AMD Radeon HD 6950 vs NVIDIA GeForce GTX 670MX Comparison

Head-to-Head Benchmarks

The recorded data includes a single direct comparison between these two graphics cards, the Geekbench OpenCL test. In this workload, the AMD Radeon HD 6950 scores 6210, while the NVIDIA GeForce GTX 670MX scores 6125. That gives AMD a 1.4% advantage, a narrow margin that places both cards in essentially the same performance tier for compute workloads.

The AMD card's score of 6210 sits at the 36th percentile among all GPUs in the database. Its nearest rivals bracket it tightly: the AMD FirePro W600 scores 6223 (0.2% higher), the NVIDIA GeForce RTX 4070 Ti SUPER AD102 scores 6270 (0.9% higher), and the NVIDIA Quadro K620 scores 6282 (1.1% higher). The fact that a 2010 desktop card lands within roughly one percent of much newer parts in this particular OpenCL test says more about the test's characteristics than about raw gaming performance, but it does indicate that the HD 6950 retains respectable compute throughput.

The GTX 670MX, meanwhile, scores 6125 in OpenCL, placing it at the 33rd percentile overall. Its closest competitors in the database include the Intel Iris Pro Graphics P6300 at 5712 (0.2% lower), the NVIDIA GeForce GTX 550 Ti at 5731 (0.2% higher), the AMD Radeon HD 8790M at 5691 (0.5% higher), and the NVIDIA Quadro M500M at 5604 (2.1% higher). The GTX 670MX's average benchmark score across all recorded tests, including its Vulkan result, drops to 5721, which reflects the inclusion of that second workload where it scores 5316.

The NVIDIA card also has a Vulkan result in the database, scoring 5316. The AMD Radeon HD 6950 has no recorded Vulkan score and also lists no Vulkan API support in its specification data, so no direct comparison is possible in that API. The difference in API support is one of the more significant separators between these two cards, as the GTX 670MX explicitly supports Vulkan 1.2.175 while the HD 6950 lists no Vulkan support at all.

When considering the OpenCL head-to-head alone, the margin is small enough that run-to-run variance could plausibly flip the result. The delta of 1.4% is within the range of differences seen among the HD 6950's own nearest rivals, where deltas range from 0.2% to 1.1%. The practical takeaway is that neither card holds a commanding lead in this compute workload, but the AMD part does edge ahead in the recorded measurement.

FAQ

Q: Which card wins the Geekbench OpenCL benchmark?

A: The AMD Radeon HD 6950 wins with a score of 6210 versus 6125 for the NVIDIA GeForce GTX 670MX, a difference of 1.4%.

Q: Does the GTX 670MX support Vulkan?

A: Yes, the NVIDIA card lists Vulkan support at version 1.2.175 and has a recorded Geekbench Vulkan score of 5316. The AMD Radeon HD 6950 lists no Vulkan support in the database.

Q: How do these cards compare to their nearest rivals?

A: The HD 6950's nearest rival, the AMD FirePro W600, scores 6223, which is 0.2% higher. The GTX 670MX's nearest rival, the Intel Iris Pro Graphics P6300, scores 5712, which is 0.2% lower than the NVIDIA card's average score.

Q: What are the average benchmark scores for each card?

A: The AMD Radeon HD 6950 has an average benchmark score of 6210, based on its single recorded OpenCL result. The NVIDIA GeForce GTX 670MX has an average score of 5721, which combines its OpenCL score of 6125 and its Vulkan score of 5316.

Q: Which card has the higher pixel fill rate?

A: The AMD Radeon HD 6950 has a pixel rate of 25.60 GPixel/s, substantially higher than the GTX 670MX's 12.02 GPixel/s.

Q: What is the transistor density difference between the two chips?

A: The GTX 670MX's GK104 chip has a transistor density of 12.0M per mm², while the HD 6950's Cayman chip has 6.8M per mm².

