NVIDIA GeForce GTX 690 vs NVIDIA Tesla K40c Comparison

The Verdict

The data presents a close contest between two end-of-life Kepler cards, but the choice is clear based on workload. The NVIDIA Tesla K40c is the pick for compute-heavy tasks, as its Geekbench OpenCL score of 17,468 edges out the GTX 690's 17,399 by a slim 0.4% margin. The Tesla K40c also carries significantly more memory at 12 GB versus 2 GB, making it the only viable option for large datasets that exceed the GTX 690's frame buffer.

The NVIDIA GeForce GTX 690, however, is the only one of the two with display outputs, so it is the practical choice for any system that needs to drive a monitor. Its 3x DVI and 1x mini-DisplayPort 1.2 outputs mean it can function as a daily-use card, while the Tesla K40c has "No outputs" and is strictly a compute accelerator.

If you need a card for general use and light compute, the GTX 690 wins outright because it can actually show you a desktop. If you are building a dedicated compute node where video output is handled by another card, the Tesla K40c's extra memory and slightly higher OpenCL score make it the better investment, despite its higher launch MSRP of 7,699 USD versus 999 USD for the GTX 690. The percentile data reinforces the closeness: the Tesla K40c sits at the 61st percentile of all GPUs, while the GTX 690 sits at the 60th.

Architecture Differences

Both cards are built on NVIDIA's Kepler architecture and use TSMC's 28 nm process node, but they are fundamentally different chips. The Tesla K40c uses the GK180 chip, a large die measuring 561 mm² and packing 7,080 million transistors. The GTX 690 uses the GK104 chip, which is much smaller at 294 mm² and contains 3,540 million transistors.

The transistor density tells a story of design priorities: the Tesla K40c has a density of 12.6M transistors per mm², while the GTX 690 has 12.0M per mm². This indicates the GK180 is slightly more densely packed, likely due to its compute-focused design. The Tesla K40c belongs to the Tesla Kepler (Kxx) generation and is the successor to Tesla Fermi, while the GTX 690 is part of the GeForce 600 series and succeeds the GeForce 500.

The Tesla K40c has substantially more execution resources: 2,880 shading units, 240 texture mapping units, and 48 ROPs. The GTX 690 has 1,536 shading units, 128 TMUs, and 32 ROPs. Both cards support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, so API compatibility is identical. Neither card has ray tracing or tensor cores, as these were not part of the Kepler architecture. The Tesla K40c is the predecessor to Tesla Maxwell, while the GTX 690 is succeeded by GeForce 700.

FAQ

Q: Which card has more memory, and why does that matter?

A: The Tesla K40c has 12 GB of GDDR5 memory on a 384-bit bus, while the GTX 690 has 2 GB on a 256-bit bus. The Tesla's larger frame buffer allows it to hold datasets six times larger, which is critical for compute workloads like machine learning or scientific simulations that cannot fit in 2 GB.

Q: Does the GTX 690 support video output?

A: Yes. The GTX 690 has 3x DVI and 1x mini-DisplayPort 1.2 outputs. The Tesla K40c has no display outputs at all, meaning it cannot be used as a primary graphics card unless paired with a separate GPU for display.

Q: What is the performance difference in the available benchmark?

A: In the Geekbench OpenCL test, the Tesla K40c scores 17,468 versus the GTX 690's 17,399. The Tesla K40c wins by 0.4%. This is a very small margin, indicating near-parity in raw compute throughput for that specific workload.

Q: Which card has a higher boost clock?

A: The GTX 690 has a boost clock of 1,019 MHz and a base clock of 915 MHz. The Tesla K40c has a boost clock of 876 MHz and a base clock of 745 MHz. The GTX 690 runs at higher frequencies, but the Tesla K40c compensates with more shading units.

Q: Are these cards still supported by modern APIs?

A: Yes. Both cards support DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. This means they can run modern applications, though the 11_0 feature level for DirectX 12 may limit some advanced rendering features.

Q: What power connector configuration does each card use?

A: The Tesla K40c uses one 6-pin and one 8-pin power connector, with a suggested PSU of 550 W. The GTX 690 uses two 8-pin connectors and requires a 700 W power supply. The GTX 690 has a TDP of 300 W, while the Tesla K40c has a TDP of 245 W.

Specification Differences

The two cards diverge on nearly every specification except for architecture, process node, foundry, memory type, memory clock, bus interface, and API support. Both use Kepler, are built on 28 nm at TSMC, use GDDR5 memory clocked at 1,502 MHz (6 Gbps effective), and connect via PCIe 3.0 x16.

