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

AMD Radeon HD 8750M and NVIDIA GeForce GTX 670MX are both end-of-life mobile graphics solutions from the 2012–2013 era, but the benchmark data positions them very differently. The GTX 670MX wins the only shared test (Geekbench OpenCL) with a 2.5% margin, yet the HD 8750M posts a higher average benchmark score across its own results. This creates a nuanced picture: the NVIDIA part has a higher peak in one workload, while the AMD part shows more consistent scoring in its available data. The GTX 670MX also holds advantages in raw hardware specifications—more shading units, texture units, ROPs, memory bandwidth, and a larger frame buffer—but the HD 8750M counters with a higher base clock and a more modern DirectX feature level (12_1 vs 11_0). The data does not support a universal winner; instead, each card wins in specific contexts that this analysis will break down.

Where Each One Wins

The NVIDIA GeForce GTX 670MX wins the only direct head-to-head benchmark, the Geekbench OpenCL test, with a score of 6125 against the AMD Radeon HD 8750M’s 5970. That is a 2.5% lead, which is meaningful but not overwhelming. The GTX 670MX also achieves a higher percentile placement in its own benchmark set: it scores 6125 in OpenCL and 5316 in Vulkan, giving it a broader software workload profile. Its average benchmark score of 5721, however, is dragged down by the Vulkan result, which is notably lower than its OpenCL score.

The AMD Radeon HD 8750M wins on consistency. It has only one benchmark score (5970 in OpenCL), but that single score is higher than the GTX 670MX’s average of 5721. The HD 8750M’s percentile rank of 34 is also one point higher than the GTX 670MX’s 33, indicating that in the broader GPU landscape, the AMD part sits slightly better relative to all other GPUs. The HD 8750M also has a higher base clock (775 MHz vs 601 MHz) and boost clock (825 MHz vs 601 MHz), which suggests it may hold an advantage in frequency-sensitive workloads, even if the NVIDIA card compensates with far more execution resources.

In terms of memory, the GTX 670MX is the clear victor: 3 GB of GDDR5 on a 192-bit bus delivers 67.20 GB/s of bandwidth, versus the HD 8750M’s 1 GB of DDR3 on a 128-bit bus at 28.80 GB/s. That 2.33x bandwidth advantage means the NVIDIA card wins in any scenario where texture streaming or large data sets are involved, even if the raw compute scores are close. Conversely, the HD 8750M’s smaller memory footprint and slower memory type would be a bottleneck in modern workloads, but its higher clocks could still give it an edge in short, compute-bound bursts.

FAQ

Q: Which GPU has the higher single benchmark score?

A: The NVIDIA GeForce GTX 670MX scores 6125 in Geekbench OpenCL, which is 2.5% higher than the AMD Radeon HD 8750M’s 5970 in the same test.

Q: Does the AMD Radeon HD 8750M outperform the GTX 670MX in any benchmark?

A: The head-to-head data shows no wins for the AMD card (0 wins vs 1 win for NVIDIA). However, the HD 8750M’s average benchmark score of 5970 is higher than the GTX 670MX’s average of 5721, reflecting that the AMD part’s only result is above the NVIDIA card’s blended average.

Q: How do the two compare in memory capacity and bandwidth?

A: The GTX 670MX has 3 GB of GDDR5 memory on a 192-bit bus, yielding 67.20 GB/s bandwidth. The HD 8750M has 1 GB of DDR3 on a 128-bit bus, yielding 28.80 GB/s—the NVIDIA card offers more than double the bandwidth.

Q: Which GPU supports a newer version of DirectX?

A: The AMD Radeon HD 8750M supports DirectX 12 (11_1), while the NVIDIA GeForce GTX 670MX supports DirectX 12 (11_0). The AMD part’s feature level is one step higher.

Q: What are the transistor and die size differences?

A: The GTX 670MX uses 3,540 million transistors on a 294 mm² die, while the HD 8750M uses 950 million transistors on a 77 mm² die. The NVIDIA chip is substantially larger and more complex.

Q: Do both GPUs have the same Vulkan support?

A: No. The GTX 670MX supports Vulkan 1.2.175, while the HD 8750M supports Vulkan 1.2.170. The NVIDIA driver stack offers a slightly newer Vulkan version.

Head-to-Head Benchmarks

The only direct benchmark comparison available is Geekbench OpenCL, and the results are close. The NVIDIA GeForce GTX 670MX scores 6125, while the AMD Radeon HD 8750M scores 5970. The delta is -2.5% for the AMD card, meaning NVIDIA wins by that margin. In absolute terms, that is a 155-point gap, which is small enough to be attributed to driver optimization or workload-specific behavior rather than a fundamental architectural superiority.

Looking at the broader benchmark context, the GTX 670MX also has a Geekbench Vulkan score of 5316, which is 13.2% lower than its own OpenCL score. This suggests that the NVIDIA card’s performance is highly API-dependent, and its OpenCL win over the HD 8750M may not translate to Vulkan-based workloads. The HD 8750M does not have a Vulkan score in the data, so it is impossible to say whether it would outperform or underperform the GTX 670MX in Vulkan.