Where Each One Wins

The AMD Radeon HD 6950 wins in raw compute throughput as measured by OpenCL, and it also leads in several specification-level metrics that point toward fill-rate-heavy workloads. Its pixel rate of 25.60 GPixel/s is more than double the GTX 670MX's 12.02 GPixel/s, and its texture rate of 70.40 GTexel/s comfortably exceeds the NVIDIA part's 48.08 GTexel/s. The 2.253 TFLOPS FP32 throughput is also nearly double the GTX 670MX's 1,153.9 GFLOPS. These figures suggest the AMD card would hold an advantage in applications that scale with raw shading throughput and fill rate, at least where its older architecture is well supported.

The AMD card also has the higher memory bandwidth at 160.0 GB/s over a 256-bit bus, compared to 67.20 GB/s over a 192-bit bus for the NVIDIA part. This is a large gap, roughly 2.4 times the bandwidth, which could matter in memory-bound scenarios such as high-resolution texture streaming or compute kernels that access large data sets. Its 2 GB of GDDR5 is smaller than the GTX 670MX's 3 GB, however, so the NVIDIA card can hold more data on board even if it moves that data more slowly.

The NVIDIA GeForce GTX 670MX wins in areas that the OpenCL score does not capture. It has a Vulkan score of 5316, and it lists Vulkan 1.2.175 support, which opens up a broader range of modern API workloads. The HD 6950 lists no Vulkan support at all, capping its software compatibility. The GTX 670MX also has a much lower TDP at 75 W versus 200 W, and it requires no external power connectors, making it viable in portable systems. Its PCIe 3.0 x16 interface is a generation newer than the HD 6950's PCIe 2.0 x16, which can matter for data transfer in bandwidth-sensitive tasks even if the GPU itself is the bottleneck in most rendering scenarios.

The GTX 670MX also supports newer API versions in some categories. It lists DirectX 12 (11_0) and OpenGL 4.6, while the HD 6950 lists DirectX 11.2 (11_0) and OpenGL 4.4. For software that relies on these newer API entry points, the NVIDIA card is the only one of the two that can run them.

Specification Differences

The two cards differ across nearly every major specification category. The AMD Radeon HD 6950 uses the Cayman chip on a 40 nm process from TSMC, while the GTX 670MX uses the GK104 chip on a 28 nm process, also from TSMC. The AMD chip contains 2,640 million transistors on a 389 mm² die, giving it a density of 6.8M per mm². The NVIDIA chip packs 3,540 million transistors onto a smaller 294 mm² die, yielding 12.0M per mm².

Memory configurations diverge sharply. The HD 6950 has 2 GB of GDDR5 on a 256-bit bus with 160.0 GB/s bandwidth. The GTX 670MX has 3 GB of GDDR5 on a 192-bit bus with 67.20 GB/s bandwidth. The memory clock values also differ: the AMD card runs at 1250 MHz (5 Gbps effective), while the NVIDIA card runs at 700 MHz (2.8 Gbps effective).

Shader resources favor AMD. The HD 6950 has 1408 shading units, 88 TMUs, and 32 ROPs. The GTX 670MX has 960 shading units, 80 TMUs, and 24 ROPs. The resulting throughput rates are 25.60 GPixel/s and 70.40 GTexel/s for AMD versus 12.02 GPixel/s and 48.08 GTexel/s for NVIDIA. FP32 compute is 2.253 TFLOPS for AMD versus 1,153.9 GFLOPS for NVIDIA.

Power and physical requirements are heavily in NVIDIA's favor. The HD 6950 has a 200 W TDP, requires two 6-pin power connectors, and needs a 550 W suggested PSU. The GTX 670MX has a 75 W TDP, requires no external power connectors, and has no listed PSU suggestion. The AMD card is dual-slot with dimensions of 286 mm length, 126 mm height, and 42 mm width. The NVIDIA card has no listed dimensions, consistent with its mobile-oriented design and "Portable Device Dependent" display outputs.

The bus interface differs as well: PCIe 2.0 x16 for AMD versus PCIe 3.0 x16 for NVIDIA. Display outputs on the HD 6950 include 2x DVI, 1x HDMI 1.4a, and 2x mini-DisplayPort 1.2. The GTX 670MX's outputs are listed simply as "Portable Device Dependent." Release dates are about 22 months apart, with the HD 6950 launching on December 13, 2010, and the GTX 670MX on September 30, 2012.

