GPU Comparison

NVIDIA GeForce 820A and NVIDIA GeForce GTX 860M are both end-of-life mobile graphics solutions from NVIDIA, sharing the same 28 nm TSMC process node but diverging sharply in almost every other measurable way. The benchmark database places both at the 19th percentile among all GPUs, yet their average scores are nearly identical, 2983 for the 820A versus 2967 for the 860M, a statistical tie that masks wildly different architectural approaches and test coverage.

Head-to-Head Benchmarks

The sole direct head-to-head benchmark available is Geekbench OpenCL, and the result is decisive. The GTX 860M scores 10461, while the 820A scores 2983, giving the 860M a 71.5% advantage. This is not a marginal win; it is a routing. The deltaPct of -71.5% (from the 820A's perspective) indicates that the 820A delivers less than one-third of the 860M's raw compute throughput in this specific workload. For context, the 820A's nearest rivals in the database include the GTX 860M itself (deltaPct 0.5%), GTX 750 Ti (1%), Quadro P600 (2.1%), and RTX 4060 Ti 8 GB (2.4%), all within a narrow 2.4% band of its 2983 average. The GTX 860M's nearest rivals are nearly the same list, but with the 820A at -0.5%, GTX 750 Ti at 0.5%, Quadro P600 at 1.5%, and RTX 4060 Ti 8 GB at 1.9%. What this means is that in the aggregate average benchmark score, these two GPUs are functionally indistinguishable from each other and from a set of very different cards. Yet the OpenCL subtest tells a completely different story, one where the 860M's Kepler architecture with 1152 shading units simply overwhelms the 820A's Fermi 2.0 design with only 96 shading units.

The head-to-head table shows a single entry, and the 820A records zero wins across all compared benchmarks. The 860M records one win. This asymmetry is the central finding: on the one test both cards share, the 860M is superior by a factor of roughly 3.5. The 820A's average benchmark score of 2983 comes from that single OpenCL result, whereas the 860M's average of 2967 is pulled down by its inclusion of PassMark tests, DirectX 9 (55), DirectX 10 (15), DirectX 11 (23), DirectX 12 (13), G2D (226), G3D (3081), and GPU Compute (1251), which are not run on the 820A at all. This means the 820A's score is based on one compute-heavy test, while the 860M's average balances that same test against a broader suite of legacy and directX workloads.

FAQ

Q: Which GPU wins the only head-to-head benchmark available?

A: The GTX 860M wins the Geekbench OpenCL test with a score of 10461 against the 820A's 2983, a deltaPct of -71.5% from the 820A's perspective.

Q: Are these GPUs similar in overall performance?

A: Their average benchmark scores are nearly identical, 2983 for the 820A and 2967 for the 860M, but the 820A's average is derived from a single OpenCL test, while the 860M includes multiple PassMark subtests in its average.

Q: What is the transistor count difference?

A: The 820A uses 585 million transistors on a 116 mm² die, while the 860M uses 3,540 million transistors on a 294 mm² die. The 860M has roughly six times more transistors.

Q: Do both support the same DirectX version?

A: Yes, both report DirectX 12 (11_0) and OpenGL 4.6. The 860M additionally reports Vulkan 1.2.175, while the 820A lists no Vulkan support.

Q: What memory configurations do they use?

A: The 820A has 1024 MB of DDR3 on a 64-bit bus with 16.02 GB/s bandwidth. The 860M has 2 GB of GDDR5 on a 128-bit bus with 80.00 GB/s bandwidth.

Q: How do their compute rates compare?

A: The 820A delivers 240.0 GFLOPS FP32, while the 860M delivers 2.108 TFLOPS FP32, the 860M is roughly 8.8 times higher in raw floating-point throughput.

Architecture Differences

The two GPUs come from different NVIDIA architectures despite sharing the 28 nm TSMC process. The 820A uses the GF117 chip based on Fermi 2.0, while the 860M uses the GK104 chip based on Kepler. This is a generational gap, not just a clock speed difference. The Fermi 2.0 design pairs 96 shading units with 16 texture mapping units and 8 raster output units, producing a pixel rate of 2.500 GPixel/s and a texture rate of 10.00 GTexel/s. The Kepler design on the 860M scales this up dramatically: 1152 shading units, 96 TMUs, and 16 ROPs, yielding 21.96 GPixel/s and 87.84 GTexel/s. The shading unit count is 12 times higher on the 860M, and the texture rate is 8.8 times higher.

