AMD Radeon HD 8790M vs NVIDIA GeForce GTX 670MX Comparison

The GeForce GTX 670MX and Radeon HD 8790M are a study in contrasting design philosophies from the 28 nm era, both landing at the 33rd percentile of all GPUs. Their average benchmark scores are separated by a mere 0.5% — 5721 for the NVIDIA part versus 5691 for the AMD part — making them effectively equivalent in overall synthetic performance. Yet the data reveals a sharp split: the GTX 670MX dominates OpenCL workloads, while the HD 8790M takes a decisive lead in Vulkan, suggesting that the "better" card depends entirely on the API in question.

Where Each One Wins

The workload split is unambiguous. The NVIDIA GeForce GTX 670MX is the clear winner in OpenCL compute tasks, delivering a Geekbench OpenCL score of 6125 against the AMD Radeon HD 8790M's 5017. That margin of 22.1% is substantial and indicates that applications leveraging OpenCL — commonly found in older compute tasks, video encoding, and some productivity suites — will see a marked advantage on the NVIDIA solution.

The AMD Radeon HD 8790M, conversely, claims victory in Vulkan-based workloads. Its Geekbench Vulkan score of 6365 outpaces the GTX 670MX's 5316 by a significant 16.5%. This is the more forward-looking win, as Vulkan is the modern cross-platform graphics and compute API. For users running contemporary games or applications built on Vulkan, the HD 8790M offers a tangible performance edge that the NVIDIA card cannot match.

This is a classic case of API-specific optimization. The GTX 670MX's architecture leans heavily toward compute throughput in legacy APIs, while the HD 8790M's GCN design appears better suited to the lower-overhead, more parallel-friendly Vulkan environment. The benchmark data shows a 1-1 split in wins, but the magnitude of each victory is nearly identical, reinforcing the notion that neither card is universally superior. Users prioritizing OpenCL should choose NVIDIA; those on Vulkan should choose AMD.

Architecture Differences

The two GPUs are built on the same 28 nm TSMC process but diverge completely at the silicon level. The NVIDIA GeForce GTX 670MX uses the GK104 chip, a large Kepler-architecture die measuring 294 mm² and packing 3,540 million transistors. The AMD Radeon HD 8790M, in contrast, uses the Mars chip with GCN 1.0 architecture, a much smaller 77 mm² die with just 950 million transistors.

Despite the massive difference in physical size, the transistor density is nearly identical — 12.0M per mm² for NVIDIA versus 12.3M per mm² for AMD. This means the GTX 670MX's size advantage comes purely from having a larger budget for silicon, not from denser packing. The Kepler architecture on the GTX 670MX is designed around a wide array of 960 shading units, 80 texture mapping units, and 24 ROPs. The GCN 1.0 architecture on the HD 8790M is more modest in raw counts, with 384 shading units, 24 TMUs, and just 8 ROPs.

Clock speeds tell a different story. The AMD part runs at a base clock of 850 MHz with a boost to 900 MHz, while the NVIDIA part sits at a static 601 MHz with no boost. That higher frequency helps AMD close the gap in raw throughput calculations. The GTX 670MX's pixel rate is 12.02 GPixel/s versus 7.200 GPixel/s for the HD 8790M, and its texture rate of 48.08 GTexel/s more than doubles AMD's 21.60 GTexel/s. In terms of FP32 compute, NVIDIA leads with 1,153.9 GFLOPS against AMD's 691.2 GFLOPS.

API support differs slightly as well. The GTX 670MX supports DirectX 12 (11_0) and Vulkan 1.2.175, while the HD 8790M supports DirectX 12 (11_1) and Vulkan 1.2.170. Both offer OpenGL 4.6. The NVIDIA card uses a PCIe 3.0 x16 interface, whereas the AMD card is a removable MXM-A (3.0) module.

Head-to-Head Benchmarks

The Geekbench OpenCL test delivers the GTX 670MX's biggest win. With a score of 6125 versus 5017, the NVIDIA card finishes 22.1% ahead of the AMD part. This is a decisive margin that speaks to the raw compute capacity of the larger Kepler die. The GTX 670MX's 80 TMUs and 24 ROPs, combined with its 1,153.9 GFLOPS of FP32 throughput, provide ample resources for OpenCL's parallel workload patterns. In this test, the HD 8790M's higher clocks cannot compensate for its smaller execution engine.

