NVIDIA GeForce GTX 660 vs NVIDIA GeForce MX350 Comparison

Head-to-Head Benchmarks

The recorded data shows a split decision between these two NVIDIA parts, with each claiming one benchmark win. The GeForce GTX 660 takes the OpenCL test decisively, scoring 11,347 against the MX350's 8,689. That is a 23.4% advantage for the older card, a substantial margin that reflects its broader compute resources. In raw throughput terms, the GTX 660's 1.981 TFLOPS FP32 rating edges out the MX350's 1.879 TFLOPS, and the gap in texture rate is even larger: 82.56 GTexel/s versus 46.98 GTexel/s.

The MX350 fights back in the Vulkan API test, where it posts 13,077 points against the GTX 660's 11,415. That 14.6% lead is notable because Vulkan tends to expose architectural efficiency rather than just raw shader count. The MX350's Pascal architecture implements Vulkan 1.4 support, while the GTX 660's Kepler architecture tops out at Vulkan 1.2.175. The newer driver stack and API compliance give the MX350 a measurable edge in this workload.

Looking at the broader database context, the MX350 sits at the 49th percentile of all GPUs, while the GTX 660 sits at the 45th. Their average benchmark scores tell a similar story: the MX350 averages 10,883, the GTX 660 averages 9,022. However, those aggregate figures mask the per-test reality. The GTX 660's OpenCL performance is closer to its nearest rivals, with the NVIDIA TITAN V CEO Edition just 0.2% behind and the GeForce GTX 560 0.4% behind. The MX350's nearest rivals include the GeForce GTX 1650 SUPER at 1.5% ahead and the AMD Radeon RX 550 at 1.7% ahead, showing it competes in a slightly faster neighborhood.

The Vulkan result is the more telling one for modern workloads. The MX350's 13,077 score places it ahead of the GTX 660 by a comfortable margin, and this aligns with its newer architecture and driver support. The GTX 660's OpenCL win, while large in percentage terms, comes from a benchmark that heavily weights raw compute throughput, an area where the older card's 960 shading units and 80 texture mapping units simply outmuscle the MX350's 640 shading units and 32 TMUs.

Architecture Differences

The two GPUs come from different foundries and process nodes. The MX350 uses Samsung's 14 nm process, packing 3,300 million transistors into a 132 mm² die. The GTX 660 uses TSMC's 28 nm process, with 2,540 million transistors on a much larger 221 mm² die. The density difference is stark: the MX350 achieves 25.0 million transistors per square millimeter, while the GTX 660 manages only 11.5 million. This explains how the MX350 delivers comparable compute in a far smaller package with dramatically lower power draw.

Both cards use GDDR5 memory with 2 GB capacity, but the memory subsystems diverge sharply. The MX350 has a 64-bit bus with 56.06 GB/s bandwidth, while the GTX 660 has a 192-bit bus with 144.2 GB/s bandwidth. The GTX 660's memory bandwidth is roughly 2.6 times higher, which historically benefits texture-heavy and bandwidth-bound workloads. The MX350 compensates with faster effective memory speed: 7 Gbps versus 6 Gbps, but the narrower bus limits overall throughput.

The shader configurations reflect their respective generations. The MX350's Pascal architecture supports DirectX 12 (12_1), while the GTX 660's Kepler architecture supports DirectX 12 (11_0). Both support OpenGL 4.6, but Vulkan support differs: the MX350 lists Vulkan 1.4, the GTX 660 lists Vulkan 1.2.175. The MX350 also records FP16 performance at 29.36 GFLOPS with a 1:64 ratio, while the GTX 660 has no listed FP16 capability, a sign of its older compute design.

Power and physical requirements diverge completely. The MX350 has a 20 W TDP with no power connectors and no suggested PSU rating, reflecting its mobile-oriented design. The GTX 660 has a 140 W TDP, requires a single 6-pin power connector, and suggests a 300 W power supply. The GTX 660 also occupies a dual-slot form factor and measures 241 mm in length, while the MX350's dimensions are listed as portable-device dependent. The bus interface differs as well: the MX350 uses PCIe 3.0 x4, the GTX 660 uses PCIe 3.0 x16, giving the older card more host bandwidth in theory.

FAQ

Q: Which GPU has higher raw compute throughput?

A: The GTX 660 edges out the MX350 in FP32 performance, posting 1.981 TFLOPS versus 1.879 TFLOPS. The GTX 660 also has far higher texture fill rate at 82.56 GTexel/s compared to 46.98 GTexel/s.

Q: Why does the MX350 win in Vulkan despite lower compute numbers?

A: The MX350's Pascal architecture supports Vulkan 1.4, while the GTX 660's Kepler architecture only supports Vulkan 1.2.175. In the recorded Vulkan benchmark, the MX350 scores 13,077 against the GTX 660's 11,415, a 14.6% advantage that reflects architectural efficiency and newer API compliance rather than raw shader count.

Q: How do their memory bandwidth specifications compare?

A: The GTX 660 has substantially more bandwidth at 144.2 GB/s with a 192-bit bus, while the MX350 provides 56.06 GB/s with a 64-bit bus. The MX350's memory runs at 7 Gbps effective, faster than the GTX 660's 6 Gbps, but the wider bus gives the GTX 660 the overall throughput advantage.

Q: What are their power requirements?

A: The MX350 has a 20 W TDP and requires no power connectors. The GTX 660 has a 140 W TDP, requires one 6-pin power connector, and lists a suggested 300 W power supply. The GTX 660 is also a dual-slot card at 241 mm length.

