NVIDIA GeForce GTX 660 Ti vs NVIDIA Tesla M4 Comparison

The Verdict

The NVIDIA Tesla M4 is the stronger compute card in the recorded database. Its OpenCL score of 16,932 beats the GTX 660 Ti’s 15,113 by 12%. The M4 also sits at the 60th percentile of all GPUs, while the GTX 660 Ti sits at the 57th. If your workload is compute-focused, particularly OpenCL, the M4 is the clear pick.

The GTX 660 Ti is the better choice for anyone needing display outputs and a wider memory bus. It has 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2, while the M4 has no outputs at all. The GTX 660 Ti also carries a 192-bit bus with 144.2 GB/s bandwidth versus the M4’s 128-bit bus with 88.00 GB/s. For a desktop gaming or general-purpose card, the GTX 660 Ti’s display connectivity and higher memory bandwidth make it more practical.

However, the M4 consumes only 50 W TDP versus the GTX 660 Ti’s 150 W. It is also single-slot, while the GTX 660 Ti is dual-slot and requires two 6-pin power connectors. The M4’s suggested PSU is 250 W, the GTX 660 Ti’s is 450 W. The M4 is a low-power, compute-oriented accelerator. The GTX 660 Ti is a conventional graphics card with higher power draw and legacy display support.

The launch MSRP for the GTX 660 Ti was 299 USD. The M4 has no recorded launch MSRP. The M4 was released in November 2015, the GTX 660 Ti in August 2012. Both are end-of-life products.

Where Each One Wins

NVIDIA Tesla M4 wins in compute efficiency and raw OpenCL performance. The benchmark data shows a 12% lead over the GTX 660 Ti in geekbench_opencl. The M4 achieves this with a 50 W TDP, which is one-third of the GTX 660 Ti’s power draw. The M4’s transistor density is also slightly higher at 12.9M per mm² versus 12.0M per mm², indicating a more modern design on the same 28 nm node.

NVIDIA GeForce GTX 660 Ti wins in memory bandwidth and texture throughput. The GTX 660 Ti’s 144.2 GB/s bandwidth is 64% higher than the M4’s 88.00 GB/s. Its texture rate of 109.8 GTexel/s is 60% higher than the M4’s 68.61 GTexel/s. The GTX 660 Ti also has a higher FP32 throughput at 2.634 TFLOPS versus the M4’s 2.195 TFLOPS. These strengths favor workloads that are bandwidth-bound or texture-heavy, such as traditional rasterization and gaming.

GTX 660 Ti also wins on shading units and TMUs. It has 1,344 shading units and 112 TMUs, while the M4 has 1,024 shading units and 64 TMUs. The M4 counters with 32 ROPs versus the GTX 660 Ti’s 24 ROPs, giving it a higher pixel rate of 34.30 GPixel/s versus 27.44 GPixel/s.

Architecture Differences

The two cards come from different NVIDIA architectures. The Tesla M4 uses the GM206 chip on Maxwell 2.0, while the GTX 660 Ti uses the GK104 chip on Kepler. Both are built on TSMC’s 28 nm process, but the M4 packs 2,940 million transistors on a 228 mm² die, while the GTX 660 Ti has 3,540 million transistors on a 294 mm² die. The M4’s transistor density is 12.9M per mm², slightly denser than the GTX 660 Ti’s 12.0M per mm².

The M4 supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. The GTX 660 Ti supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The M4 has a higher DirectX feature level and a newer Vulkan version, reflecting its later release date.

Memory configurations differ significantly. The M4 has 4 GB of GDDR5 on a 128-bit bus. The GTX 660 Ti has 2 GB of GDDR5 on a 192-bit bus. The M4’s memory clock is 1375 MHz (5.5 Gbps effective), while the GTX 660 Ti’s is 1502 MHz (6 Gbps effective). Despite the GTX 660 Ti’s higher memory clock and wider bus, the M4’s larger capacity may matter for compute workloads that need more VRAM.

The M4’s base clock is 872 MHz with a boost of 1072 MHz. The GTX 660 Ti’s base clock is 915 MHz with a boost of 980 MHz. The GTX 660 Ti has a higher base clock, but the M4 has a higher boost clock. The M4 is a single-slot card with no power connectors, while the GTX 660 Ti is dual-slot with 2x 6-pin connectors.

FAQ

Q: Which card has higher OpenCL performance in the database?

A: The NVIDIA Tesla M4 scores 16,932 in geekbench_opencl, which is 12% higher than the GTX 660 Ti’s 15,113.

Q: Can the Tesla M4 output video to a display?

A: No. The M4 has no display outputs. The GTX 660 Ti has 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

Q: Which card requires less power?