The Tesla K40c has the larger chip (GK180, 561 mm²) and more transistors (7,080 million). Its memory configuration is 12 GB on a 384-bit bus, yielding 288.4 GB/s of bandwidth. It has 2,880 shading units, 240 TMUs, and 48 ROPs. Its clocks are lower: 745 MHz base and 876 MHz boost. The pixel rate is 52.56 GPixel/s, and the texture rate is 210.2 GTexel/s. FP32 performance is 5.046 TFLOPS. It consumes 245 W, is dual-slot, uses 1x 6-pin + 1x 8-pin power, and requires a 550 W PSU. It has no display outputs and is 267 mm long. It was released on 2013-10-07 with a launch MSRP of 7,699 USD.

The GTX 690 uses the GK104 chip (294 mm²) with 3,540 million transistors. Its memory is 2 GB on a 256-bit bus, providing 192.3 GB/s. It has 1,536 shading units, 128 TMUs, and 32 ROPs. Clocks are higher: 915 MHz base and 1,019 MHz boost. The pixel rate is 32.61 GPixel/s, and the texture rate is 130.4 GTexel/s. FP32 is 3.130 TFLOPS. It draws 300 W, is dual-slot, uses 2x 8-pin connectors, and needs a 700 W PSU. It has 3x DVI and 1x mini-DisplayPort 1.2 outputs and is 279 mm long, 111 mm high, and 38 mm wide. It was released on 2012-05-02 with a launch MSRP of 999 USD.

Head-to-Head Benchmarks

The only head-to-head benchmark available is Geekbench OpenCL. The Tesla K40c scores 17,468, and the GTX 690 scores 17,399. The Tesla K40c wins by a narrow 0.4% margin. This is a remarkably close result given the architectural differences.

The Tesla K40c achieves this with 88% more shading units (2,880 versus 1,536) and 88% more TMUs (240 versus 128). Its FP32 throughput is 5.046 TFLOPS versus 3.130 TFLOPS for the GTX 690, a 61% advantage. However, the GTX 690's higher clocks (1,019 MHz boost versus 876 MHz boost) and faster memory clock offset some of that advantage.

The GTX 690 also has a Vulkan score of 16,675, but no corresponding Vulkan result exists for the Tesla K40c, so a direct comparison cannot be made. In the OpenCL test, the deltaPct of 0.4% is within the noise of typical benchmark variance, suggesting that for this specific workload, the two cards are effectively tied.

The average benchmark scores tell a similar story. The Tesla K40c has an average score of 17,468, while the GTX 690 averages 17,037 across its two benchmarks. The GTX 690's average is dragged down by its Vulkan score of 16,675, which is 4.2% lower than its OpenCL score. The Tesla K40c's nearest rival is the AMD Radeon Pro 460 at 17,509 (-0.2%), while the GTX 690's nearest rival is the AMD Radeon RX 7600 XT at 17,083 (-0.3%).

Where Each One Wins

The Tesla K40c wins in raw compute density. Its 12 GB memory capacity is the dominant advantage, allowing it to work with datasets that would completely overflow the GTX 690's 2 GB buffer. For tasks like neural network training, scientific computing, or large-scale data processing, the memory capacity alone makes the Tesla K40c the only choice. It also wins on bandwidth with 288.4 GB/s versus 192.3 GB/s, a 50% advantage that matters for memory-bound kernels. The Tesla K40c has a higher FP32 rating (5.046 TFLOPS) and more execution resources, making it superior for parallel workloads that can utilize all 2,880 shading units.

The GTX 690 wins in practicality and frequency. It has display outputs, so it can serve as a functional graphics card in a desktop system. Its higher base clock (915 MHz versus 745 MHz) and boost clock (1,019 MHz versus 876 MHz) give it an advantage in latency-sensitive or lightly-threaded tasks that cannot fully utilize the Tesla's extra compute units. The GTX 690 also wins on power efficiency in a sense: despite a higher TDP of 300 W versus 245 W, it still delivers competitive OpenCL performance while using a much smaller chip (294 mm² versus 561 mm²).

For gaming, the GTX 690 is the obvious pick because it has video outputs. The Tesla K40c cannot output a signal, so it is useless for any interactive workload. For compute, the Tesla K40c's memory capacity is a decisive advantage that no clock speed can overcome. The benchmark data shows they are nearly equal in raw OpenCL throughput, but the Tesla K40c's 12 GB frame buffer makes it the only card that can handle modern large-model workloads without spilling to system memory.