NVIDIA GeForce GTX 960M vs NVIDIA Tesla K20c Comparison

FAQ

Q: What is the performance gap between the NVIDIA Tesla K20c and the GeForce GTX 960M in the recorded database?

A: In the only shared benchmark, Geekbench OpenCL, the Tesla K20c scores 11,479 points against 11,045 for the GTX 960M, a 3.9% advantage for the Tesla part.

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

A: The Tesla K20c sits at the 51st percentile, while the GTX 960M sits at the 46th percentile, placing the Tesla in a slightly higher overall standing among all recorded GPUs.

Q: Which GPU has the higher memory bandwidth?

A: The Tesla K20c has a 320-bit memory bus and 208.0 GB/s of bandwidth, while the GTX 960M has a 128-bit bus and 80.19 GB/s of bandwidth. The Tesla's bandwidth is roughly 2.6 times higher.

Q: What are the memory capacities of these two cards?

A: The Tesla K20c has 5 GB of GDDR5 memory, while the GTX 960M has 4 GB of GDDR5 memory.

Q: Do both cards support DirectX 12?

A: Yes, both list DirectX 12 (11_0) support in the database. Both also support OpenGL 4.6, and the GTX 960M lists Vulkan 1.4 support while the Tesla K20c lists Vulkan 1.2.175.

Q: What are the power requirements for each card?

A: The Tesla K20c has a 225 W TDP and requires a 550 W suggested power supply, plus one 6-pin and one 8-pin power connector. The GTX 960M has a 75 W TDP and no power connectors, as it is an MXM module.

Where Each One Wins

The recorded data shows a single head-to-head benchmark, and the Tesla K20c wins that test. However, the broader database profile reveals where each card has distinct strengths beyond that one score.

The Tesla K20c is built for compute throughput. Its 2,496 shading units, 208 texture mapping units, and 40 ROPs are all far higher than the GTX 960M's 640 shading units, 40 TMUs, and 16 ROPs. The Tesla's FP32 output of 3.524 TFLOPS is more than double the GTX 960M's 1.505 TFLOPS. For raw number crunching, OpenCL workloads, or any task that scales with shader count and memory bandwidth, the Tesla K20c is the clear pick. Its 5 GB frame buffer and 208.0 GB/s bandwidth also support larger datasets than the GTX 960M can hold.

The GTX 960M wins in a different category: deployment flexibility. It is an MXM module with no external power connectors, a 75 W TDP, and portable-device-dependent display outputs. This makes it suitable for laptops and compact systems where the Tesla's dual-slot, 267 mm length, and 225 W power draw would not fit. The GTX 960M also has a boost clock of 1176 MHz and a base clock of 1097 MHz, whereas the Tesla K20c has no recorded base or boost clock values, suggesting its clock behavior is less straightforward for comparison.

The GTX 960M also has a Vulkan 1.4 API listing, which is newer than the Tesla's Vulkan 1.2.175. For workloads that rely on the latest Vulkan features, the mobile card has an edge in software compatibility.

Architecture Differences

The Tesla K20c uses the GK110 chip under the Kepler architecture, while the GTX 960M uses the GM107 chip under the Maxwell architecture. Both are manufactured on a 28 nm process at TSMC, and both have the same transistor density of 12.6M per mm². The similarities end there.

The Tesla's GK110 is a large compute-oriented die. It packs 7,080 million transistors into a 561 mm² die. The GTX 960M's GM107 is much smaller, with 1,870 million transistors on a 148 mm² die. The transistor density is identical, but the physical scale is very different. The Tesla has nearly four times the transistor count and a die that is roughly 3.8 times larger.

Architecturally, Kepler and Maxwell differ in how they handle shader workloads. The Tesla K20c has 2,496 shading units, 208 TMUs, and 40 ROPs, giving it a much wider execution width. The GTX 960M has 640 shading units, 40 TMUs, and 16 ROPs, a narrower configuration that is more power-efficient per unit of work. Maxwell generally improves instruction scheduling and power efficiency over Kepler, but the Tesla's sheer scale gives it a large throughput advantage.

Memory architecture also differs sharply. The Tesla K20c uses a 320-bit bus with 5 GB of GDDR5, achieving 208.0 GB/s. The GTX 960M uses a 128-bit bus with 4 GB of GDDR5, achieving 80.19 GB/s. The Tesla's memory system is designed for bandwidth-hungry compute kernels, while the GTX 960M's is sized for mobile gaming and lighter workloads.

The Tesla K20c has no display outputs, confirming its role as a compute accelerator. The GTX 960M's display outputs are listed as "Portable Device Dependent," meaning it is intended to drive a screen in a laptop. The Tesla's power delivery requires a 6-pin and an 8-pin connector, while the GTX 960M draws power through its MXM slot with no additional connectors.

Specification Differences

The two cards diverge on nearly every specification field. The Tesla K20c has 5 GB of GDDR5 memory on a 320-bit bus, while the GTX 960M has 4 GB of GDDR5 on a 128-bit bus. Memory bandwidth is 208.0 GB/s versus 80.19 GB/s. The Tesla's memory clock is recorded at 1300 MHz with 5.2 Gbps effective, while the GTX 960M's is 1253 MHz with 5 Gbps effective.

