AMD Radeon HD 8750M vs NVIDIA GeForce GTX 670M Comparison

Head-to-Head Benchmarks

The only directly comparable benchmark in the database is Geekbench OpenCL, and the results show a decisive win for the NVIDIA GeForce GTX 670M. It scores 6513 points against 5970 for the AMD Radeon HD 8750M, a lead of 9.1%. That is a meaningful gap in raw compute workloads, placing the GTX 670M clearly ahead of the HD 8750M in general-purpose GPU tasks.

Placing both in context, the GTX 670M sits at the 38th percentile of all GPUs in the database, while the HD 8750M sits at the 34th percentile. That percentile difference, though modest in rank terms, is backed by the 9.1% score advantage in the head-to-head measurement. The GTX 670M’s nearest rival is the NVIDIA GeForce GT 555M, which scores 6493, just 0.3% behind. The GTX 670M also edges out the NVIDIA Quadro M5000M at 6481 (0.5% ahead) and the AMD Radeon Vega 10 Mobile at 6476 (0.6% ahead), while trailing the Intel UHD Graphics P750 at 6554 by 0.6%. So the GTX 670M is part of a tight cluster of mobile GPUs, but it holds its own at the upper edge of that group.

The AMD Radeon HD 8750M, by contrast, sits in a slightly lower cluster. Its nearest rival is the NVIDIA Quadro K4000M at 5986, which is 0.3% ahead. The NVIDIA Quadro K4000 is 0.2% ahead at 5982, and the AMD Radeon HD 8730M is 0.3% behind at 5955. The NVIDIA Quadro K620M is 0.2% behind at 5957. This means the HD 8750M is essentially in the middle of its own peer group, with no rival more than 0.3% away in either direction. The delta between the two GPUs in this comparison, 9.1%, is far larger than any of the deltas within their respective rival clusters, which reinforces that the GTX 670M and HD 8750M are not really competing in the same performance tier.

Architecture Differences

The architectural divide between these two mobile GPUs is substantial. The NVIDIA GeForce GTX 670M uses the GF114 chip, built on the Fermi 2.0 architecture, manufactured on a 40 nm process at TSMC. The AMD Radeon HD 8750M uses the Mars chip, built on GCN 1.0, on a 28 nm process, also at TSMC. The process node difference alone, 40 nm versus 28 nm, explains a great deal about the efficiency and density characteristics of each part, even if the database does not record direct power consumption figures for the AMD side.

Transistor counts tell a similar story. The GTX 670M packs 1,950 million transistors on a 332 mm² die, giving it a transistor density of 5.9M per mm². The HD 8750M has 950 million transistors on a 77 mm² die, which works out to 12.3M per mm². So the AMD chip is far smaller and denser, while the NVIDIA chip is larger overall with roughly twice the transistor budget. This is a classic late-life Fermi part versus a mid-life GCN part: the older architecture needs more die area and more transistors to deliver its performance, while the newer architecture achieves a more compact design.

Memory configurations also diverge sharply. The GTX 670M ships with 1536 MB of GDDR5 on a 192-bit bus, yielding 72.00 GB/s of bandwidth. The HD 8750M ships with 1024 MB of DDR3 on a 128-bit bus, yielding only 28.80 GB/s. That is a 43.20 GB/s difference in memory bandwidth, which is enormous for mobile GPUs of this era. For any workload that depends on feeding data to the compute units quickly, the GTX 670M has a fundamental advantage. The HD 8750M’s memory clock is listed at 900 MHz (1800 Mbps effective), while the GTX 670M’s memory runs at 750 MHz (3 Gbps effective). The GTX 670M’s wider bus and faster effective memory speed combine for that large bandwidth lead.

Compute resources differ in interesting ways. The GTX 670M has 336 shading units, 56 texture mapping units, and 24 ROPs. The HD 8750M has 384 shading units, 24 TMUs, and 8 ROPs. So AMD has more shaders on paper, but NVIDIA has far more TMUs and triple the ROP count. This shows up in the rate figures: the GTX 670M achieves 8.372 GPixel/s pixel rate and 33.49 GTexel/s texture rate, while the HD 8750M manages 6.600 GPixel/s and 19.80 GTexel/s. The FP32 throughput is 803.7 GFLOPS for the GTX 670M versus 633.6 GFLOPS for the HD 8750M. The GTX 670M wins every rate metric despite having fewer shading units, which points to higher clock efficiency in the NVIDIA design and the benefit of a much wider memory interface.

Clock speeds are only partially recorded. The GTX 670M has no base or boost clock listed, only its memory clock. The HD 8750M has a base clock of 775 MHz and a boost clock of 825 MHz. The GTX 670M’s TDP is recorded as 75 W, while the HD 8750M’s TDP is not recorded in the database. The GTX 670M uses an MXM module form factor with an MXM-B (3.0) bus interface and no power connectors, while the HD 8750M uses PCIe 3.0 x8. Display outputs on the GTX 670M are listed as portable device dependent, and the HD 8750M has no display output data recorded.

API support also differs. The GTX 670M supports DirectX 12 (11_0) and OpenGL 4.6, with no Vulkan entry. The HD 8750M supports DirectX 12 (11_1), OpenGL 4.6, and Vulkan 1.2.170. So the AMD part has a more modern feature set in terms of API coverage, particularly the Vulkan support, which the NVIDIA Fermi part lacks entirely.

FAQ

Q: Which GPU is faster in the database’s benchmark results?

A: The NVIDIA GeForce GTX 670M scores 6513 in Geekbench OpenCL, while the AMD Radeon HD 8750M scores 5970. The GTX 670M wins by 9.1%.

Q: How do these two compare to their closest rivals?

