GPU Comparison

The NVIDIA GeForce GT 645M and the NVIDIA GeForce 940MX are two mobile graphics solutions from different eras of NVIDIA’s laptop lineup. The GT 645M is a Kepler-generation part from the GeForce 600M series, while the 940MX is a Maxwell-based chip from the GeForce 900M family. The benchmark data reveals a clear split: the 940MX dominates in OpenCL compute workloads, while the older GT 645M edges out a narrow victory in Vulkan graphics performance. This page breaks down those results and the architectural choices that explain them.

Head-to-Head Benchmarks

The two GPUs were tested head-to-head in two distinct workloads, and the results show a dramatic divergence in their strengths. In the Geekbench OpenCL test, the NVIDIA GeForce 940MX delivers a score of 4939, which is a massive 45.7% higher than the GT 645M’s 2680. This is not a marginal win; it is a near-doubling of compute performance. The deltaPct figure of -45.7% (expressed from the perspective of the lower-scoring GT 645M) underscores the scale of the 940MX’s advantage in general-purpose compute tasks. For any workload that leverages OpenCL, such as video encoding, physics simulations, or data processing, the 940MX is in a different league.

However, the story flips when the focus shifts to Vulkan graphics. In the Geekbench Vulkan test, the NVIDIA GeForce GT 645M scores 4875, narrowly beating the 940MX’s 4749. The delta here is just 2.7%, which indicates a statistical tie in real-world terms. The GT 645M’s win is notable but slim, suggesting that the two architectures handle Vulkan’s low-level API demands in a surprisingly similar fashion. The data shows that while the 940MX is the superior compute part, the GT 645M can still hold its own in a modern graphics API.

The overall win count is evenly split at one win apiece. Yet the magnitude of those wins is far from equal. The 940MX’s OpenCL victory is a landslide, while the GT 645M’s Vulkan victory is a photo finish. Looking at the average benchmark scores across all tested workloads, the 940MX posts 4844, which is 9.8% higher than the GT 645M’s 4411. This aggregate figure suggests that the 940MX is the more capable overall GPU, despite the GT 645M’s single-test triumph.

Architecture Differences

The architectural divide between these two GPUs explains much of the performance gap. The GT 645M is built on the GK107 chip using the Kepler architecture, manufactured on a 28 nm process at TSMC. It packs 1,270 million transistors into a die size of 118 mm², yielding a transistor density of 10.8 million transistors per square millimeter. The 940MX, in contrast, uses the GM107 chip based on the newer Maxwell architecture. It also uses a 28 nm process at TSMC, but the die is larger at 148 mm² and holds significantly more hardware, 1,870 million transistors, for a density of 12.6M / mm². This 47% increase in transistor count is a fundamental reason why the 940MX achieves higher raw compute throughput.

The memory subsystems are also fundamentally different. The GT 645M uses 2 GB of DDR3 memory on a 128-bit bus, delivering a bandwidth of 28.80 GB/s. The 940MX uses 2 GB of GDDR5 memory on a narrower 64-bit bus, yet its bandwidth is higher at 40.10 GB/s. The switch from DDR3 to GDDR5 is a key architectural upgrade; despite halving the bus width, the faster memory clock yields a 39% bandwidth advantage for the 940MX. This bandwidth increase is critical for compute-heavy tasks that constantly feed data to the shader cores.

Clock speeds also favor the 940MX. Its base clock is 795 MHz with a boost of 861 MHz, compared to the GT 645M’s 709 MHz base and 780 MHz boost. The memory clocks diverge even more sharply: the 940MX runs at 1253 MHz (5 Gbps effective), while the GT 645M runs at 900 MHz (1800 Mbps effective). These higher clocks, combined with the larger transistor budget, allow the 940MX to reach 881.7 GFLOPS of FP32 performance, versus 599.0 GFLOPS for the GT 645M. That is a 47% improvement in raw floating-point throughput.

The shading hardware also differs. The 940MX has 512 shading units and 32 TMUs, but only 8 ROPs. The GT 645M has 384 shading units, 32 TMUs, and 16 ROPs. The 940MX has a 33% advantage in shading units, which directly fuels its OpenCL dominance. However, the GT 645M’s double the ROP count (16 vs. 8) helps explain why it remains competitive in Vulkan rasterization tests, more ROPs can mean better fill-rate efficiency in certain rendering scenarios. The pixel rate of the GT 645M is 6.240 GPixel/s, and its texture rate is 24.96 GTexel/s; the 940MX counters with 6.888 GPixel/s and 27.55 GTexel/s, respectively.

Where Each One Wins

The benchmark results point to a clear use-case split. The NVIDIA GeForce 940MX is the definitive winner for compute-oriented workloads. Its OpenCL score of 4939 versus 2680 means that any application leveraging GPU compute, be it for image processing, machine learning inference, or scientific calculations, will run dramatically faster on the 940MX. The higher FP32 throughput (881.7 GFLOPS) and superior memory bandwidth (40.10 GB/s) make it the logical choice for tasks that stress the shader array and memory bus.

