AMD Radeon HD 8870M vs NVIDIA GeForce GTX 680M Comparison

Where Each One Wins

The benchmark data splits these two mobile GPUs cleanly, though the sample size is limited to a single shared test. The NVIDIA GeForce GTX 680M takes the only head-to-head win, outperforming the AMD Radeon HD 8870M in Geekbench OpenCL by 8.3%. That is the sole direct comparison available, and it favors NVIDIA across the board in that workload.

Looking at the broader database, the AMD Radeon HD 8870M sits at the 43rd percentile of all GPUs, while the NVIDIA GeForce GTX 680M lands at the 39th percentile. Despite the lower percentile ranking, the GTX 680M posts a higher OpenCL score in the head-to-head test. The percentile figures reflect each card's position relative to the entire GPU population, not just the matchup between these two specific rivals, so the GTX 680M's 39th percentile is pulled down by the larger number of faster modern parts above it.

The AMD card's average benchmark score is 8462, which comes from its single OpenCL result. The GTX 680M averages 7023 across two tests, one OpenCL and one Metal result of 4815. That Metal score drags its average down substantially, even though its OpenCL score beats the AMD card's only recorded number. The database shows the GTX 680M's nearest rivals include the NVIDIA T600 at 7035, only 0.2% higher, and the AMD Radeon R5 M240 at 6975, which is 0.7% lower. The GTX 675M trails by 1.1% at 6946, while the GTX 970 sits 1.9% ahead at 7157.

The AMD Radeon HD 8870M's nearest rivals tell a different story. The NVIDIA GeForce MX330 scores 8458, just 0.1% behind the AMD card, and the AMD Radeon 880M is 0.3% behind at 8436. The GTX 675MX trails by 0.4% at 8427, while the Intel Arc A380 runs 1.1% ahead at 8558. These tight deltas show the HD 8870M clustering with the MX330 and the integrated Radeon 880M in raw compute, a much closer grouping than the GTX 680M experiences against its rivals.

The GTX 680M wins where it matters in this comparison: the single compute benchmark both cards share. The HD 8870M does not win any recorded benchmark in this matchup, so its case rests entirely on its percentile position and the competitive scores of its nearest rivals rather than any direct victory over the NVIDIA part.

Architecture Differences

The two chips come from different architectural families and different eras of mobile GPU design. The AMD Radeon HD 8870M uses the Venus chip built on GCN 1.0 architecture, part of the Solar System generation under the HD 8800M line. The NVIDIA GeForce GTX 680M uses the GK104 chip on Kepler architecture, from the GeForce 600M generation. Both are manufactured on a 28 nm process at TSMC, so the process node is identical, but the physical design diverges sharply.

Transistor counts reveal the scale difference. The AMD chip packs 1,500 million transistors on a 123 mm² die, giving a transistor density of 12.2 million per square millimeter. The NVIDIA chip carries 3,540 million transistors on a 294 mm² die, with a slightly lower density of 12.0 million per square millimeter. The GTX 680M's die is more than twice the physical size and holds well over twice the transistor count, which explains much of its compute advantage.

Execution resources differ dramatically. The AMD Radeon HD 8870M has 640 shading units, 40 texture mapping units, and 16 render output units. The NVIDIA GeForce GTX 680M has 1,344 shading units, 112 TMUs, and 32 ROPs. In every execution category, the NVIDIA part has roughly double the hardware, and that structural advantage shows up in the computed rates. The AMD card delivers 12.40 GPixel/s of pixel rate and 31.00 GTexel/s of texture rate. The GTX 680M delivers 21.22 GPixel/s and 84.90 GTexel/s. The NVIDIA card also reaches 2.038 TFLOPS of FP32 compute, over twice the AMD card's 992.0 GFLOPS.

Clock speeds are close. The AMD card runs a 725 MHz base clock with a 775 MHz boost. The NVIDIA card runs 719 MHz base with 758 MHz boost. The AMD silicon actually clocks slightly higher, but the NVIDIA chip's much larger execution resource pool overwhelms that small clock advantage.