Q: Which card has better driver and API longevity?

A: The MX350 lists support for DirectX 12 (12_1) and Vulkan 1.4, both newer than the GTX 660's DirectX 12 (11_0) and Vulkan 1.2.175. Both support OpenGL 4.6, and both are marked end-of-life in the database.

Q: How do their average benchmark scores compare?

A: The MX350 averages 10,883 across recorded tests, placing it at the 49th percentile of all GPUs. The GTX 660 averages 9,022, placing it at the 45th percentile. The MX350's nearest rival is the AMD Radeon Pro 450 at 0.7% behind, while the GTX 660's nearest rival is the NVIDIA TITAN V CEO Edition at 0.2% behind.

Specification Differences

The two cards differ across nearly every core specification. The MX350 uses a 14 nm Samsung process with 3,300 million transistors on a 132 mm² die, while the GTX 660 uses a 28 nm TSMC process with 2,540 million transistors on a 221 mm² die. Transistor density favors the MX350 at 25.0M per mm² versus 11.5M per mm².

Clock speeds differ modestly: the MX350 runs at 1354 MHz base and 1468 MHz boost, while the GTX 660 runs at 980 MHz base and 1032 MHz boost. Memory clocks also differ, with the MX350 at 1752 MHz (7 Gbps effective) and the GTX 660 at 1502 MHz (6 Gbps effective).

The compute configuration is where the GTX 660 pulls ahead. It has 960 shading units, 80 TMUs, and 24 ROPs, versus the MX350's 640 shading units, 32 TMUs, and 16 ROPs. Pixel rate favors the MX350 slightly at 23.49 GPixel/s versus 20.64 GPixel/s, but texture rate heavily favors the GTX 660 at 82.56 GTexel/s versus 46.98 GTexel/s.

Physical and power specifications show the generational divide. The MX350 has a 20 W TDP, no power connectors, and a PCIe 3.0 x4 interface. The GTX 660 has a 140 W TDP, one 6-pin connector, a suggested 300 W PSU, a dual-slot form factor, and a PCIe 3.0 x16 interface. Display outputs also differ: the MX350 is portable-device dependent, while the GTX 660 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

API support diverges on DirectX and Vulkan. The MX350 lists DirectX 12 (12_1) and Vulkan 1.4. The GTX 660 lists DirectX 12 (11_0) and Vulkan 1.2.175. Both support OpenGL 4.6. The MX350 has FP16 performance listed at 29.36 GFLOPS (1:64), while the GTX 660 has no FP16 listing.

Release dates are separated by over seven years: the MX350 launched on 2020-02-09, the GTX 660 on 2012-09-05. The GTX 660 has a listed launch MSRP of 229 USD. The GTX 660's predecessor is GeForce 500 and its successor is GeForce 700, while the MX350 has no predecessor or successor listed.

The Verdict

The data supports a clear split recommendation. For OpenCL-heavy compute workloads and any task that benefits from high memory bandwidth and texture throughput, the GTX 660 is the stronger choice. Its 23.4% OpenCL lead, 144.2 GB/s bandwidth, and 82.56 GTexel/s texture rate are decisive advantages. The 192-bit bus gives it a fundamental memory throughput edge that the MX350 cannot overcome with its faster clocked but narrower 64-bit interface.

For modern API workloads, particularly Vulkan, the MX350 is the better pick. Its 14.6% Vulkan lead comes with full Vulkan 1.4 support, DirectX 12 (12_1), and a 20 W TDP that makes it suitable for portable or low-power systems. The Pascal architecture's efficiency shows in the percentile rankings: the MX350 sits at the 49th percentile versus the GTX 660's 45th, and its average benchmark score of 10,883 exceeds the GTX 660's 9,022.

The GTX 660's 140 W power draw, dual-slot size, and 241 mm length make it a desktop-only proposition. The MX350's 20 W TDP and portable-device-dependent outputs make it the obvious choice for thin-and-light systems. Neither card is current production; both are marked end-of-life. The GTX 660's 229 USD launch MSRP reflects its 2012 positioning, while the MX350 has no recorded launch price.

Where Each One Wins

NVIDIA GeForce GTX 660 wins in: OpenCL compute workloads, where it leads by 23.4%. Memory-bandwidth-bound scenarios, given its 144.2 GB/s bandwidth versus 56.06 GB/s. Texture-heavy rendering, with 82.56 GTexel/s versus 46.98 GTexel/s. Any workload that scales with shading units (960 versus 640) and TMUs (80 versus 32). Desktop builds with a 300 W PSU and room for a 241 mm dual-slot card.

NVIDIA GeForce MX350 wins in: Vulkan API workloads, where it leads by 14.6%. Modern API compliance, with Vulkan 1.4 and DirectX 12 (12_1). Power-constrained systems, with a 20 W TDP versus 140 W. Compact or portable designs, given its lack of power connectors and portable-device-dependent outputs. Pixel fill rate, with 23.49 GPixel/s versus 20.64 GPixel/s. Overall database ranking, with the 49th percentile and an average score of 10,883 versus 9,022.

For users prioritizing raw compute and bandwidth in a stationary desktop, the GTX 660's data is compelling. For users needing a low-power, modern-API-capable GPU in a portable chassis, the MX350's record is the one to follow. The benchmark split is even at one win each, but the context of each win matters: the GTX 660's OpenCL victory is larger in percentage terms, while the MX350's Vulkan victory reflects better forward compatibility.