Transistor density tells a similar story. The 820A packs 585 million transistors into 116 mm², a density of 5.0M / mm². The 860M fits 3,540 million transistors into 294 mm², a density of 12.0M / mm². Even accounting for the larger die, the Kepler architecture uses the silicon far more efficiently. The memory subsystem also differs fundamentally: the 820A uses DDR3 at 1001 MHz (2 Gbps effective) with a 64-bit bus, while the 860M uses GDDR5 at 1250 MHz (5 Gbps effective) with a 128-bit bus. The 860M's memory bandwidth of 80.00 GB/s is exactly five times the 820A's 16.02 GB/s. The 860M also has a base clock of 797 MHz and a boost clock of 915 MHz, whereas the 820A lists no base or boost clocks in the data.

The API support also differs. Both support DirectX 12 (11_0) and OpenGL 4.6, but the 860M adds Vulkan 1.2.175 support, which the 820A lacks entirely. The bus interface is another point of divergence: the 820A uses PCIe 2.0 x16, while the 860M uses PCIe 3.0 x16. Power consumption is lower on the 820A at 15 W TDP versus 75 W TDP for the 860M, but both are listed as IGP slot width with no power connectors. Neither has any RT cores or tensor cores, consistent with their pre-ray-tracing architectures.

The Verdict

The data points to a clear performance hierarchy, but the choice is not straightforward. The GTX 860M wins the only shared benchmark by 71.5%, and it does so with a higher transistor count, more shading units, faster memory, and support for Vulkan. If the sole criterion is raw compute in OpenCL, the 860M is the superior part. However, the average benchmark scores are essentially tied, 2983 versus 2967, and both sit at the 19th percentile among all GPUs. This suggests that in the aggregate, the 860M's broader test suite brings its average down to the 820A's level, even though the 860M's individual OpenCL result is far higher. The 820A, with only one benchmark recorded, does not have the data to demonstrate any multi-test consistency. For a user prioritizing compute performance in OpenCL workloads, the 860M is the only rational pick. For a user who values lower power consumption (15 W versus 75 W) and a smaller die (116 mm² versus 294 mm²), the 820A offers a more efficient package, though with a fraction of the throughput. The 860M's 2 GB of GDDR5 versus the 820A's 1 GB of DDR3 also matters for memory-intensive tasks. The verdict from the data is that the 860M is the performance winner, while the 820A is the efficiency-oriented option with a much lower thermal and power envelope.

Specification Differences

The two GPUs differ in nearly every specification field. The process node is identical (28 nm, TSMC), but the chip, architecture, transistor count, and die size all diverge. The 820A uses GF117 (Fermi 2.0) with 585 million transistors on 116 mm²; the 860M uses GK104 (Kepler) with 3,540 million transistors on 294 mm². The 820A has no base or boost clock listed, while the 860M has a 797 MHz base and 915 MHz boost. Memory clocks differ: 1001 MHz (2 Gbps effective) for the 820A versus 1250 MHz (5 Gbps effective) for the 860M. Memory size is 1024 MB DDR3 versus 2 GB GDDR5, with bus widths of 64-bit versus 128-bit. Bandwidth is 16.02 GB/s versus 80.00 GB/s. Shading units are 96 versus 1152, TMUs are 16 versus 96, and ROPs are 8 versus 16. Pixel rate is 2.500 GPixel/s versus 21.96 GPixel/s, and texture rate is 10.00 GTexel/s versus 87.84 GTexel/s. FP32 performance is 240.0 GFLOPS versus 2.108 TFLOPS. TDP is 15 W versus 75 W. The bus interface is PCIe 2.0 x16 versus PCIe 3.0 x16. The 820A lacks Vulkan support, while the 860M supports Vulkan 1.2.175. Both have no power connectors and portable-device-dependent display outputs. Release dates are within a week: the 860M on 2014-03-09 and the 820A on 2014-03-16.

Where Each One Wins

The GTX 860M wins the single head-to-head OpenCL benchmark by 71.5%, and it also wins on every raw performance metric in the specification table: shading units, texture rate, pixel rate, FP32, memory bandwidth, and memory capacity. It supports Vulkan, which the 820A does not, making it the better choice for any workload that leverages that API. The 860M's 2 GB GDDR5 frame buffer is double the 820A's 1 GB DDR3, which is relevant for textures and larger datasets. The 820A wins on power efficiency, with a 15 W TDP versus 75 W, a five-fold difference, and on transistor density (5.0M / mm² versus 12.0M / mm² is actually lower, so the 820A uses less silicon per transistor, but the die is smaller overall). The 820A also has a smaller die (116 mm² versus 294 mm²), which could be relevant for thermal-constrained designs, though both are IGP form factors. In terms of benchmark wins, the 820A records zero wins, while the 860M records one. The 820A's only claim is its lower power draw and smaller physical footprint. For any application that values compute throughput, memory bandwidth, or modern API support, the 860M is the clear choice. For a fanless or low-power portable device where 15 W is the maximum allowable, the 820A is the only viable option between the two. The data does not support any scenario where the 820A outperforms the 860M in actual rendering or compute performance; the only context where the 820A leads is in power consumption and die area.