NVIDIA GeForce GT 645M vs NVIDIA GeForce GTX 650 Comparison

The benchmark database presents a clear split between the NVIDIA GeForce GT 645M and the NVIDIA GeForce GTX 650. The GT 645M wins two of the three recorded head-to-head tests, including a dominant victory in Geekbench Metal, while the GTX 650 secures a substantial win in Geekbench OpenCL. However, the average benchmark scores tell a different story, with the GT 645M holding a 15.4% overall lead (4411 vs 3823). This suggests that the GT 645M is the stronger all-around performer in the recorded tests, despite the GTX 650's superior memory bandwidth and higher raw compute ratings.

Head-to-Head Benchmarks

The most striking result is in Geekbench Metal, where the GT 645M scores 5679 against the GTX 650's 2400. This represents a 136.6% advantage, the largest margin in any test. The GT 645M more than doubles the GTX 650's Metal score, indicating a significant difference in how the two GPUs handle Apple's Metal API. This is not a marginal win; it is a complete reversal of expectations given the GTX 650's higher theoretical performance.

In Geekbench Vulkan, the GT 645M again takes the lead, scoring 4875 versus 4524 for the GTX 650. The 7.8% delta is more modest, but it still shows the GT 645M outperforming its desktop counterpart in this cross-platform graphics test. The GTX 650's Vulkan score of 4524 is close, but the GT 645M's 4875 puts it ahead by a comfortable margin.

The GTX 650's only head-to-head victory comes in Geekbench OpenCL, where it scores 4545 against the GT 645M's 2680. This is a 41% deficit for the GT 645M, a massive swing in the opposite direction. The GTX 650 nearly doubles the GT 645M's OpenCL score, suggesting that the GTX 650's architecture is far better optimized for OpenCL workloads, or that its faster memory and higher clock rates pay off in this specific test.

When these results are averaged, the GT 645M's overall benchmark score of 4411 comfortably exceeds the GTX 650's 3823. The GTX 650's OpenCL win is not enough to overcome the GT 645M's dominance in Metal and its narrower but still decisive edge in Vulkan. The data shows the GT 645M as the winner in two out of three tests, with its Metal performance being the single biggest differentiator.

Architecture Differences

Both GPUs are built on NVIDIA's Kepler architecture and use the same 28 nm process node at TSMC. The similarities end there, as the chips themselves diverge significantly. The GT 645M uses the GK107 chip, which contains 1,270 million transistors on a 118 mm² die. The GTX 650 uses the larger GK106 chip, packing 2,540 million transistors into a 221 mm² die. This gives the GTX 650 a transistor density of 11.5M per mm², slightly higher than the GT 645M's 10.8M per mm².

Core counts are identical on paper: both GPUs feature 384 shading units, 32 texture mapping units, and 16 raster output units. The real difference lies in how those cores are fed. The GT 645M relies on 2 GB of DDR3 memory on a 128-bit bus, delivering 28.80 GB/s of bandwidth. The GTX 650 uses 1024 MB of GDDR5 memory on the same 128-bit bus, but its bandwidth jumps to 80.00 GB/s. That is nearly three times the memory bandwidth, a factor that directly impacts texture-heavy and high-resolution workloads.

Clock behavior also separates the two. The GT 645M has a base clock of 709 MHz and a boost clock of 780 MHz, while the GTX 650's base and boost clocks are not recorded in the database. The GTX 650's memory clock is listed at 1250 MHz with 5 Gbps effective speed, whereas the GT 645M's memory runs at 900 MHz with 1800 Mbps effective. These clock differences feed into the pixel and texture rates: the GTX 650 achieves 8.464 GPixel/s and 33.86 GTexel/s, while the GT 645M manages 6.240 GPixel/s and 24.96 GTexel/s. The GTX 650 also leads in FP32 throughput at 812.5 GFLOPS versus 599.0 GFLOPS.

Power and physical design are starkly different. The GT 645M is an integrated-class part with a 32 W TDP, IGP slot width, and no power connectors. The GTX 650 is a discrete card with a 65 W TDP, single-slot width, and requires a 1x 6-pin power connector, with a suggested PSU of 250 W. The GTX 650 measures 147 mm (5.8 inches) in length, while the GT 645M has no recorded dimensions as it is portable-device dependent. Both support PCIe 3.0 x16, DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, but the GT 645M's display outputs are portable-device dependent, while the GTX 650 offers 1x DVI and 2x DisplayPort 1.2.

FAQ

Q: Which GPU wins in Geekbench Metal, and by how much?

A: The NVIDIA GeForce GT 645M wins decisively, scoring 5679 against the GTX 650's 2400, a 136.6% advantage.

Q: Does the GTX 650 have any benchmark where it beats the GT 645M?

A: Yes, in Geekbench OpenCL, the GTX 650 scores 4545 while the GT 645M scores 2680, giving the GTX 650 a 41% lead.

Q: What is the average benchmark score difference between the two GPUs?

A: The GT 645M averages 4411 across all recorded tests, while the GTX 650 averages 3823. The GT 645M leads by 588 points, or 15.4%.

Q: How do the memory types and bandwidths compare?