The NVIDIA GeForce GT 645M finds its niche in specific graphics API scenarios. Its Vulkan score of 4875 slightly edges out the 940MX’s 4749, suggesting that in Vulkan-based games or applications, the two are effectively interchangeable, with a marginal nod to the older card. The GT 645M’s higher ROP count and wider 128-bit memory bus may contribute to this parity in rasterization-heavy tasks, even though its overall bandwidth is lower. For users running legacy DirectX 11 titles or Vulkan workloads, the GT 645M remains surprisingly viable.

Beyond raw scores, the 940MX also offers better efficiency on paper. Its TDP is 23 W, compared to 32 W for the GT 645M. This means the 940MX delivers higher performance while drawing less power, a critical factor in laptops where thermals and battery life are paramount. The 940MX also supports a newer Vulkan version (1.4) compared to the GT 645M’s Vulkan 1.2.175, which could offer better driver-level optimization for future titles.

Specification Differences

The two GPUs diverge across several key specification fields. The chip and architecture differ entirely: GK107 with Kepler versus GM107 with Maxwell. The transistor count jumps from 1,270 million to 1,870 million, and the die size grows from 118 mm² to 148 mm². Transistor density increases from 10.8M / mm² to 12.6M / mm².

Clock speeds are uniformly higher on the 940MX: base clock is 795 MHz vs. 709 MHz, boost clock is 861 MHz vs. 780 MHz, and memory clock is 1253 MHz (5 Gbps effective) vs. 900 MHz (1800 Mbps effective). The memory type changes from DDR3 to GDDR5, and while the bus width narrows from 128-bit to 64-bit, the bandwidth rises from 28.80 GB/s to 40.10 GB/s.

The shading unit count increases from 384 to 512, while TMUs stay constant at 32. The ROP count drops from 16 to 8. Pixel rate rises from 6.240 GPixel/s to 6.888 GPixel/s, and texture rate goes from 24.96 GTexel/s to 27.55 GTexel/s. FP32 performance jumps from 599.0 GFLOPS to 881.7 GFLOPS.

Power draw is significantly lower on the 940MX at 23 W versus 32 W. The slot width changes from IGP to MXM Module, and the bus interface narrows from PCIe 3.0 x16 to PCIe 3.0 x8. The 940MX supports Vulkan 1.4, while the GT 645M is limited to 1.2.175. Both share the same DirectX 12 (11_0) and OpenGL 4.6 support, and both have 2 GB of memory and no power connectors.

FAQ

Q: Which GPU is faster in OpenCL compute workloads?

A: The NVIDIA GeForce 940MX is substantially faster. Its Geekbench OpenCL score is 4939, which is 45.7% higher than the GT 645M’s 2680.

Q: Does the older GT 645M win any benchmark?

A: Yes. The GT 645M wins the Geekbench Vulkan test with a score of 4875, edging out the 940MX’s 4749 by a 2.7% margin.

Q: What is the overall average benchmark score for each GPU?

A: The 940MX has an average benchmark score of 4844, while the GT 645M averages 4411. This gives the 940MX a 9.8% higher aggregate score.

Q: How do the memory configurations compare?

A: Both have 2 GB of memory, but the GT 645M uses DDR3 on a 128-bit bus with 28.80 GB/s bandwidth, while the 940MX uses GDDR5 on a 64-bit bus with 40.10 GB/s bandwidth.

Q: Which GPU has more shading units and higher FP32 performance?

A: The 940MX has 512 shading units and delivers 881.7 GFLOPS, compared to the GT 645M’s 384 shading units and 599.0 GFLOPS.

Q: What are the TDP differences between the two?

A: The 940MX has a TDP of 23 W, which is lower than the GT 645M’s 32 W, indicating better power efficiency for the newer part.

The Verdict

The data paints a clear picture for most users: the NVIDIA GeForce 940MX is the superior GPU. It wins the only major compute benchmark by a landslide (45.7% in OpenCL), posts a higher average benchmark score (4844 vs. 4411), and does so while drawing 9 W less power. Its higher transistor count, faster GDDR5 memory, and greater FP32 throughput make it the rational choice for anyone prioritizing compute performance or modern workloads.

The NVIDIA GeForce GT 645M is not without merit, though. Its Vulkan score of 4875 shows that in one specific graphics API, it can outperform the 940MX, albeit by a narrow 2.7%. This suggests that for users running Vulkan-based applications, the GT 645M is a viable option, especially given its wider 128-bit memory bus and double the ROP count. However, these advantages do not translate into broader wins.

The choice ultimately depends on the workload. For compute-heavy tasks, video processing, or any OpenCL-accelerated application, the 940MX is the definitive pick, its 45.7% lead is too large to ignore. For users locked into Vulkan-only workloads where every frame counts, the GT 645M’s marginal edge could be relevant, but the 940MX’s overall efficiency and higher average score make it the safer recommendation for a general-purpose laptop GPU. The benchmark data strongly favors the 940MX as the more capable and future-proof option.