The Vulkan benchmark flips the script entirely. The AMD Radeon HD 8790M scores 6365, which is 16.5% higher than the GTX 670MX's 5316. This result is counterintuitive at first glance, given the NVIDIA card's superior raw specs. However, GCN 1.0 architecture has historically demonstrated strong Vulkan driver efficiency and scheduling behavior. The HD 8790M's 900 MHz boost clock likely provides an advantage in latency-sensitive Vulkan workloads where the GTX 670MX's fixed 601 MHz clock becomes a bottleneck.

Looking at the broader competitive landscape, both cards sit in a tight cluster. The GTX 670MX's average score of 5721 places it just 0.2% above the Intel Iris Pro Graphics P6300 (5712) and 0.5% above the HD 8790M itself. The AMD card's average of 5691 is 0.4% below the Intel part and 0.7% below the NVIDIA GeForce GTX 550 Ti (5731). The only notable separation comes from the NVIDIA Quadro M500M, which trails the GTX 670MX by 2.1% and the HD 8790M by 1.6%.

Specification Differences

The two cards diverge on nearly every measurable specification. The GTX 670MX uses the GK104 chip with 3,540 million transistors on a 294 mm² die, while the HD 8790M uses the Mars chip with 950 million transistors on a 77 mm² die. Clock speeds favor AMD: 850 MHz base and 900 MHz boost versus NVIDIA's flat 601 MHz. Memory configurations differ as well — the GTX 670MX offers 3 GB of GDDR5 on a 192-bit bus for 67.20 GB/s bandwidth, while the HD 8790M provides 2 GB of GDDR5 on a 128-bit bus for 64.00 GB/s bandwidth.

The compute resource gap is substantial. NVIDIA fields 960 shading units, 80 TMUs, and 24 ROPs; AMD counters with 384 shading units, 24 TMUs, and 8 ROPs. Resulting throughput figures favor NVIDIA heavily: 12.02 GPixel/s versus 7.200 GPixel/s pixel rate, 48.08 GTexel/s versus 21.60 GTexel/s texture rate, and 1,153.9 GFLOPS versus 691.2 GFLOPS FP32. The GTX 670MX draws 75 W and uses no power connectors, while the HD 8790M's TDP is not specified but it ships as an MXM module. The NVIDIA card is from the GeForce 600M generation, released in 2012, while the HD 8790M belongs to the Solar System (HD 8700M) generation from 2013. Both are end-of-life products.

FAQ

Q: Which GPU has higher raw compute performance?

A: The NVIDIA GeForce GTX 670MX is decisively ahead in raw throughput, with 1,153.9 GFLOPS of FP32 performance versus 691.2 GFLOPS for the AMD Radeon HD 8790M. It also leads in pixel rate (12.02 vs 7.200 GPixel/s) and texture rate (48.08 vs 21.60 GTexel/s).

Q: Why does the AMD card win in Vulkan if it has fewer cores?

A: Despite having 384 shading units versus 960 for NVIDIA, the HD 8790M's higher clock speeds (900 MHz boost versus 601 MHz) and GCN 1.0 architecture's Vulkan efficiency allow it to score 6365 in Geekbench Vulkan, which is 16.5% higher than the GTX 670MX's 5316.

Q: Which card has more memory and bandwidth?

A: The NVIDIA GeForce GTX 670MX has 3 GB of GDDR5 on a 192-bit bus, yielding 67.20 GB/s bandwidth. The AMD Radeon HD 8790M has 2 GB of GDDR5 on a 128-bit bus, providing 64.00 GB/s. NVIDIA leads in both capacity and bandwidth.

Q: Are these cards comparable in overall performance?

A: Yes. Their average benchmark scores are nearly identical — 5721 for the GTX 670MX and 5691 for the HD 8790M, a difference of just 0.5%. Both sit at the 33rd percentile of all GPUs.

Q: What is the architectural generation difference?

A: The GTX 670MX uses NVIDIA's Kepler architecture on the GK104 chip, while the HD 8790M uses AMD's GCN 1.0 architecture on the Mars chip. Both are built on TSMC's 28 nm process, with the AMD chip being significantly smaller at 77 mm² versus 294 mm².

Q: Which card is better for modern Vulkan-based games?

A: Based on benchmark data, the AMD Radeon HD 8790M is clearly better for Vulkan workloads, winning the Geekbench Vulkan test by 16.5%. The GTX 670MX's Vulkan score of 5316 falls substantially behind, making AMD the preferable choice for Vulkan-focused applications.