A: The GT 645M uses 2 GB of DDR3 with 28.80 GB/s bandwidth, while the GTX 650 uses 1024 MB of GDDR5 with 80.00 GB/s bandwidth. The GTX 650 has nearly triple the memory bandwidth.

Q: Are the core counts the same on both GPUs?

A: Yes, both have 384 shading units, 32 TMUs, and 16 ROPs. The differences lie in clock speeds, memory, and chip size, not raw core counts.

Q: What is the power consumption difference?

A: The GT 645M has a 32 W TDP, while the GTX 650 has a 65 W TDP. The GTX 650 also requires a 6-pin power connector and a 250 W suggested PSU, while the GT 645M has none.

Specification Differences

The two GPUs differ in nearly every hardware specification except core counts and API support. The GT 645M uses the GK107 chip with 1,270 million transistors on a 118 mm² die, while the GTX 650 uses the GK106 chip with 2,540 million transistors on a 221 mm² die. Process node is identical at 28 nm TSMC, but transistor density is slightly higher on the GTX 650 at 11.5M / mm² versus 10.8M / mm².

Clock speeds diverge: the GT 645M has a base clock of 709 MHz and boost of 780 MHz, while the GTX 650's base and boost clocks are unrecorded. Memory clocks are 900 MHz (1800 Mbps effective) for the GT 645M and 1250 MHz (5 Gbps effective) for the GTX 650. Memory configuration differs in size and type: 2 GB DDR3 for the GT 645M versus 1024 MB GDDR5 for the GTX 650, with bandwidths of 28.80 GB/s and 80.00 GB/s respectively.

Pixel rate favors the GTX 650 at 8.464 GPixel/s versus 6.240 GPixel/s, as does texture rate at 33.86 GTexel/s versus 24.96 GTexel/s. FP32 performance is 812.5 GFLOPS for the GTX 650 and 599.0 GFLOPS for the GT 645M. TDP is 32 W for the GT 645M and 65 W for the GTX 650. Slot width is IGP for the GT 645M and single-slot for the GTX 650, with power connectors listed as "None" versus "1x 6-pin". The GTX 650 has a suggested PSU of 250 W, while the GT 645M has none. The GTX 650 has a recorded length of 147 mm (5.8 inches); the GT 645M has no dimensions. Display outputs are "Portable Device Dependent" for the GT 645M and "1x DVI, 2x DisplayPort 1.2" for the GTX 650. Release dates differ: the GT 645M launched on 2012-09-30, while the GTX 650 came later on 2013-11-26.

The Verdict

The data points to the NVIDIA GeForce GT 645M as the superior performer in the recorded benchmark suite. Its average score of 4411 places it in the 26th percentile of all GPUs, while the GTX 650's 3823 average sits in the 22nd percentile. The GT 645M wins two of three head-to-head tests, with its Metal and Vulkan scores leading by 136.6% and 7.8% respectively. The GTX 650's OpenCL win is notable, but it cannot compensate for the GT 645M's dominant Metal showing.

The GTX 650 does hold advantages in raw specifications: higher memory bandwidth, faster pixel and texture rates, and greater FP32 throughput. These numbers suggest the GTX 650 should be faster in many traditional gaming or compute workloads. Yet the benchmark results contradict that expectation, at least in the tests recorded here. The GT 645M's 2 GB memory capacity and higher average score make it the better overall choice for users prioritizing the measured APIs.

For those seeking the strongest single test result, the GT 645M's Metal score of 5679 is the standout figure. The GTX 650's best result is its OpenCL score of 4545, which still trails the GT 645M's Metal output by over 1000 points. The verdict is clear: the GT 645M is the more balanced and higher-scoring GPU in this comparison.

Where Each One Wins

The NVIDIA GeForce GT 645M wins in Geekbench Metal and Geekbench Vulkan, making it the preferred option for workloads that rely on these APIs. Its Metal score of 5679 is exceptional, and its Vulkan score of 4875 leads the GTX 650 by 351 points. Users running Metal-based applications, such as certain macOS or iOS development tools, will see a clear benefit from the GT 645M. Its 2 GB memory capacity also provides an edge in memory-hungry tasks, even if the slower DDR3 bandwidth limits sustained throughput.

The NVIDIA GeForce GTX 650 wins in Geekbench OpenCL, scoring 4545 against the GT 645M's 2680. This 41% margin makes it the clear choice for OpenCL compute workloads, which are common in scientific simulations, video encoding, and some rendering tasks. The GTX 650's GDDR5 memory and 80.00 GB/s bandwidth give it a massive data throughput advantage, and its higher pixel and texture rates (8.464 GPixel/s and 33.86 GTexel/s) support this. For users who know their applications rely heavily on OpenCL, the GTX 650 is the better fit.

The GT 645M also wins on efficiency, with a 32 W TDP versus the GTX 650's 65 W. This makes it suitable for portable devices and low-power systems. The GTX 650 requires a discrete power connector and a 250 W PSU, limiting its use to desktop builds with adequate power delivery. In summary, the GT 645M excels in Metal and Vulkan, while the GTX 650 dominates OpenCL and offers faster memory and higher throughput metrics. The choice depends entirely on which API and workload matters most.