Memory configurations also differ. The AMD Radeon HD 8870M has 2 GB of GDDR5 on a 128 bit bus, with memory clocked at 1125 MHz or 4.5 Gbps effective, yielding 72.00 GB/s of bandwidth. The GTX 680M has 4 GB of GDDR5 on a 256 bit bus, with memory at 900 MHz or 3.6 Gbps effective, yielding 115.2 GB/s. The NVIDIA card's wider bus and larger capacity give it 60% more bandwidth despite lower memory clocks.

API support varies slightly. The AMD card supports DirectX 12 (11_1), OpenGL 4.6, and Vulkan 1.2.170. The NVIDIA card supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The Vulkan revision is newer on the NVIDIA side, while the AMD card's DirectX feature level is one notch higher. Both are end-of-life products.

The GTX 680M carries a 100 W TDP, while the database does not list a TDP for the HD 8870M. The NVIDIA card is an MXM Module with no power connectors and portable-device-dependent display outputs. The AMD card uses a PCIe 3.0 x16 bus interface, while the NVIDIA card uses MXM-B (3.0).

Head-to-Head Benchmarks

The single recorded head-to-head benchmark is Geekbench OpenCL, and the result is clear. The NVIDIA GeForce GTX 680M scores 9230, while the AMD Radeon HD 8870M scores 8462. The delta is 8.3% in NVIDIA's favor, making the GTX 680M the winner in the only direct comparison the database provides.

That 8462 score from the AMD card is remarkably close to the scores of its nearest rivals. The GeForce MX330 posts 8458, just 0.1% behind. The Radeon 880M posts 8436, 0.3% behind. The GTX 675MX posts 8427, 0.4% behind. Only the Intel Arc A380 at 8558 beats the HD 8870M among its nearest rivals, and only by 1.1%. The AMD card sits in a very tight cluster where every nearby GPU falls within about one percent of its score.

The GTX 680M's 9230 OpenCL result sits in a different competitive zone. Among its nearest rivals, the T600 at 7035 is 0.2% ahead of the GTX 680M's average, which includes that weaker Metal score. The R5 M240 at 6975 is 0.7% behind, the GTX 675M at 6946 is 1.1% behind, and the GTX 970 at 7157 is 1.9% ahead. These numbers reflect the GTX 680M's average across two tests, not its OpenCL peak, so the direct OpenCL comparison against the AMD card is the more telling figure.

The GTX 680M's Metal score of 4815 has no AMD counterpart in the database, since the HD 8870M was only tested with OpenCL. That Metal result drags the NVIDIA card's average down to 7023, which is why its percentile sits at 39 despite winning the head-to-head OpenCL test. The AMD card's average equals its OpenCL score at 8462, since no other benchmark was recorded for it.

The 8.3% gap in OpenCL is meaningful but not enormous. It means the GTX 680M completes roughly 8.3% more compute work per unit time in that workload, consistent with its much larger shading unit count and wider memory bus. The AMD card's higher clock speeds and newer DirectX feature level do not close the gap in raw compute throughput.

The Verdict

The data points to the NVIDIA GeForce GTX 680M as the stronger GPU in this pairing, based on the only benchmark both cards share. The GTX 680M wins the OpenCL test by 8.3%, and its architecture carries roughly double the shading units, TMUs, and ROPs of the Radeon HD 8870M. Its FP32 throughput of 2.038 TFLOPS is more than twice the AMD card's 992.0 GFLOPS, and its 115.2 GB/s of memory bandwidth is 60% higher than the AMD card's 72.00 GB/s.

The AMD Radeon HD 8870M is not without merit. It sits at the 43rd percentile of all GPUs, four points higher than the GTX 680M's 39th percentile, and its single OpenCL score of 8462 places it in a tight competitive cluster with the MX330, Radeon 880M, and GTX 675MX. It also supports DirectX 12 at the 11_1 feature level, one step above the NVIDIA card's 11_0, and its Vulkan driver reports version 1.2.170 against the NVIDIA card's 1.2.175. The AMD card uses less than half the transistors of the NVIDIA part and has a smaller die, so it fits in more power-constrained designs.

