NVIDIA GeForce GT 740M vs NVIDIA GeForce GTX 650 Comparison

The NVIDIA GeForce GTX 650 and NVIDIA GeForce GT 740M are both end-of-life Kepler parts from 2013, yet they serve fundamentally different purposes: the GTX 650 is a single-slot desktop card, while the GT 740M is a mobile MXM module. Benchmark results show a clear hierarchy, but the margins vary significantly depending on the API tested, and the GT 740M’s lower power consumption tells a different story than its raw performance deficit.

Head-to-Head Benchmarks

The GTX 650 wins both available head-to-head comparisons, but the degree of dominance is not uniform. In Geekbench OpenCL, the GTX 650 scores 4545 against the GT 740M’s 3974, a 14.4% lead. That is a solid, consistent advantage that reflects the desktop card’s superior memory bandwidth and full PCIe 3.0 x16 interface. The GT 740M is not embarrassed here; its score is within striking distance, and in real-world compute tasks that scale well across shading units, the gap would be noticeable but not crushing.

The Vulkan result is where the GTX 650 pulls away decisively. It scores 4524 versus the GT 740M’s 3459, a 30.8% margin. This is a massive difference, nearly double the OpenCL gap. The GTX 650’s score in Vulkan is almost identical to its OpenCL score, suggesting it handles the API efficiently. The GT 740M, by contrast, drops 13% from its OpenCL result when running Vulkan, indicating that its narrower 64-bit memory bus and cut-down GK208 chip become a bottleneck under Vulkan’s lower-level driver overhead. If you are targeting modern Vulkan-based games or applications, the GTX 650 is the only serious choice between these two.

Looking at the broader benchmark context, the GTX 650’s average benchmark score is 3823, placing it in the 22nd percentile of all GPUs. Its nearest rival is the NVIDIA GeForce MX110, which scores 3834, a mere 0.3% higher. That is effectively a tie. The GT 740M averages 3717, also in the 22nd percentile, with its closest competitor being the NVIDIA Quadro 3000M at 3718, a 0% delta. Both cards sit in the same performance tier, but the GTX 650 sits at the top of that tier while the GT 740M sits near the bottom. The difference between their averages is 106 points, about 2.9%, which is meaningful enough to tip the scales in the desktop card’s favor for any compute workload.

Architecture Differences

Both GPUs are built on TSMC’s 28 nm process, but they are fundamentally different dies. The GTX 650 uses the GK106 chip, a 2,540-million-transistor design measuring 221 mm². The GT 740M uses the GK208 chip, which is dramatically smaller at 1,020 million transistors and 87 mm². Transistor density is nearly identical — 11.5M per mm² for GK106 versus 11.7M per mm² for GK208 — so the size difference is purely a matter of how much silicon each chip carries.

The architectural split is telling. Both have 384 shading units and 32 texture mapping units, so pixel-pushing and texturing throughput are nearly equal on paper. The GTX 650’s pixel rate is 8.464 GPixel/s versus 8.264 GPixel/s for the GT 740M, and its texture rate is 33.86 GTexel/s versus 33.06 GTexel/s. These are marginal differences. The real divergence comes in memory and ROPs. The GTX 650 has 16 ROPs, twice the GT 740M’s 8 ROPs, and it uses a 128-bit GDDR5 interface with 1024 MB of memory. The GT 740M has a 64-bit DDR3 interface with 2 GB. Memory bandwidth tells the whole story: 80.00 GB/s for the GTX 650 versus 14.40 GB/s for the GT 740M — a 5.5x disparity. That bandwidth gap explains why the GTX 650 wins OpenCL by 14.4% and Vulkan by 30.8% despite nearly identical shader counts; the GT 740M simply starves its cores.

Clock speeds are another differentiator, though not in the GTX 650’s favor. The GT 740M has a base clock of 980 MHz and a boost clock of 1033 MHz, while the GTX 650’s base and boost clocks are not listed in the data. The GT 740M’s memory runs at 900 MHz (1800 Mbps effective), while the GTX 650’s memory runs at 1250 MHz (5 Gbps effective). The GTX 650 compensates for any clock deficit with vastly more effective memory speed and double the ROPs. The GT 740M’s FP32 throughput is 793.3 GFLOPS, nearly matching the GTX 650’s 812.5 GFLOPS, but that theoretical parity never translates to real-world results because of the memory bottleneck.

The Verdict

The data is unambiguous: the GTX 650 is the superior performer in every measured benchmark. It wins both head-to-head tests, has a higher average benchmark score (3823 versus 3717), and holds its own against modern low-end parts like the MX110, while the GT 740M is barely distinguishable from the Quadro 3000M. For anyone building a desktop system or upgrading an older rig, the GTX 650 is the pick. Its 80 GB/s of bandwidth and 16 ROPs make it viable for older titles and light compute tasks, and its Vulkan performance is strong enough to handle modern API overhead.

