NVIDIA GRID K2 vs NVIDIA Tesla C2070 Comparison

FAQ

Q: Which GPU has the higher OpenCL benchmark score?

A: The NVIDIA GRID K2 scores 10,602 in Geekbench OpenCL, while the NVIDIA Tesla C2070 scores 9,716. The GRID K2 wins this head-to-head comparison by 8.4%.

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

A: The Tesla C2070 offers 6 GB of GDDR5 memory on a 384-bit bus, delivering 143.4 GB/s of bandwidth. The GRID K2 has 4 GB of GDDR5 memory on a 256-bit bus, providing 160.0 GB/s of bandwidth, which is higher despite the smaller capacity.

Q: What are the transistor counts and die sizes of each GPU?

A: The Tesla C2070 uses the GF100 chip with 3,100 million transistors on a 529 mm² die. The GRID K2 uses the GK104 chip with 3,540 million transistors on a 294 mm² die, resulting in a much higher transistor density of 12.0M per mm² versus 5.9M per mm².

Q: Which card supports PCIe 3.0?

A: The GRID K2 features a PCIe 3.0 x16 bus interface, while the Tesla C2070 is limited to PCIe 2.0 x16. This gives the GRID K2 a more modern interconnect for data transfer.

Q: What display outputs do these cards provide?

A: The Tesla C2070 has a single DVI output. The GRID K2 has no display outputs at all, indicating it is intended purely for compute or virtualized workloads rather than direct display attachment.

Q: What is the percentile ranking of each GPU in the database?

A: The Tesla C2070 sits at the 47th percentile of all GPUs, while the GRID K2 sits at the 42nd percentile. Despite the GRID K2 winning the direct OpenCL comparison, the C2070 ranks slightly higher overall in the database distribution.

Where Each One Wins

The GRID K2 takes the sole direct benchmark victory in this comparison. In the Geekbench OpenCL test, the GRID K2 records 10,602 points against the Tesla C2070's 9,716 points, a margin of 8.4%. This makes the GRID K2 the clear choice for raw compute throughput in OpenCL workloads, at least based on the recorded measurement.

The Tesla C2070, however, holds advantages in areas that the head-to-head benchmark does not capture. It offers 6 GB of memory versus 4 GB on the GRID K2, which can matter for workloads with large datasets that exceed the smaller frame buffer. It also has a DVI output, making it usable in configurations that require a local display, whereas the GRID K2 has no display outputs at all. The C2070's larger memory bus of 384 bits versus 256 bits may also favor certain access patterns, even though the GRID K2's effective bandwidth is higher.

In terms of database percentile placement, the Tesla C2070 ranks at the 47th percentile, slightly ahead of the GRID K2's 42nd percentile. This suggests that in the broader pool of all GPUs, the C2070 sits a bit higher in the distribution, even though the direct comparison favors the GRID K2. The GRID K2's average benchmark score of 8,080 is dragged down by its additional Metal benchmark result of 5,557, which is not available for the C2070.

Architecture Differences

The two GPUs represent different NVIDIA architectures. The Tesla C2070 is built on the Fermi architecture using the GF100 chip, while the GRID K2 uses the Kepler architecture with the GK104 chip. This architectural generation gap is significant: Fermi was NVIDIA's first GPU architecture with unified L2 cache and concurrent kernel execution, while Kepler introduced significant efficiency improvements.

The manufacturing process differs substantially. The C2070 uses a 40 nm process at TSMC, while the GRID K2 uses a 28 nm process, also at TSMC. This node shrink allows the GRID K2 to pack 3,540 million transistors into a 294 mm² die, compared to the C2070's 3,100 million transistors on a much larger 529 mm² die. The transistor density tells the story clearly: the GRID K2 achieves 12.0M transistors per mm² versus 5.9M per mm² for the C2070.

Shader and texture resources are dramatically different. The GRID K2 has 1,536 shading units and 128 texture mapping units, while the C2070 has 448 shading units and 56 TMUs. This gives the GRID K2 more than three times the shader count and more than double the texture units. However, the C2070 has 48 ROPs versus 32 on the GRID K2, meaning the C2070 retains an advantage in certain rasterization output stages.

The memory architecture also differs. The C2070 has a 384-bit memory bus with 6 GB of GDDR5, while the GRID K2 has a 256-bit bus with 4 GB. Despite the narrower bus, the GRID K2 achieves higher memory bandwidth at 160.0 GB/s versus 143.4 GB/s, thanks to its faster memory clock of 1,250 MHz compared to the C2070's 747 MHz. The GRID K2 also supports PCIe 3.0, while the C2070 is limited to PCIe 2.0.

