AMD Radeon HD 8850M vs NVIDIA GeForce GTX 660 Comparison

The NVIDIA GeForce GTX 660 and the AMD Radeon HD 8850M occupy very different corners of the same 28 nm era, and the database makes that contrast immediately obvious. One is a desktop card with a 140 W thermal profile and a 221 mm² die; the other is a mobile part built on a 123 mm² chip. Both are end-of-life, both carry 2 GB of GDDR5, and both connect over PCIe 3.0 x16. Yet the recorded data shows the GTX 660 pulling decisively ahead wherever the two meet in measurement. The question worth investigating is not really who wins, but how large the gap is and what it implies about the two design philosophies behind these chips.

Where Each One Wins

The head-to-head record is unambiguous: the GTX 660 wins the only test where both parts have recorded scores. In Geekbench OpenCL, the GTX 660 posts 11347 against the HD 8850M's 7447, a 52.4 percent advantage for the NVIDIA card. There are no benchmark wins recorded for the HD 8850M in this comparison.

That single data point is consistent with the broader context in the database. The GTX 660 sits at the 45th percentile against all GPUs with an average benchmark score of 9022, while the HD 8850M sits at the 40th percentile with an average of 7447. The percentile gap is modest, only five points, which is curious given the size of the head-to-head delta. One explanation the data suggests: the GTX 660's average is dragged down by its relatively low Geekbench Metal score of 4305, a test where the AMD part has no recorded entry at all. Strip that out and the NVIDIA card's remaining scores, 11347 in OpenCL and 11415 in Vulkan, tower over the HD 8850M's solitary OpenCL result.

So the use-case split is straightforward. For any workload measured here, compute via OpenCL or Vulkan, the GTX 660 is the stronger choice. The HD 8850M's case rests entirely on what it is rather than what it scores: a mobile chip, with no power connectors listed and no slot width, designed for machines where a dual-slot desktop card drawing from a 6-pin connector is simply not an option.

Architecture Differences

Both GPUs were fabricated by TSMC on a 28 nm process, which makes the architectural contrast cleaner: same node, very different silicon budgets. The GTX 660 uses the GK106 chip on NVIDIA's Kepler architecture, part of the GeForce 600 generation, with 2,540 million transistors on a 221 mm² die. That works out to a transistor density of 11.5M per mm². The HD 8850M uses the Venus chip on AMD's GCN 1.0 architecture, from the Solar System generation of mobile parts, with 1,500 million transistors on a 123 mm² die, giving a slightly higher density of 12.2M per mm².

The resource counts tell the story of the desktop-versus-mobile divide. The GTX 660 fields 960 shading units, 80 texture mapping units, and 24 render output units. The HD 8850M counters with 640 shading units, 40 TMUs, and 16 ROPs. In proportional terms the NVIDIA part has 50 percent more shading units, double the TMUs, and 50 percent more ROPs. Clocks widen the gap further: the GTX 660 runs a 980 MHz base and 1032 MHz boost, against 575 MHz base and 625 MHz boost for the Venus chip.

Memory is another differentiator. Both carry 2 GB of GDDR5, but the GTX 660's 192-bit bus at 1502 MHz (6 Gbps effective) delivers 144.2 GB/s of bandwidth, more than double the HD 8850M's 64.00 GB/s from its 128-bit bus at 1000 MHz (4 Gbps effective). The throughput figures follow: 20.64 GPixel/s pixel rate and 82.56 GTexel/s texture rate for the GTX 660, against 10.00 GPixel/s and 25.00 GTexel/s for the HD 8850M. Raw FP32 compute lands at 1.981 TFLOPS for NVIDIA versus 800.0 GFLOPS for AMD, roughly a 2.5-to-1 ratio.

Feature support is close but not identical. Both report DirectX 12 capability, with the GTX 660 at feature level 11_0 and the HD 8850M at 11_1, a point in AMD's favor on paper. Both list OpenGL 4.6, while Vulkan support differs slightly: version 1.2.175 on the NVIDIA card versus 1.2.170 on the AMD part. Neither GPU has RT cores or tensor cores, as expected for this generation.

The I/O picture also diverges. The GTX 660 is a dual-slot card with one 6-pin power connector, a 140 W TDP, a suggested 300 W PSU, and a length of 241 mm (9.5 inches). Its display outputs include two DVI ports, HDMI 1.4a, and DisplayPort 1.2. The HD 8850M has no recorded TDP, connectors, slot width, dimensions, or display outputs in the database, fitting for an integrated mobile solution rather than a retail card.

Head-to-Head Benchmarks

Only one head-to-head test exists in the database, but it is decisive. In Geekbench OpenCL, the GTX 660 scores 11347 and the HD 8850M scores 7447. The delta is 52.4 percent in NVIDIA's favor, the largest single number in this comparison and the clearest verdict the data offers. For any OpenCL-accelerated workload, the GTX 660 delivers roughly one and a half times the throughput of the HD 8850M.

