NVIDIA GeForce GT 640 vs NVIDIA GeForce GT 740M Comparison

The NVIDIA GeForce GT 640 and NVIDIA GeForce GT 740M are both end-of-life Kepler-based parts, but they target different segments: the GT 640 is a desktop card, while the GT 740M is a mobile module. The benchmark data shows a near-total split in performance character, with each GPU taking one of the two available head-to-head tests. The GT 740M wins in OpenCL compute, while the GT 640 counters with a decisive victory in Vulkan graphics. Their average benchmark scores differ by roughly 15.8%, favoring the mobile part, yet the architectural decisions behind each chip explain why the desktop card still holds a lead in a modern graphics API.

Where Each One Wins

The GT 740M is the clear winner in raw compute throughput. Its OpenCL score of 3974 beats the GT 640’s 3736 by 6%, and its average benchmark score of 3717 sits at the 22nd percentile of all GPUs, compared to the GT 640’s 20th percentile and 3210 average. This makes the GT 740M the better choice for general-purpose compute workloads, such as OpenCL-accelerated rendering or physics simulations, where raw shader output matters more than memory bandwidth or ROP throughput.

The GT 640, conversely, wins decisively in the Vulkan graphics test. Its score of 3809 outperforms the GT 740M’s 3459 by 10.1%, a substantial margin that indicates better driver efficiency or hardware scheduling in that API. This suggests the GT 640 is the stronger option for modern game titles that leverage Vulkan, as well as for any workload that relies on the graphics pipeline rather than pure compute. The GT 640’s single head-to-head win in Vulkan is its only benchmark victory, but it is a meaningful one given Vulkan’s growing adoption.

For users who need a mix of both, the data points to a trade-off: the GT 740M offers higher peak compute and a better overall average score, while the GT 640 delivers superior graphics API performance. The GT 640 also benefits from a wider 128-bit memory bus, which translates to 28.51 GB/s of bandwidth versus the GT 740M’s 14.40 GB/s, a factor that likely contributes to its Vulkan advantage. In practical terms, the GT 740M suits compute-focused laptops, while the GT 640 fits desktops where Vulkan gaming or bandwidth-sensitive tasks take priority.

Architecture Differences

Both GPUs are built on TSMC’s 28 nm process, but they diverge in chip design. The GT 640 uses the GK107 chip with 1,270 million transistors on a 118 mm² die, yielding a transistor density of 10.8M per mm². The GT 740M uses the smaller GK208 chip, packing 1,020 million transistors into 87 mm², which raises density to 11.7M per mm². Despite the smaller transistor count, the GT 740M achieves higher clock speeds: its base clock is 980 MHz with a boost of 1033 MHz, while the GT 640 has no listed base or boost clocks, only a memory clock of 891 MHz (1782 Mbps effective).

The two chips share identical shader resources—384 shading units and 32 texture mapping units—but differ in ROPs. The GT 640 has 16 ROPs, while the GT 740M has only 8. This halving of ROPs is offset by the GT 740M’s higher clocks, which push its pixel rate to 8.264 GPixel/s versus the GT 640’s 7.216 GPixel/s. Texture rate also favors the GT 740M at 33.06 GTexel/s compared to 28.86 GTexel/s, and FP32 throughput is 793.3 GFLOPS versus 692.7 GFLOPS, a 14.5% advantage.

Memory configurations are starkly different. The GT 640 uses a 128-bit bus with 2 GB DDR3, delivering 28.51 GB/s. The GT 740M uses a 64-bit bus with the same 2 GB DDR3 capacity, but bandwidth drops to just 14.40 GB/s—less than half. This bandwidth gap is critical for memory-intensive workloads, and it explains why the GT 640 can win in Vulkan despite lower raw compute. The GT 740M compensates with a much lower TDP of 33 W versus 65 W for the GT 640, and it runs on a PCIe 3.0 x8 interface rather than the GT 640’s x16. Both support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, so API compatibility is identical.

Head-to-Head Benchmarks

The two available head-to-head tests reveal a clear split. In Geekbench OpenCL, the GT 740M scores 3974 against the GT 640’s 3736, a 6% delta that favors the mobile chip. This is consistent with its higher FP32 throughput (793.3 GFLOPS vs 692.7 GFLOPS) and faster texture rate. The GT 740M’s boost clock of 1033 MHz, applied across the same 384 shading units, gives it a compute edge that the GT 640 cannot match despite its wider memory bus.