Architecture Differences

The AMD Radeon HD 6950 is built on TeraScale 3, part of the Northern Islands generation (HD 6900 family), with the Cayman chip. The NVIDIA GeForce GTX 670MX is built on Kepler, part of the GeForce 600M generation, with the GK104 chip. These are fundamentally different GPU designs from different eras, and the architectural choices reflect their target markets: AMD's card is a desktop performance part, while NVIDIA's is a mobile part.

The manufacturing process is a clear generational step. AMD uses 40 nm, while NVIDIA uses 28 nm. The smaller process allows NVIDIA to pack more transistors, 3,540 million versus 2,640 million, into a smaller die, 294 mm² versus 389 mm². That translates to nearly double the transistor density, 12.0M per mm² versus 6.8M per mm². The newer process also explains, at least in part, the enormous TDP difference: 75 W for NVIDIA versus 200 W for AMD.

The memory architectures reflect different priorities. AMD's 256-bit bus with 160.0 GB/s bandwidth is a classic desktop design favoring raw throughput. NVIDIA's 192-bit bus with 67.20 GB/s bandwidth is a more constrained mobile design, but it pairs with a larger 3 GB frame buffer. The result is that AMD moves data faster while NVIDIA stores more of it locally.

API support shows the Kepler architecture's more modern feature set. NVIDIA lists DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. AMD lists DirectX 11.2 (11_0) and OpenGL 4.4, with no Vulkan support. The TeraScale 3 architecture predates Vulkan's specification, so this is an expected limitation, but it is a real one for software compatibility.

The feature sets are otherwise relatively sparse in the recorded data. Neither card lists ray tracing cores or tensor cores, which is consistent with their respective generations. Neither lists FP16 compute capability. The GTX 670MX has a base and boost clock of 601 MHz, while the HD 6950 has no base or boost clock listed, only its memory clock of 1250 MHz.

The Verdict

The data supports a clear split based on workload type. For OpenCL compute performance, the AMD Radeon HD 6950 takes the lead with a 1.4% advantage in the head-to-head benchmark, and its specification-level throughput metrics are dramatically higher. The 2.253 TFLOPS FP32 figure, 25.60 GPixel/s pixel rate, and 70.40 GTexel/s texture rate all point to a card that can push raw geometry and pixels far faster than the GTX 670MX. Anyone running compute workloads that are well supported on TeraScale 3, or fill-rate-heavy rendering tasks, should favor the AMD card based on the recorded data.

The NVIDIA GeForce GTX 670MX is the better choice for modern API compatibility. It supports Vulkan 1.2.175, DirectX 12 (11_0), and OpenGL 4.6, and it has a recorded Vulkan score of 5316. The AMD card has no Vulkan support and older DirectX and OpenGL versions. For software that requires Vulkan, the GTX 670MX is the only option between the two.

The GTX 670MX also wins decisively on power and physical integration. Its 75 W TDP and lack of power connectors make it viable in systems where the HD 6950's 200 W TDP and two 6-pin connectors would be impractical. The lack of recorded dimensions for the NVIDIA card is consistent with a mobile part designed to fit into a laptop chassis, while the HD 6950 is a 286 mm dual-slot desktop card.

The larger frame buffer on the GTX 670MX, 3 GB versus 2 GB, could help in scenarios where capacity matters more than bandwidth, though the AMD card's 160.0 GB/s bandwidth versus 67.20 GB/s suggests the NVIDIA part will be more constrained in moving data around.

The verdict is a matter of context. The AMD Radeon HD 6950 delivers substantially higher raw throughput and wins the only direct benchmark comparison, making it the compute-oriented choice. The NVIDIA GeForce GTX 670MX offers modern API support, far lower power draw, and a larger memory pool, making it the compatibility-oriented and mobile-friendly choice. Neither card is dominant across all categories, and the selection should depend on whether the workload prioritizes raw throughput or software compatibility and power efficiency.