Context from each card's rival list reinforces the result. The GTX 660's average score of 9022 sits within a fraction of a percent of some unusual neighbors: the NVIDIA TITAN V CEO Edition at 9037 (a 0.2 percent gap) and the GeForce GTX 560 at 9058 (0.4 percent). It is narrowly ahead of the AMD Radeon 550X (1.2 percent) and the Radeon Pro WX 5100 (1.8 percent). The HD 8850M's average of 7447 puts it in a cluster with the Intel UHD Graphics 750 (0.1 percent), the AMD Radeon R7 350 (0.3 percent), the NVIDIA GeForce GTX 1650 (0.3 percent behind), and the Intel Arc A310 (1.4 percent ahead). These are integrated and entry-level comparisons, which frames the Venus chip's competitive neighborhood.

The GTX 660's own benchmark spread invites scrutiny. Its Vulkan score of 11415 essentially matches its OpenCL result of 11347, indicating consistent throughput across the two compute APIs. The Metal score of 4305 is dramatically lower, less than half the OpenCL figure, which suggests that API rather than raw hardware capability is the limiting factor there. No comparable split exists for the HD 8850M, since only its OpenCL result is recorded.

FAQ

Q: Which GPU is faster in the benchmarks?

A: The NVIDIA GeForce GTX 660. It wins the only shared test, Geekbench OpenCL, by 52.4 percent (11347 versus 7447) and holds a higher average benchmark score, 9022 versus 7447.

Q: Do both GPUs have the same amount of memory?

A: Yes, both ship with 2 GB of GDDR5. The difference is in bandwidth: the GTX 660 provides 144.2 GB/s across a 192-bit bus, while the HD 8850M provides 64.00 GB/s across a 128-bit bus.

Q: Are they built on the same manufacturing process?

A: Yes. Both are TSMC 28 nm parts, but the GTX 660's GK106 die is much larger at 221 mm² with 2,540 million transistors, versus the HD 8850M's 123 mm² Venus die with 1,500 million transistors.

Q: Which card supports the newer DirectX feature level?

A: The HD 8850M lists DirectX 12 with feature level 11_1, while the GTX 660 lists feature level 11_0. Both support OpenGL 4.6.

Q: How much power does each one need?

A: The GTX 660 has a 140 W TDP, one 6-pin power connector, and a suggested 300 W PSU. The database records no TDP or power connector data for the HD 8850M, consistent with its mobile designation.

Q: Where does each GPU rank against all GPUs in the database?

A: The GTX 660 sits at the 45th percentile; the HD 8850M sits at the 40th percentile.

Specification Differences

• Manufacturer and chip: NVIDIA GK106 (Kepler, GeForce 600) versus AMD Venus (GCN 1.0, HD 8800M)
• Release date: September 2012 for the GTX 660, March 2013 for the HD 8850M
• Die and transistors: 221 mm² and 2,540 million versus 123 mm² and 1,500 million
• Transistor density: 11.5M/mm² versus 12.2M/mm²
• Clocks: 980 MHz base and 1032 MHz boost versus 575 MHz base and 625 MHz boost
• Memory bus and bandwidth: 192-bit, 144.2 GB/s versus 128-bit, 64.00 GB/s
• Memory speed: 6 Gbps effective versus 4 Gbps effective
• Shading units, TMUs, ROPs: 960 / 80 / 24 versus 640 / 40 / 16
• Pixel and texture rates: 20.64 GPixel/s and 82.56 GTexel/s versus 10.00 GPixel/s and 25.00 GTexel/s
• FP32 compute: 1.981 TFLOPS versus 800.0 GFLOPS
• TDP and power: 140 W, one 6-pin connector, 300 W suggested PSU versus no recorded data
• Form factor: dual-slot, 241 mm long, versus no recorded dimensions
• Display outputs: two DVI, HDMI 1.4a, DisplayPort 1.2 versus none listed
• API support: DirectX 12 (11_0) and Vulkan 1.2.175 versus DirectX 12 (11_1) and Vulkan 1.2.170
• Launch MSRP: 229 USD for the GTX 660; no figure recorded for the HD 8850M

The Verdict

The data leaves little room for debate on performance. The GTX 660 wins the sole head-to-head benchmark by 52.4 percent, leads in every specification that affects throughput, from shading units to memory bandwidth, and ranks five percentile points higher against the full GPU database. Anyone choosing between these two for compute or gaming workloads should pick the GTX 660, provided the platform can host it: it needs a dual-slot space of 241 mm, a 6-pin power connector, and a 300 W-class power supply.

The HD 8850M's case is not about winning benchmarks. It is a mobile GCN 1.0 part with no external power requirements recorded, a smaller die, and a DirectX feature level one step higher on paper. For systems where the GTX 660 physically cannot fit, that distinction is the whole argument. But on the numbers alone, across the one test they share and the specification sheet as a whole, the GeForce GTX 660 is the stronger GPU in every measured dimension, and by a wide margin.