecture has no RT cores or tensor cores. It only supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, so ray tracing is not possible in hardware.

Q: What is the launch MSRP of the GTX 650 Ti Boost?

A: The launch MSRP is 169 USD. This was a mid-range price point at release, but the card is now end-of-life.

Q: Is the 134 W TDP suitable for a small form factor build?

A: The TDP is 134 W with a recommended 300 W PSU and a single 6-pin connector. The 241 mm length and dual-slot profile may require careful case selection, but it is manageable for most mid-tower systems.

Q: Which API provides better performance on this card?

A: The Vulkan score of 10,039 is 8.2% higher than the OpenCL score of 9,278, indicating that the card performs better in Vulkan-based applications due to lower driver overhead.