NVIDIA GeForce GTX 660 vs NVIDIA GRID K2 Comparison

Head-to-Head Benchmarks

The two GPUs split their recorded benchmark comparisons exactly one win apiece, but the margins tell very different stories. In the Geekbench Metal test, the NVIDIA GRID K2 posts 5557 points against 4305 for the GeForce GTX 660, a 22.5% advantage. That is a decisive gap, placing the GRID K2 clearly ahead in this particular workload. The GeForce GTX 660 counters in Geekbench OpenCL, scoring 11347 versus 10602, a 7% lead. The asymmetry is notable: the GRID K2 wins by a wide margin in Metal, while the GTX 660 wins by a narrower but still meaningful margin in OpenCL.

Looking at overall averages, the GeForce GTX 660 holds an average benchmark score of 9022, while the GRID K2 trails at 8080. That is a difference of roughly 11.7% in favor of the GTX 660 when aggregating across all recorded tests. The percentile rankings reinforce this: the GTX 660 sits at the 45th percentile against all GPUs, while the GRID K2 sits at the 42nd. The GTX 660's nearest rivals in the database include the NVIDIA TITAN V CEO Edition at 9037 (0.2% ahead), the NVIDIA GeForce GTX 560 at 9058 (0.4% ahead), the AMD Radeon 550X at 8918 (1.2% behind), and the AMD Radeon Pro WX 5100 at 8863 (1.8% behind). The GRID K2's nearest rivals are clustered much tighter: the NVIDIA GeForce GTX 650 Ti Boost at 8067 (0.2% ahead), the NVIDIA GeForce 945M at 8099 (0.2% behind), the NVIDIA GeForce GTX 650 Ti at 8053 (0.3% ahead), and the NVIDIA GeForce GTX 880M at 8040 (0.5% ahead). The data suggests the GRID K2 lands in a performance tier populated by mid-range GeForce parts, whereas the GTX 660 sits in a slightly higher bracket.

What does this imply? The GRID K2's single strong Metal showing does not translate into a higher aggregate score, because its OpenCL result is substantially lower than the GTX 660's. The GTX 660's OpenCL score of 11347 is nearly 7% higher than the GRID K2's 10602, and that advantage outweighs the GRID K2's Metal lead when averaged out. The database records no Vulkan score for the GRID K2, while the GTX 660 posts 11415 in that test, further widening the aggregate gap. It is curious that the GRID K2, with substantially more shading units and memory bandwidth, fails to convert that hardware into a consistent benchmark advantage.

Architecture Differences

Both GPUs share the same Kepler architecture and are fabricated on TSMC's 28 nm process, so the fundamental design philosophy is identical. The chips, however, differ in size and scale. The GeForce GTX 660 uses the GK106 chip, measuring 221 mm² in die size, containing 2,540 million transistors at a density of 11.5M per mm². The GRID K2 uses the larger GK104 chip, measuring 294 mm², with 3,540 million transistors at a density of 12.0M per mm². The GRID K2's chip is physically larger by about a third by area, and it packs roughly 39% more transistors.

The compute resources scale accordingly. The GRID K2 has 1536 shading units, 128 texture mapping units, and 32 render output units. The GTX 660 has 960 shading units, 80 TMUs, and 24 ROPs. That means the GRID K2 fields 60% more shading units, 60% more TMUs, and 33% more ROPs. The theoretical peak rates follow: the GRID K2 reaches 23.84 GPixel/s pixel rate and 95.36 GTexel/s texture rate, while the GTX 660 reaches 20.64 GPixel/s and 82.56 GTexel/s. The GRID K2's FP32 throughput is 2.289 TFLOPS versus 1.981 TFLOPS for the GTX 660, a 15.5% advantage. Neither card features ray tracing cores or tensor cores, and neither lists FP16 performance.

Memory is another clear divergence. The GTX 660 uses 2 GB of GDDR5 on a 192-bit bus, delivering 144.2 GB/s of bandwidth. The GRID K2 doubles capacity to 4 GB, widens the bus to 256-bit, and achieves 160.0 GB/s. The GRID K2's memory clock is listed at 1250 MHz with 5 Gbps effective, while the GTX 660 runs at 1502 MHz with 6 Gbps effective. So the GRID K2 wins on capacity and total bandwidth, but the GTX 660 has faster memory clock speeds.

Clock speeds for the GPU core itself are only recorded for the GTX 660: 980 MHz base and 1032 MHz boost. The GRID K2 has no base or boost clock listed in the database. That absence makes direct clock-for-clock comparison impossible, but the measured benchmark scores provide the practical picture. The GRID K2's power envelope is much higher at 225 W TDP versus 140 W for the GTX 660, and it requires both a 6-pin and an 8-pin power connector, while the GTX 660 needs only a single 6-pin. The suggested PSU scales accordingly: 550 W for the GRID K2, 300 W for the GTX 660. Physical dimensions also differ: the GRID K2 is 267 mm long versus 241 mm for the GTX 660. Both are dual-slot cards.

The GRID K2 is clearly not a consumer graphics card. It has no display outputs at all, whereas the GTX 660 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. The GRID K2's generation label is "GRID (K2)", pointing to a virtualization or datacenter role. Both cards share identical API support: DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. Both are end-of-life products, with the GTX 660 released on 2012-09-05 and the GRID K2 on 2013-05-10.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA GeForce GTX 660, with an average benchmark score of 9022 compared to 8080 for the NVIDIA GRID K2.