Shading units: 2,496 versus 640. TMUs: 208 versus 40. ROPs: 40 versus 16. Pixel rate: 36.71 GPixel/s versus 18.82 GPixel/s. Texture rate: 146.8 GTexel/s versus 47.04 GTexel/s. FP32: 3.524 TFLOPS versus 1.505 TFLOPS. The Tesla leads every raw throughput metric.

The GTX 960M has recorded base and boost clocks of 1097 MHz and 1176 MHz respectively, while the Tesla K20c has no recorded base or boost clock values. The Tesla has a 225 W TDP and a dual-slot form factor with a 267 mm length. The GTX 960M has a 75 W TDP and is an MXM module with no length recorded.

Power connectors differ completely: the Tesla requires one 6-pin and one 8-pin, while the GTX 960M has none. The suggested PSU for the Tesla is 550 W, while the GTX 960M has no suggested PSU listed. The bus interface is PCIe 2.0 x16 for the Tesla and MXM-B (3.0) for the GTX 960M.

API support is similar but not identical. Both support DirectX 12 (11_0) and OpenGL 4.6. The Tesla lists Vulkan 1.2.175, while the GTX 960M lists Vulkan 1.4. The GTX 960M has a newer Vulkan version.

Release timing also differs. The Tesla K20c launched in November 2012 with a launch MSRP of 3,199 USD. The GTX 960M launched in March 2015 with no launch MSRP recorded. The Tesla's predecessor is Tesla Fermi and its successor is Tesla Maxwell. The GTX 960M's predecessor is GeForce 800M and its successor is GeForce 10 Mobile.

Head-to-Head Benchmarks

The database contains one direct comparison between these two GPUs: Geekbench OpenCL. The Tesla K20c scores 11,479, while the GTX 960M scores 11,045. This gives the Tesla a 3.9% lead.

That 3.9% gap is modest given the specifications. The Tesla has nearly four times the shading units and more than twice the FP32 throughput. The fact that the GTX 960M comes within 3.9% in OpenCL suggests that Maxwell's architecture extracts more real-world performance per shader and per watt than Kepler. The GTX 960M's boost clock of 1176 MHz likely helps close the gap in latency-bound workloads, while the Tesla's higher bandwidth and wider execution width matter more for parallel throughput.

The Tesla's nearest rivals in the database are the AMD Radeon Pro 5500M with an average score of 11,528 (0.4% higher), the AMD Radeon RX 7800 XT at 11,627 (1.3% higher), and the NVIDIA GeForce GTX 1660 at 11,680 (1.7% higher). The NVIDIA GeForce GTX 780M trails at 11,261, which is 1.9% below the Tesla's average of 11,479.

The GTX 960M's average benchmark score is 9,645, which is notably lower than its OpenCL score of 11,045 because the database also includes its Geekbench Vulkan score of 8,245. The Tesla's average benchmark score is 11,479, identical to its OpenCL score since that is its only recorded benchmark.

The GTX 960M's nearest rivals cluster tightly around its average. The NVIDIA Quadro K5000 scores 9,637 (0.1% below), the AMD Radeon Pro WX 2100 scores 9,653 (0.1% above), the NVIDIA Quadro P4000 scores 9,665 (0.2% above), and the NVIDIA Tesla C2070 scores 9,716 (0.7% above). This tight cluster suggests the GTX 960M sits in a crowded performance band where small architectural differences determine rankings.

The head-to-head delta of 3.9% is smaller than the gap in average scores. The Tesla's average score of 11,479 is 19% higher than the GTX 960M's average of 9,645. That larger gap reflects the GTX 960M's weaker Vulkan showing, which pulls down its overall average.

The Verdict

The data points to a clear split between these two GPUs. The Tesla K20c is the stronger compute performer in the one shared benchmark, leading by 3.9% in Geekbench OpenCL. It also holds commanding leads in memory bandwidth (208.0 GB/s versus 80.19 GB/s), shading units (2,496 versus 640), and FP32 throughput (3.524 TFLOPS versus 1.505 TFLOPS). For anyone running OpenCL compute workloads, large data sets, or tasks that need more than 4 GB of memory, the Tesla K20c is the better choice.

The GTX 960M is the more practical option for portable systems. Its 75 W TDP, MXM form factor, and lack of power connectors make it installable in laptops. Its Vulkan 1.4 support is newer than the Tesla's Vulkan 1.2.175, and its recorded base and boost clocks give it predictable performance scaling. The GTX 960M's OpenCL score of 11,045 is only 434 points behind the Tesla, showing that Maxwell's efficiency offsets much of Kepler's raw scale.

The percentile rankings reinforce this. The Tesla sits at the 51st percentile, the GTX 960M at the 46th. Both are mid-pack GPUs in the overall database, not top-tier performers. The Tesla's nearest rivals are all desktop-class cards with average scores just above 11,500, while the GTX 960M's nearest rivals are workstation and older Tesla parts clustered around 9,600 to 9,700.

For a desktop workstation with power headroom, the Tesla K20c wins on compute density and memory capacity. For a laptop or compact chassis, the GTX 960M is the only viable option given its power and thermal profile. The 3.9% OpenCL delta is real but small; the deciding factors are the form factor, power draw, and memory size differences. The Tesla K20c is end-of-life, and the GTX 960M is also end-of-life, so neither is a future-proof purchase. The database shows two aging GPUs with different strengths, and the right pick depends entirely on the physical and electrical constraints of the target system.