A: The GTX 670M’s nearest rival is the NVIDIA GeForce GT 555M at 6493, just 0.3% behind. The HD 8750M’s nearest rival is the NVIDIA Quadro K4000M at 5986, which is 0.3% ahead of the AMD part.

Q: Does the AMD Radeon HD 8750M have more shading units?

A: Yes, the HD 8750M has 384 shading units, while the GTX 670M has 336. Despite this, the GTX 670M delivers higher FP32 throughput at 803.7 GFLOPS versus 633.6 GFLOPS.

Q: What is the memory bandwidth difference?

A: The GTX 670M has 72.00 GB/s of bandwidth from 1536 MB of GDDR5 on a 192-bit bus. The HD 8750M has 28.80 GB/s from 1024 MB of DDR3 on a 128-bit bus.

Q: Which GPU has better API support?

A: The HD 8750M supports Vulkan 1.2.170 and DirectX 12 (11_1). The GTX 670M has no Vulkan support and DirectX 12 (11_0). Both support OpenGL 4.6.

Q: Are these GPUs still in production?

A: No, both are listed as end-of-life products. The GTX 670M released earlier and was succeeded by the GeForce 700M, while the HD 8750M was succeeded by the Gem System.

The Verdict

The data points to a clear winner for raw performance. The GTX 670M beats the HD 8750M by 9.1% in the only benchmark recorded for both, and it wins every measured throughput metric: pixel rate, texture rate, and FP32. The GTX 670M also has more than double the memory bandwidth, which alone makes it the better choice for bandwidth-sensitive workloads. The HD 8750M does have some advantages in the record: a smaller die, higher transistor density, a more modern 28 nm process, and Vulkan support. But none of those translate into a benchmark win.

For anyone choosing between these two in a laptop or MXM upgrade context, the GTX 670M is the stronger performer. It sits at the 38th percentile of all GPUs, above the HD 8750M’s 34th percentile. Its nearest rival comparisons show it clustered with parts like the GeForce GT 555M and Quadro M5000M, all within 0.6%, while the HD 8750M sits with a slower group centered around the Quadro K4000 series. The performance gap between the two is not a rounding error; it is a real tier difference.

The HD 8750M is not without merit. Its GCN 1.0 architecture, 28 nm process, and Vulkan support make it a more modern design in some respects. It uses far fewer transistors at 950 million versus 1,950 million, and its 77 mm² die is much smaller than 332 mm². For a system builder focused on compatibility or feature support rather than peak speed, the HD 8750M could still be acceptable. But the benchmark evidence is unambiguous: the GTX 670M is the faster GPU.

Specification Differences

| Specification | NVIDIA GeForce GTX 670M | AMD Radeon HD 8750M |

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

| Chip | GF114 | Mars |

| Architecture | Fermi 2.0 | GCN 1.0 |

| Process Node | 40 nm | 28 nm |

| Transistors | 1,950 million | 950 million |

| Die Size | 332 mm² | 77 mm² |

| Transistor Density | 5.9M / mm² | 12.3M / mm² |

| Memory Size | 1536 MB | 1024 MB |

| Memory Type | GDDR5 | DDR3 |

| Memory Bus Width | 192 bit | 128 bit |

| Memory Bandwidth | 72.00 GB/s | 28.80 GB/s |

| Memory Clock | 750 MHz (3 Gbps effective) | 900 MHz (1800 Mbps effective) |

| Base Clock | Not recorded | 775 MHz |

| Boost Clock | Not recorded | 825 MHz |

| Shading Units | 336 | 384 |

| TMUs | 56 | 24 |

| ROPs | 24 | 8 |

| Pixel Rate | 8.372 GPixel/s | 6.600 GPixel/s |

| Texture Rate | 33.49 GTexel/s | 19.80 GTexel/s |

| FP32 | 803.7 GFLOPS | 633.6 GFLOPS |

| TDP | 75 W | Not recorded |

| Bus Interface | MXM-B (3.0) | PCIe 3.0 x8 |

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

| OpenGL | 4.6 | 4.6 |

| Vulkan | Not recorded | 1.2.170 |

| Release Date | 2012-03-21 | 2013-02-25 |

| Predecessor | GeForce 500M | London |

| Successor | GeForce 700M | Gem System |

| Production Status | End-of-life | End-of-life |

Where Each One Wins

The GTX 670M wins in every directly measured performance category. Its 9.1% lead in Geekbench OpenCL is the headline result, but the supporting data makes the case even stronger. It has higher pixel rate at 8.372 GPixel/s versus 6.600 GPixel/s, higher texture rate at 33.49 GTexel/s versus 19.80 GTexel/s, higher FP32 at 803.7 GFLOPS versus 633.6 GFLOPS, and more than double the memory bandwidth at 72.00 GB/s versus 28.80 GB/s. Any workload that stresses the GPU compute pipeline or memory subsystem will favor the GTX 670M. Its 56 TMUs and 24 ROPs, compared to 24 and 8 respectively, give it a structural advantage in texture-heavy and pixel-heavy rendering paths. For gaming, general compute, or any task that needs sustained throughput, the GTX 670M is the pick.

The HD 8750M wins in architectural efficiency and modern feature coverage. Its 28 nm process and 77 mm² die with 12.3M transistors per mm² represent a much denser, more modern design than the GTX 670M’s 40 nm, 332 mm² die. It supports Vulkan 1.2.170, which the GTX 670M does not, and its DirectX 12 (11_1) support is a minor step ahead of the GTX 670M’s DirectX 12 (11_0). It also has more shading units at 384 versus 336, though that does not translate into a performance win. For a user who values newer API compatibility, lower transistor count, or the efficiency of a smaller chip, the HD 8750M has something to offer. But in raw benchmark results, the GTX 670M is the clear winner.