In Geekbench Vulkan, the results flip dramatically. The GT 640 scores 3809, while the GT 740M falls to 3459, giving the desktop card a 10.1% victory. This is a larger margin than the GT 740M’s OpenCL win, and it points to the GT 640’s 16 ROPs and 28.51 GB/s bandwidth as decisive factors. Vulkan workloads often stress memory access and pixel throughput, areas where the GT 640’s architecture is clearly superior. The GT 740M’s 8 ROPs and 14.40 GB/s bandwidth appear to bottleneck its Vulkan performance, even with higher clocks.

The average benchmark scores reinforce this pattern. The GT 740M averages 3717, which is 15.8% higher than the GT 640’s 3210, but this average is skewed by the inclusion of the OpenCL test where the GT 740M excels. The GT 640’s nearest rivals include the Intel Arc Pro B60 (3182, 0.9% delta) and NVIDIA Quadro P1000 (3163, 1.5% delta), placing it in a cluster of similarly performing parts. The GT 740M, meanwhile, sits near the NVIDIA Quadro 3000M (3718, 0% delta) and AMD Radeon HD 6770 (3649, 1.9% delta), showing it competes at a higher performance tier overall.

FAQ

Q: Which GPU has higher raw compute performance?

A: The GT 740M, with 793.3 GFLOPS FP32 output compared to the GT 640’s 692.7 GFLOPS. This is reflected in its OpenCL win of 3974 vs 3736, a 6% delta.

Q: Why does the GT 640 win in Vulkan despite lower compute?

A: The GT 640 has 16 ROPs versus the GT 740M’s 8, and a memory bandwidth of 28.51 GB/s against 14.40 GB/s. These factors give it a 10.1% edge in the Geekbench Vulkan test (3809 vs 3459).

Q: Are the memory configurations the same?

A: No. Both have 2 GB DDR3, but the GT 640 uses a 128-bit bus, while the GT 740M uses a 64-bit bus. This results in nearly double the bandwidth for the GT 640 (28.51 GB/s vs 14.40 GB/s).

Q: Which GPU has better power efficiency?

A: The GT 740M, with a TDP of 33 W versus the GT 640’s 65 W. This makes the GT 740M suitable for portable devices, as indicated by its MXM module form factor.

Q: How do their average benchmark scores compare?

A: The GT 740M averages 3717, placing it at the 22nd percentile of all GPUs, while the GT 640 averages 3210 at the 20th percentile. The GT 740M’s average is 15.8% higher.

Q: Do they support the same graphics APIs?

A: Yes. Both support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, so API-level compatibility is identical.

The Verdict

The data presents a clear choice based on workload. For compute-intensive tasks that rely on OpenCL, the GT 740M is the superior option. Its higher FP32 throughput (793.3 GFLOPS), faster texture rate (33.06 GTexel/s), and 6% OpenCL benchmark win make it the stronger candidate for GPGPU applications. Its lower 33 W TDP also makes it viable for thin-and-light laptops, where the GT 640’s 65 W desktop requirement would be impractical.

For graphics-centric workloads, particularly those using Vulkan, the GT 640 is the better pick. Its 10.1% Vulkan victory stems from a doubling of ROPs (16 vs 8) and a memory bus twice as wide (128-bit vs 64-bit), which provide 28.51 GB/s of bandwidth—critical for modern rendering. The GT 640’s higher pixel rate (7.216 GPixel/s) and superior memory subsystem make it more capable for gaming, despite its lower compute scores.

The average benchmark scores favor the GT 740M (3717 vs 3210), but this figure is heavily influenced by the OpenCL test where it excels. Users should ignore the aggregate and focus on the specific API they intend to use. The GT 640 is a desktop part with a launch MSRP of 99 USD, while the GT 740M has no listed price, reflecting its OEM mobile nature. Ultimately, the GT 740M wins on compute and efficiency, while the GT 640 wins on graphics API performance and memory bandwidth. Choose based on the workload, not the average.