NVIDIA GeForce GT 645M vs NVIDIA GeForce GTX 970M Comparison

The NVIDIA GeForce GTX 970M and the NVIDIA GeForce GT 645M represent two very different eras of mobile graphics, separated by a generational gap that shows up clearly in the benchmark data. The GTX 970M is a high-end part from the Maxwell 2.0 generation, while the GT 645M is an older, entry-level Kepler chip. Across the two shared benchmark tests, the GTX 970M wins decisively in both, but the GT 645M still holds a place in the database's lower tier. This analysis breaks down where each GPU wins, what architectural changes drove the performance gap, and how the specifications compare directly.

Where Each One Wins

The data is unambiguous in a head-to-head comparison: the GTX 970M wins both available shared benchmarks, claiming a 2–0 victory. The GT 645M has no wins in any direct comparison. However, looking at the broader benchmark suite, each card has its own profile of strengths that matters depending on the workload.

The GTX 970M's wins are massive in raw compute and modern API workloads. In Geekbench OpenCL, it scores 18,946 versus the GT 645M's 2,680, a delta of 606.9%. In Geekbench Vulkan, the GTX 970M scores 18,292 against 4,875, a 275.2% advantage. These are not marginal gains; the GTX 970M is in a completely different performance class for compute-heavy tasks and applications that leverage Vulkan. Beyond the head-to-head, the GTX 970M also has a robust set of additional benchmarks, including a Passmark G3D score of 5,704 and a Passmark GPU Compute score of 2,289, indicating strong DirectX and OpenCL performance across the board.

The GT 645M, while losing the shared tests, shows a capability in one specific area: its Geekbench Metal score of 5,679 is the only benchmark in its list that exceeds its OpenCL and Vulkan results. This suggests that on Apple's Metal API, the card performs comparatively better than on other compute interfaces. It is also worth remembering the GT 645M’s average benchmark score of 4,411 is close to the GTX 970M’s average of 4,628, but this is misleading because the averages are drawn from different test sets. The GT 645M only has three benchmarks listed, while the GTX 970M has ten, skewing the comparison. In practical terms, the GTX 970M is the clear winner for any modern gaming or compute workload, while the GT 645M is best suited for legacy DirectX 9 or 10-era titles, where its Passmark scores (99 in DX9 and 28 in DX10) are its only competitive figures.

Architecture Differences

The architectural gulf between these two GPUs is the primary reason for the performance disparity. The GTX 970M is built on the GM204 chip, using NVIDIA's Maxwell 2.0 architecture, while the GT 645M uses the GK107 chip with the older Kepler architecture. Both are manufactured on the same 28 nm process at TSMC, but that is where the similarity ends.

The transistor count tells a story of scale. The GTX 970M packs 5,200 million transistors into a 398 mm² die, resulting in a transistor density of 13.1M per mm². The GT 645M, in contrast, has just 1,270 million transistors on a 118 mm² die, with a density of 10.8M per mm². This means the GTX 970M has over four times the transistor budget, which translates directly into more execution resources. The GTX 970M features 1,280 shading units, 80 texture mapping units (TMUs), and 48 raster operation units (ROPs). The GT 645M is far leaner, with 384 shading units, 32 TMUs, and 16 ROPs. That is a 3.3x advantage in shaders, a 2.5x advantage in TMUs, and a 3x advantage in ROPs for the newer card.

Clock speeds also differ significantly. The GTX 970M has a base clock of 924 MHz and a boost clock of 1,038 MHz, while the GT 645M runs at a base of 709 MHz and a boost of 780 MHz. Combined with the higher shader count, this gives the GTX 970M a theoretical FP32 performance of 2.657 TFLOPS, compared to just 599.0 GFLOPS for the GT 645M. The memory subsystem is another major architectural split. The GTX 970M uses 6 GB of GDDR5 on a 192-bit bus, delivering 120.3 GB/s of bandwidth. The GT 645M uses 2 GB of DDR3 on a 128-bit bus, managing only 28.80 GB/s. This bandwidth difference is critical for texture-heavy scenes and high-resolution rendering. The GTX 970M also supports a newer API set, including DirectX 12 (12_1) and Vulkan 1.4, whereas the GT 645M tops out at DirectX 12 (11_0) and Vulkan 1.2.175.

Head-to-Head Benchmarks

The two shared benchmarks between these GPUs paint a stark picture of generational improvement. The first is Geekbench OpenCL, a compute test that measures raw throughput for general-purpose GPU tasks. Here, the GTX 970M scores 18,946, while the GT 645M manages only 2,680. The delta is 606.9%, meaning the GTX 970M is roughly seven times faster. This is not just a matter of higher clocks; it is the combination of more shaders, higher memory bandwidth, and a more efficient architecture that allows the Maxwell part to obliterate the older Kepler chip in compute workloads.

