GPU Comparison

The NVIDIA GeForce GT 640 and NVIDIA GeForce 820A are both end-of-life mobile and desktop graphics solutions from NVIDIA, yet they occupy different performance tiers. The data shows the GT 640 is the clear winner, delivering a 25.2% higher score in the only shared benchmark, Geekbench OpenCL. The GT 640 is the choice for users who need a discrete, single-slot card with more memory and a wider feature set, while the 820A is a low-power integrated solution for portable devices where efficiency and compactness outweigh raw performance.

The Verdict

Based strictly on the benchmark data, the NVIDIA GeForce GT 640 is the superior performer. In the Geekbench OpenCL test, the GT 640 scores 3736, which is 25.2% higher than the 820A’s 2983. This is a decisive margin that places the GT 640 in a different performance class. The GT 640 also holds a higher overall percentile rank, sitting at the 20th percentile of all GPUs, while the 820A is just below at the 19th percentile. For any user prioritizing compute performance or gaming, the GT 640 is the only viable option between the two.

The 820A, however, is not without its merits. Its 15 W TDP is drastically lower than the GT 640’s 65 W, and it is classified as an IGP (integrated graphics processor), meaning it requires no slot width and has no power connectors. This makes it suitable for thin, portable devices where the GT 640’s 145 mm length and single-slot footprint would be impractical. The verdict is simple: choose the GT 640 for performance, choose the 820A for ultra-low-power integration.

Architecture Differences

The two GPUs are built on fundamentally different architectures. The GT 640 uses the GK107 chip based on the Kepler architecture, while the 820A uses the GF117 chip based on the older Fermi 2.0 architecture. Both are manufactured on a 28 nm process at TSMC, but the similarities end there. The Kepler chip is significantly more complex, packing 1,270 million transistors on a 118 mm² die, resulting in a transistor density of 10.8M / mm². In contrast, the Fermi 2.0 chip has only 585 million transistors on a 116 mm² die, yielding a density of 5.0M / mm².

This architectural gap translates directly into compute resources. The GT 640 features 384 shading units, 32 texture mapping units (TMUs), and 16 raster output units (ROPs). The 820A has just 96 shading units, 16 TMUs, and 8 ROPs. Consequently, the GT 640’s pixel rate is 7.216 GPixel/s and its texture rate is 28.86 GTexel/s, compared to the 820A’s 2.500 GPixel/s and 10.00 GTexel/s. The FP32 floating-point performance also reflects this, with the GT 640 delivering 692.7 GFLOPS versus the 820A’s 240.0 GFLOPS.

Memory configurations also diverge sharply. The GT 640 comes with 2 GB of DDR3 memory on a 128-bit bus, providing a bandwidth of 28.51 GB/s. The 820A is limited to 1024 MB of DDR3 on a 64-bit bus, resulting in just 16.02 GB/s of bandwidth. The memory clock is also higher on the GT 640 at 891 MHz (1782 Mbps effective) versus the 820A’s 1001 MHz (2 Gbps effective), though the wider bus gives the GT 640 the overall advantage.

Connectivity and API support further differentiate them. The GT 640 uses a PCIe 3.0 x16 interface, while the 820A uses the older PCIe 2.0 x16. Both support DirectX 12 (11_0) and OpenGL 4.6, but the GT 640 also supports Vulkan 1.2.175, whereas the 820A has no Vulkan support listed. The GT 640 offers discrete display outputs (1x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2), while the 820A’s outputs are described as "Portable Device Dependent."

Head-to-Head Benchmarks

The only direct comparison available is the Geekbench OpenCL test, and the results are unambiguous. The GT 640 scores 3736, while the 820A scores 2983. The delta percentage is 25.2%, meaning the GT 640 is roughly a quarter faster than the 820A in this compute workload. This is a substantial gap that underscores the difference in shading units, memory bandwidth, and overall architecture efficiency.

Looking at the nearest rivals provides context for these scores. The GT 640’s average benchmark score is 3210, placing it near the NVIDIA Quadro P1000 (3163, 1.5% slower) and slightly below the NVIDIA GeForce 920M (3287, -2.3% delta). The 820A, with an average score of 2983, sits close to the NVIDIA GeForce GTX 860M (2967, 0.5% slower) and the NVIDIA GeForce GTX 750 Ti (2953, 1% slower). Interestingly, the 820A’s score is also within 2.4% of the NVIDIA GeForce RTX 4060 Ti 8 GB (2913), though this is likely an anomaly in the benchmark dataset given the massive architectural differences.

The GT 640 also has additional benchmark data points that the 820A lacks. It scores 2086 in Geekbench Metal and 3809 in Geekbench Vulkan, indicating solid performance across multiple compute APIs. The 820A has no such scores, meaning the GT 640 offers a more versatile benchmark profile for users who rely on these specific APIs.

FAQ

Q: Which GPU is faster in the Geekbench OpenCL test?

A: The NVIDIA GeForce GT 640 is significantly faster, scoring 3736 compared to the 820A’s 2983, a 25.2% advantage.

Q: What are the memory sizes and bandwidths of each GPU?

A: The GT 640 has 2 GB of DDR3 memory on a 128-bit bus with 28.51 GB/s bandwidth. The 820A has 1024 MB of DDR3 on a 64-bit bus with 16.02 GB/s bandwidth.

Q: How do their power requirements compare?

A: The GT 640 has a TDP of 65 W and a suggested PSU of 250 W, while the 820A has a TDP of only 15 W and no suggested PSU listed.

Q: Do both GPUs support the same APIs?

A: Both support DirectX 12 (11_0) and OpenGL 4.6. However, the GT 640 also supports Vulkan 1.2.175, while the 820A does not list Vulkan support.

Q: Which GPU has more shading units?

A: The GT 640 has 384 shading units, while the 820A has only 96 shading units.

Q: What is the production status of these GPUs?

A: Both are marked as End-of-life. The GT 640 was released on 2012-06-04, and the 820A was released on 2014-03-16.

Where Each One Wins

NVIDIA GeForce GT 640

The GT 640 wins decisively in raw performance metrics. Its 25.2% lead in Geekbench OpenCL makes it the clear choice for any compute-heavy task. It also has a higher average benchmark score of 3210 versus the 820A’s 2983. The GT 640’s 2 GB memory capacity is double that of the 820A, and its 128-bit bus provides nearly double the bandwidth. For gaming or general-purpose GPU computing, the GT 640’s 384 shading units and 16 ROPs provide a much stronger foundation. It also supports Vulkan 1.2.175, opening up modern graphics APIs that the 820A cannot access. The GT 640’s display outputs (DVI, HDMI, DisplayPort) make it suitable for desktop use, and its PCIe 3.0 x16 interface is more modern than the 820A’s PCIe 2.0.

NVIDIA GeForce 820A

The 820A wins in power efficiency and form factor. With a 15 W TDP, it consumes less than a quarter of the power of the GT 640’s 65 W. It is classified as an IGP, meaning it is integrated into a portable device, has no slot width, and no power connectors. This makes it ideal for laptops or compact systems where space and battery life are critical. While its 96 shading units and 8 ROPs are far fewer than the GT 640’s, they are sufficient for basic display output and light compute tasks. The 820A’s smaller die size (116 mm²) and lower transistor count (585 million) contribute to its efficiency, even though its transistor density is lower. For a device that needs basic graphics without the bulk of a discrete card, the 820A is the appropriate pick.