A: The Tesla M4 has a 50 W TDP and a suggested PSU of 250 W. The GTX 660 Ti has a 150 W TDP and a suggested PSU of 450 W.

Q: Does the GTX 660 Ti have more memory bandwidth?

A: Yes. The GTX 660 Ti has 144.2 GB/s on a 192-bit bus. The Tesla M4 has 88.00 GB/s on a 128-bit bus.

Q: Which card has more VRAM?

A: The Tesla M4 has 4 GB of GDDR5. The GTX 660 Ti has 2 GB of GDDR5.

Q: Which card supports a newer Vulkan version?

A: The Tesla M4 supports Vulkan 1.4. The GTX 660 Ti supports Vulkan 1.2.175.

Head-to-Head Benchmarks

The only recorded head-to-head benchmark is geekbench_opencl. The Tesla M4 wins with a score of 16,932 against the GTX 660 Ti’s 15,113. The delta is 12%. This is the sole benchmark in the database, and the M4 wins it outright.

To contextualize the M4’s score, its nearest rivals in the database include the AMD Radeon HD 7970M at 17,019 (-0.5%), the NVIDIA GeForce GTX 690 at 17,037 (-0.6%), the NVIDIA T400 4 GB at 16,792 (+0.8%), and the AMD Radeon RX 7600 XT at 17,083 (-0.9%). The M4 sits within 1% of all these cards, showing it is competitive with a wide range of GPUs despite its low 50 W TDP.

The GTX 660 Ti’s nearest rivals include the AMD Radeon Pro 560X at 15,082 (-0.1%), the AMD Radeon RX 7600 at 15,171 (-0.7%), the NVIDIA GeForce RTX 3050 OEM at 15,199 (-0.9%), and the AMD Radeon 680M at 15,270 (-1.4%). The GTX 660 Ti trails all of them by less than 1.5%, indicating it is at the lower end of a tight performance cluster.

The M4’s single benchmark win is significant because it achieves this with far lower power draw and a smaller die. The GTX 660 Ti’s higher FP32 (2.634 TFLOPS) and texture rate (109.8 GTexel/s) do not translate into a higher OpenCL score. This suggests the M4’s Maxwell architecture is more efficient per FLOP in this workload.

The GTX 660 Ti does have a second benchmark, geekbench_vulkan, at 15,012. However, the M4 has no recorded Vulkan score, so no direct comparison is possible. The GTX 660 Ti’s Vulkan score is nearly identical to its OpenCL score, within 0.7%.

Specification Differences

The two cards differ in nearly every major specification except the 28 nm process node and GDDR5 memory type.

Chip and die: The M4 uses GM206 (Maxwell 2.0) with 2,940 million transistors on 228 mm². The GTX 660 Ti uses GK104 (Kepler) with 3,540 million transistors on 294 mm².

Clocks: The M4 runs at 872 MHz base and 1072 MHz boost. The GTX 660 Ti runs at 915 MHz base and 980 MHz boost. Memory clocks are 1375 MHz (5.5 Gbps) for the M4 and 1502 MHz (6 Gbps) for the GTX 660 Ti.

Memory: The M4 has 4 GB on a 128-bit bus with 88.00 GB/s bandwidth. The GTX 660 Ti has 2 GB on a 192-bit bus with 144.2 GB/s bandwidth.

Compute units: The M4 has 1,024 shading units, 64 TMUs, and 32 ROPs. The GTX 660 Ti has 1,344 shading units, 112 TMUs, and 24 ROPs.

Rates: The M4’s pixel rate is 34.30 GPixel/s, texture rate is 68.61 GTexel/s, and FP32 is 2.195 TFLOPS. The GTX 660 Ti’s pixel rate is 27.44 GPixel/s, texture rate is 109.8 GTexel/s, and FP32 is 2.634 TFLOPS.

Power and physical: The M4 has a 50 W TDP, is single-slot, has no power connectors, and suggests a 250 W PSU. The GTX 660 Ti has a 150 W TDP, is dual-slot, requires 2x 6-pin connectors, and suggests a 450 W PSU. The GTX 660 Ti is 241 mm (9.5 inches) long.

Outputs and API: The M4 has no display outputs. The GTX 660 Ti has 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. The M4 supports DirectX 12 (12_1) and Vulkan 1.4. The GTX 660 Ti supports DirectX 12 (11_0) and Vulkan 1.2.175. Both support OpenGL 4.6.

Release and status: The M4 was released on 2015-11-09, the GTX 660 Ti on 2012-08-15. Both are end-of-life. The M4’s predecessor is Tesla Kepler and successor is Tesla Pascal. The GTX 660 Ti’s predecessor is GeForce 500 and successor is GeForce 700. The GTX 660 Ti has a launch MSRP of 299 USD; the M4 has none recorded.