The second shared test is Geekbench Vulkan, which exercises the GPU under a modern low-level graphics API. The GTX 970M again dominates with a score of 18,292, against the GT 645M's 4,875. The delta here is 275.2%, meaning the GTX 970M is nearly four times faster. Vulkan is designed to reduce CPU overhead and better utilize multi-core processors, but it also rewards GPUs with more raw compute and bandwidth. The GTX 970M's 192-bit GDDR5 interface provides 120.3 GB/s of bandwidth, which is over four times the 28.80 GB/s available to the GT 645M. In real-world gaming, this translates to the ability to handle higher resolutions, more detailed textures, and higher frame rates without stuttering.

Looking at the broader benchmark data, the GTX 970M also shows strength in DirectX tests that the GT 645M does not have listed. Its Passmark G3D score of 5,704 is a solid mid-range result, and its Passmark DirectX 11 score of 42 indicates playable performance in older titles. The GT 645M's best DirectX result is a Passmark DirectX 9 score of 99, which is a legacy API that modern games no longer use. The GTX 970M scores 99 in the same DirectX 9 test, tying with the older card, but it moves ahead in every other category. The conclusion from the data is that the GTX 970M is not just faster; it is in a different performance tier, with the GT 645M only remaining competitive in decade-old workloads.

Specification Differences

The two cards differ in nearly every measurable specification, reflecting their different positions in the product stack and the years of architectural progress between them.

• Chip and Architecture: The GTX 970M uses the GM204 chip with Maxwell 2.0 architecture, while the GT 645M uses the GK107 chip with Kepler architecture.
• Transistors and Die Size: The GTX 970M has 5,200 million transistors on a 398 mm² die; the GT 645M has 1,270 million transistors on a 118 mm² die. Transistor density is 13.1M / mm² versus 10.8M / mm².
• Clock Speeds: The GTX 970M runs at a base of 924 MHz and a boost of 1,038 MHz. The GT 645M runs at a base of 709 MHz and a boost of 780 MHz. Memory clocks are 5 Gbps effective for the GTX 970M and 1800 Mbps effective for the GT 645M.
• Memory Configuration: The GTX 970M has 6 GB of GDDR5 on a 192-bit bus, yielding 120.3 GB/s bandwidth. The GT 645M has 2 GB of DDR3 on a 128-bit bus, yielding 28.80 GB/s bandwidth.
• Compute Units: The GTX 970M has 1,280 shading units, 80 TMUs, and 48 ROPs. The GT 645M has 384 shading units, 32 TMUs, and 16 ROPs.
• Performance Rates: The GTX 970M achieves 49.82 GPixel/s pixel rate and 83.04 GTexel/s texture rate. The GT 645M achieves 6.240 GPixel/s and 24.96 GTexel/s. FP32 performance is 2.657 TFLOPS versus 599.0 GFLOPS.
• Power and Interface: The GT 645M has a TDP of 32 W, while the GTX 970M has no listed TDP. The GTX 970M uses an MXM-B (3.0) bus interface, while the GT 645M uses PCIe 3.0 x16. Both have no power connectors and portable-device-dependent display outputs.
• API Support: The GTX 970M supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. The GT 645M supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175.
• Release Timing: The GTX 970M was released on October 6, 2014, while the GT 645M was released on September 30, 2012. The GTX 970M's predecessor is GeForce 800M and successor is GeForce 10 Mobile; the GT 645M's predecessor is GeForce 500M and successor is GeForce 700M.

FAQ

Q: Which GPU is faster in OpenCL compute workloads?

A: The GTX 970M is vastly faster. In Geekbench OpenCL, it scores 18,946 versus the GT 645M's 2,680, a delta of 606.9%.

Q: How does the memory bandwidth compare between the two cards?

A: The GTX 970M has 120.3 GB/s of bandwidth from 6 GB of GDDR5 on a 192-bit bus. The GT 645M has only 28.80 GB/s from 2 GB of DDR3 on a 128-bit bus.

Q: What are the architectural differences that explain the performance gap?

A: The GTX 970M uses the Maxwell 2.0 architecture on a 5,200-million-transistor GM204 chip, while the GT 645M uses the older Kepler architecture on a 1,270-million-transistor GK107 chip. This results in 1,280 shading units versus 384, and a 2.657 TFLOPS FP32 rate versus 599.0 GFLOPS.

Q: Does the GT 645M win any benchmark against the GTX 970M?

A: No. In the two shared head-to-head benchmarks, the GTX 970M wins both. The GT 645M has no wins in any direct comparison.

Q: What is the API support difference?

A: The GTX 970M supports DirectX 12 (12_1) and Vulkan 1.4, while the GT 645M supports DirectX 12 (11_0) and Vulkan 1.2.175. Both support OpenGL 4.6.

Q: How do the cards compare in Vulkan performance?

A: The GTX 970M scores 18,292 in Geekbench Vulkan, compared to the GT 645M's 4,875, giving the GTX 970M a 275.2% advantage.