API support shows a notable difference in Vulkan: the GRID K2 supports Vulkan 1.2.175, while the C2070 has no Vulkan support listed. Both cards support DirectX 12 (11_0) and OpenGL 4.6.

Specification Differences

The two cards differ across nearly every specification category. The C2070 is a Fermi architecture GPU with a 40 nm process node, while the GRID K2 is Kepler with a 28 nm node. The C2070 has 3,100 million transistors on a 529 mm² die; the GRID K2 has 3,540 million on 294 mm². Memory clock speeds are 747 MHz for the C2070 versus 1,250 MHz for the GRID K2. Memory capacity is 6 GB versus 4 GB, with bus widths of 384 bit versus 256 bit, and bandwidth of 143.4 GB/s versus 160.0 GB/s.

The C2070 has 448 shading units, 56 TMUs, and 48 ROPs. The GRID K2 has 1,536 shading units, 128 TMUs, and 32 ROPs. Pixel rates are 16.07 GPixel/s for the C2070 versus 23.84 GPixel/s for the GRID K2. Texture rates are 32.14 GTexel/s versus 95.36 GTexel/s. FP32 compute is 1,027.7 GFLOPS for the C2070 versus 2.289 TFLOPS for the GRID K2.

Power consumption is 238 W for the C2070 and 225 W for the GRID K2. Both are dual-slot cards requiring a 550 W power supply, with identical power connectors of 1x 6-pin plus 1x 8-pin. The C2070 uses PCIe 2.0 x16 while the GRID K2 uses PCIe 3.0 x16. The C2070 has one DVI output; the GRID K2 has no outputs. The C2070 supports DirectX 12 (11_0) and OpenGL 4.6, with no Vulkan; the GRID K2 adds Vulkan 1.2.175. Physical length is 248 mm for the C2070 versus 267 mm for the GRID K2.

Head-to-Head Benchmarks

The only direct benchmark comparison in the database is the Geekbench OpenCL test. The GRID K2 scores 10,602, beating the Tesla C2070's 9,716 by 8.4%. This is a decisive win for the GRID K2 in compute performance. The delta of 8.4% is notable given that the two cards have very different architectural designs, with the GRID K2's Kepler architecture and far higher shader count likely driving the advantage.

Looking at the nearest rivals for context, the Tesla C2070's closest competitor is the NVIDIA Tesla M10 with a score of 9,724, a negligible 0.1% difference. The Quadro P4000 scores 9,665, putting it 0.5% behind the C2070, and the Radeon Pro WX 2100 scores 9,653, 0.7% behind. The GeForce GTX 1070 scores 9,780, which is 0.7% ahead of the C2070. These tight margins show that the C2070 sits squarely in a competitive band of GPUs around the 9,700 score level.

The GRID K2's nearest rivals tell a different story. Its average score of 8,080 places it near the GeForce GTX 650 Ti Boost at 8,067 (0.2% behind), the GeForce 945M at 8,099 (0.2% ahead), the GeForce GTX 650 Ti at 8,053 (0.3% behind), and the GeForce GTX 880M at 8,040 (0.5% behind). The gap between the GRID K2's OpenCL score of 10,602 and its average score of 8,080 is explained by its Metal benchmark result of 5,557, which significantly lowers the average. This means the GRID K2's OpenCL performance is much stronger than its overall average suggests.

The Verdict

The data points to a clear split in use cases. For pure OpenCL compute workloads, the NVIDIA GRID K2 is the better choice, posting a 10,602 score that beats the Tesla C2070 by 8.4%. Its Kepler architecture delivers over double the FP32 throughput (2.289 TFLOPS versus 1,027.7 GFLOPS), higher texture and pixel rates, and more than triple the shading units. The GRID K2 also offers PCIe 3.0 and Vulkan support, making it the more modern compute card.

However, the Tesla C2070 has its own strengths that matter for specific scenarios. It provides 6 GB of memory versus 4 GB, which is critical for workloads that need to hold larger datasets in GPU memory. It also has a DVI output, enabling direct display connection, while the GRID K2 has none. The C2070's 48 ROPs exceed the GRID K2's 32, and its 384-bit memory bus may benefit certain access patterns, even if raw bandwidth is lower.

For users needing maximum OpenCL performance and modern features, the GRID K2 is the obvious pick. For those who require larger memory capacity or a display output, the Tesla C2070 remains viable despite its older architecture. The database's percentile rankings show the C2070 at the 47th percentile versus the GRID K2's 42nd, but this is partly skewed by the GRID K2's low Metal score. In the direct OpenCL comparison that matters for compute, the GRID K2 wins decisively.