Q: Why does the GRID K2 win the Metal benchmark if its overall average is lower?

A: The GRID K2 scores 5557 in Geekbench Metal versus 4305 for the GTX 660, a 22.5% lead. However, the GTX 660 counters with 11347 in OpenCL versus 10602, and also has a recorded Vulkan score of 11415, which the GRID K2 lacks. The aggregate average thus favors the GTX 660.

Q: How do their transistor counts compare?

A: The GRID K2's GK104 chip contains 3,540 million transistors, while the GTX 660's GK106 chip contains 2,540 million. The GRID K2 also has a larger die at 294 mm² versus 221 mm².

Q: Do both cards support the same graphics APIs?

A: Yes, both report DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175 in the database.

Q: Can the GRID K2 be used for direct display output?

A: No, the GRID K2 has no display outputs listed, while the GTX 660 provides 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

Q: What are the power requirements for each card?

A: The GTX 660 has a 140 W TDP and requires a 300 W suggested PSU with a single 6-pin connector. The GRID K2 has a 225 W TDP and requires a 550 W suggested PSU with one 6-pin and one 8-pin connector.

The Verdict

The data points to two different products for two different purposes. The GeForce GTX 660 is the better choice for general compute workloads as measured by the database: it wins the OpenCL test, holds a higher aggregate average score (9022 versus 8080), and ranks higher at the 45th percentile versus the 42nd. It also carries display outputs, making it a functional graphics card for direct use. The GRID K2, by contrast, dominates the Metal benchmark with a 22.5% margin, but its overall score is dragged down by a weaker OpenCL result and a missing Vulkan entry. Its 4 GB memory capacity and wider 256-bit bus suggest a design aimed at memory-heavy virtualized workloads rather than consumer rendering. For anyone comparing these purely on recorded benchmark performance, the GTX 660 is the more consistent performer. For scenarios where Metal performance is the priority, the GRID K2 is the clear winner, but its lack of display outputs and higher power demands make it a niche option. The GTX 660's nearest rival list includes the TITAN V CEO Edition at nearly the same score, which contextualizes its strong average. The GRID K2's nearest rivals are all lower-tier GeForce parts, confirming its aggregate placement is a step below.

Specification Differences

The two cards differ across nearly every hardware category. The GRID K2 uses the GK104 chip, while the GTX 660 uses GK106. Transistor count is 3,540 million versus 2,540 million, and die size is 294 mm² versus 221 mm². The GRID K2 has 1536 shading units, 128 TMUs, and 32 ROPs, compared to 960 shading units, 80 TMUs, and 24 ROPs. Pixel rate is 23.23 GPixel/s higher at 23.84, and texture rate is 12.0 GTexel/s higher at 95.36, while the GTX 660 posts 20.64 GPixel/s and 82.56 GTexel/s. FP32 throughput stands at 2.289 TFLOPS versus 1.981 TFLOPS.

Memory configurations differ: 4 GB versus 2 GB, 256-bit versus 192-bit bus width, 160.0 GB/s versus 144.2 GB/s bandwidth. Memory clocks are 1250 MHz at 5 Gbps effective for the GRID K2, and 1502 MHz at 6 Gbps effective for the GTX 660. Power profiles diverge sharply: 225 W TDP versus 140 W, 1x 6-pin + 1x 8-pin versus 1x 6-pin, 550 W suggested PSU versus 300 W. The GRID K2 measures 267 mm in length, the GTX 267 mm, while the GTX 660 measures 241 mm. The GTX 660 has display outputs, the GRID K2 has none. Release dates are 2013-05-10 for the GRID K2 and 2013-05-05 for the GTX 660, though the launch dates in the database are 2013-05-10 for the GRID K2 and 2013-09-05 for the GTX 660. The GTX 660 is also listed with a successor in the GeForce 700 series and a predecessor in GeForce 500.

The GTX 660's launch MSRP is 229 USD, and the GRID K2's is 5,199 USD, but the power and compute differences are the more meaningful points in the database.

The GTX 660 has a faster memory clock, a lower TDP, and a smaller physical footprint, while the GRID K2 counters with a larger bus and more memory. Neither listed base clocks for the GRID K2, which the GTX 660 has at 980 MHz base and 1032 MHz boost.

Where Each One Wins

The GRID K2 wins in any scenario that prioritizes Metal compute performance. Its 5557 Metal score is the single largest margin in the comparison, and its higher FP32 throughput, larger memory pool, and wider bus all point to workloads that scale with raw resources. The GTX 660 wins in OpenCL, posting a 7% higher score, and it holds the only Vulkan result in the comparison at 11415. Its lower power draw, smaller physical size, and display outputs make it the more practical card for a standard desktop environment. The GRID K2's lack of display outputs means it cannot serve as a primary graphics output; it is a compute accelerator. The GTX 660's 2 GB memory is half the GRID K2's 4 GB, which could matter for large datasets. The GTX 660 also wins on efficiency per watt by the numbers: 140 W for a 9022 average score versus 225 W for an 8080 average. In aggregate, the GTX 660 is the stronger all-around performer, while the GRID K2 is the specialist with a single, decisive victory.