For buyers choosing between these two end-of-life mobile GPUs, the GTX 680M is the pick when raw compute performance and memory bandwidth matter most. The NVIDIA card wins the only direct benchmark, has double the VRAM at 4 GB against 2 GB, and delivers higher pixel and texture rates across the board. The HD 8870M makes sense when the lower transistor count, smaller die, and newer DirectX feature level are priorities, or when the workload is light enough that the 8.3% OpenCL gap does not matter.

The percentile rankings complicate the story. The HD 8870M ranks 43rd against all GPUs, while the GTX 680M ranks 39th, so the AMD card sits higher in the overall distribution despite losing the head-to-head test. That is because the GTX 680M's average score includes its weak Metal result of 4815, which pulls the average down to 7023. The database records only one shared benchmark, and on that test the NVIDIA part is faster.

FAQ

Q: Which GPU wins the only shared benchmark?

A: The NVIDIA GeForce GTX 680M wins Geekbench OpenCL with a score of 9230 against the AMD Radeon HD 8870M's 8462, a delta of 8.3%.

Q: How much memory does each card have?

A: The AMD Radeon HD 8870M has 2 GB of GDDR5 on a 128 bit bus, while the NVIDIA GeForce GTX 680M has 4 GB of GDDR5 on a 256 bit bus.

Q: What are the transistor counts of the two chips?

A: The AMD Venus chip has 1,500 million transistors on a 123 mm² die, and the NVIDIA GK104 chip has 3,540 million transistors on a 294 mm² die.

Q: Which card has higher compute throughput?

A: The GTX 680M reaches 2.038 TFLOPS of FP32, while the HD 8870M reaches 992.0 GFLOPS, making the NVIDIA card more than twice as fast in raw FP32 compute.

Q: How do the two cards rank against all GPUs?

A: The AMD Radeon HD 8870M sits at the 43rd percentile, and the NVIDIA GeForce GTX 680M sits at the 39th percentile.

Q: What DirectX versions do the cards support?

A: The AMD card supports DirectX 12 at the 11_1 feature level, while the NVIDIA card supports DirectX 12 at the 11_0 feature level.

Specification Differences

| Specification | AMD Radeon HD 8870M | NVIDIA GeForce GTX 680M |

|---|---|---|

| Chip | Venus | GK104 |

| Architecture | GCN 1.0 | Kepler |

| Generation | Solar System (HD 8800M) | GeForce 600M |

| Transistors | 1,500 million | 3,540 million |

| Die Size | 123 mm² | 294 mm² |

| Transistor Density | 12.2M / mm² | 12.0M / mm² |

| Base Clock | 725 MHz | 719 MHz |

| Boost Clock | 775 MHz | 758 MHz |

| Memory Clock | 1125 MHz, 4.5 Gbps effective | 900 MHz, 3.6 Gbps effective |

| Memory Size | 2 GB | 4 GB |

| Memory Bus | 128 bit | 256 bit |

| Memory Bandwidth | 72.00 GB/s | 115.2 GB/s |

| Shading Units | 640 | 1344 |

| TMUs | 40 | 112 |

| ROPs | 16 | 32 |

| Pixel Rate | 12.40 GPixel/s | 21.22 GPixel/s |

| Texture Rate | 31.00 GTexel/s | 84.90 GTexel/s |

| FP32 | 992.0 GFLOPS | 2.038 TFLOPS |

| TDP | Not listed | 100 W |

| Slot Width | Not listed | MXM Module |

| Power Connectors | Not listed | None |

| Bus Interface | PCIe 3.0 x16 | MXM-B (3.0) |

| Display Outputs | Not listed | Portable Device Dependent |

| DirectX | 12 (11_1) | 12 (11_0) |

| Vulkan | 1.2.170 | 1.2.175 |

| Release Date | 2013-03-31 | 2012-06-03 |

| Predecessor | London | GeForce 500M |

| Successor | Gem System | GeForce 700M |