The GT 740M is not without merit, but its merits are not performance-based. It draws only 33 W, exactly half the GTX 650’s 65 W TDP, and requires no external power connector — it runs entirely off the motherboard slot. It is an MXM module designed for laptops, so it is not a purchase decision but a spec sheet comparison. If you are comparing a used laptop with a GT 740M versus a desktop with a GTX 650, the desktop wins on every performance metric. If you are stuck with the GT 740M, the data shows it will handle OpenCL workloads at roughly 87% of the GTX 650’s level, but Vulkan performance falls to 76%. Neither result is terrible for a 2013 mobile chip, but neither is competitive with the desktop card.

Specification Differences

The two cards differ in nearly every major specification except for the core compute configuration. Both use 28 nm TSMC silicon, 384 shading units, and 32 TMUs, and both support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The differences are:

• Chip and die: GK106 (2,540M transistors, 221 mm²) versus GK208 (1,020M transistors, 87 mm²)
• Memory: 1024 MB GDDR5 on a 128-bit bus versus 2 GB DDR3 on a 64-bit bus
• Bandwidth: 80.00 GB/s versus 14.40 GB/s
• ROPs: 16 versus 8
• Clocks: Memory at 1250 MHz (5 Gbps effective) versus base 980 MHz, boost 1033 MHz, memory at 900 MHz (1800 Mbps effective)
• Pixel/texture rates: 8.464 GPixel/s and 33.86 GTexel/s versus 8.264 GPixel/s and 33.06 GTexel/s
• FP32: 812.5 GFLOPS versus 793.3 GFLOPS
• TDP and power: 65 W with 1x 6-pin connector and 250 W suggested PSU versus 33 W with no connector and no suggested PSU
• Form factor: Single-slot, 147 mm (5.8 inches) versus MXM Module
• Bus interface: PCIe 3.0 x16 versus PCIe 3.0 x8
• Display outputs: 1x DVI, 2x DisplayPort 1.2 versus portable-device-dependent
• Release date: November 26, 2013 versus June 19, 2013
• Successor: GeForce 700 versus GeForce 800M

FAQ

Q: Which card has higher raw compute throughput?

A: The GTX 650 has 812.5 GFLOPS FP32 performance versus 793.3 GFLOPS for the GT 740M, a difference of about 2.4%. This is a minor edge, not a deciding factor.

Q: Why does the GTX 650 win Vulkan benchmarks by so much more than OpenCL?

A: The GTX 650 scores 4524 in Vulkan versus 3459 for the GT 740M, a 30.8% gap. In OpenCL the gap is only 14.4% (4545 versus 3974). The GT 740M’s 64-bit memory bus and 14.40 GB/s bandwidth likely choke under Vulkan’s lower-level API requirements.

Q: Is the GT 740M more power efficient?

A: Yes. The GT 740M has a 33 W TDP, less than half the GTX 650’s 65 W, and it requires no external power connector. The GTX 650 needs a 1x 6-pin connector and a 250 W suggested PSU.

Q: Do both cards support the same APIs?

A: Yes. Both support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. There is no API advantage for either card.

Q: How do these cards compare to their nearest rivals?

A: The GTX 650’s average score of 3823 is within 0.3% of the MX110’s 3834 and 0.8% above the Intel UHD Graphics 710’s 3792. The GT 740M’s 3717 average is exactly matched by the Quadro 3000M at 3718, and it is 0.6% above the GeForce 825M’s 3694.

Q: Which card has more memory, and does it matter?

A: The GT 740M has 2 GB of DDR3, double the GTX 650’s 1 GB of GDDR5. It does not matter in practice because the GT 740M’s bandwidth is 14.40 GB/s versus 80.00 GB/s, a 5.5x deficit that overwhelms any capacity advantage.

Where Each One Wins

The GTX 650 wins every benchmark category where data exists. In OpenCL, its 14.4% lead means it is the clear choice for compute workloads like rendering, physics simulation, or any task that offloads parallel math to the GPU. The Vulkan result is even more lopsided; a 30.8% margin makes the GTX 650 the only viable option for modern Vulkan-based games or applications. Its 80 GB/s bandwidth and 16 ROPs also give it a theoretical edge in any fill-rate-limited scenario, even though the pixel rate difference is small (8.464 versus 8.264 GPixel/s). The GTX 650’s PCIe 3.0 x16 interface provides full bandwidth to the host system, which matters for data transfer-heavy workloads.

The GT 740M’s only genuine wins are in power efficiency and package size. At 33 W, it can run in a thin laptop with no dedicated power connector, while the GTX 650 demands a 6-pin cable and a 250 W PSU. The GT 740M’s 2 GB memory capacity is also double, which can help in scenarios with large textures that fit within 2 GB but not 1 GB — though the drastically lower bandwidth means those textures will load slowly. For mobility, the GT 740M is the only option of the two, as the GTX 650 is a 147 mm desktop card that requires a PCIe slot and tower chassis. There are no benchmark wins for the GT 740M, but its existence is justified by its ability to run without active cooling or high power draw in a portable chassis.