Examining the nearest rivals provides additional context. The HD 8750M’s closest competitor is the NVIDIA Quadro K4000, which scores 5982 (0.2% higher), and the AMD Radeon HD 8730M at 5955 (0.3% lower). This places the HD 8750M in a tight cluster where a 1% swing can change rankings. The GTX 670MX’s nearest rivals include the NVIDIA GeForce GTX 550 Ti at 5731 (0.2% higher) and the Intel Iris Pro Graphics P6300 at 5712 (0.2% lower). The GTX 670MX’s average score of 5721 is notably lower than its OpenCL peak, indicating that its Vulkan result significantly pulls down its overall standing.

Specification Differences

The two GPUs differ across nearly every major specification. The NVIDIA GeForce GTX 670MX has 960 shading units, 80 texture mapping units, and 24 ROPs, while the AMD Radeon HD 8750M has 384 shading units, 24 TMUs, and 8 ROPs. That is a 2.5x, 3.3x, and 3x advantage for NVIDIA in those respective categories. The GTX 670MX also has a much higher pixel rate (12.02 GPixel/s vs 6.60 GPixel/s) and texture rate (48.08 GTexel/s vs 19.80 GTexel/s). Its FP32 throughput is 1,153.9 GFLOPS versus 633.6 GFLOPS for the AMD card.

Clock speeds tell the opposite story. The HD 8750M runs at 775 MHz base and 825 MHz boost, while the GTX 670MX is locked at 601 MHz for both base and boost. The AMD card’s clock advantage is 28.9% at base and 37.3% at boost, partially compensating for its fewer execution units. Memory is another area of stark contrast: the GTX 670MX uses 3 GB of GDDR5 on a 192-bit bus at 67.20 GB/s, while the HD 8750M uses 1 GB of DDR3 on a 128-bit bus at 28.80 GB/s. The bus interface also differs—the NVIDIA card uses PCIe 3.0 x16, while the AMD card uses PCIe 3.0 x8.

Power consumption is only listed for the GTX 670MX at 75 W TDP with no power connectors required. The HD 8750M has no TDP listed in the data. The NVIDIA card’s die is 294 mm² with 3,540 million transistors, versus 77 mm² and 950 million for the AMD card. Transistor density is nearly identical (12.0M/mm² vs 12.3M/mm²), but the GTX 670MX packs far more hardware into its larger die.

Architecture Differences

The AMD Radeon HD 8750M is built on the Graphics Core Next (GCN) 1.0 architecture with the Mars chip, while the NVIDIA GeForce GTX 670MX uses the Kepler architecture with the GK104 chip. Both are fabricated on a 28 nm TSMC process, but the transistor counts diverge dramatically: the GK104 uses 3,540 million transistors, while Mars uses 950 million. The die size difference is even more pronounced at 294 mm² versus 77 mm².

The memory subsystems are architecturally distinct. The GTX 670MX uses GDDR5 with a 192-bit interface, while the HD 8750M uses DDR3 with a 128-bit interface. This is not just a bandwidth difference (67.20 GB/s vs 28.80 GB/s) but also a fundamental design choice—GDDR5 is optimized for high bandwidth, while DDR3 emphasizes lower power. The NVIDIA card’s memory clock is 700 MHz (2.8 Gbps effective), versus 900 MHz (1800 Mbps effective) for the AMD part.

The API support also reflects different design philosophies. The HD 8750M supports DirectX 12 (11_1) and Vulkan 1.2.170, while the GTX 670MX supports DirectX 12 (11_0) and Vulkan 1.2.175. The AMD card has a higher DirectX feature level, while the NVIDIA card has a slightly newer Vulkan version. Both support OpenGL 4.6. The HD 8750M is part of the Solar System generation (HD 8700M), succeeding the London chip and preceding Gem System, while the GTX 670MX belongs to the GeForce 600M generation, succeeding GeForce 500M and preceding GeForce 700M.

The Verdict

The data points to a clear split: the NVIDIA GeForce GTX 670MX is the stronger choice for raw throughput and memory-intensive workloads, while the AMD Radeon HD 8750M offers a more favorable balance in frequency and API feature level. The GTX 670MX wins the only shared benchmark (OpenCL) with a 2.5% margin, and its hardware specifications are overwhelmingly superior—3x more shading units, 3.3x more TMUs, 3x more ROPs, and 2.33x more memory bandwidth. The 3 GB frame buffer versus 1 GB is a decisive advantage for modern games or compute tasks that require large working sets.

However, the HD 8750M’s higher clocks (775 MHz vs 601 MHz base) and superior DirectX feature level (12_1 vs 11_0) mean it is not obsolete by any measure. Its average benchmark score of 5970 exceeds the GTX 670MX’s average of 5721, which shows that in the AMD card’s single tested workload, it performs above the NVIDIA card’s blended average. The GTX 670MX’s Vulkan score of 5316 is significantly lower than its OpenCL score, indicating that its performance is inconsistent across APIs.

For users prioritizing memory capacity and bandwidth, the GTX 670MX is the obvious pick. For those who value higher clock speeds, a newer DirectX feature level, or a smaller, more efficient die (77 mm² vs 294 mm²), the HD 8750M holds appeal. The benchmark data does not crown a single champion—it shows a NVIDIA card that dominates in resources and a AMD card that stays competitive through efficiency and frequency. The GTX 670MX is the better performer in the head-to-head test, but the HD 8750M’s higher percentile rank (34 vs 33) suggests it is marginally better positioned relative to the entire GPU landscape.