NVIDIA GeForce GTX 670 vs NVIDIA GeForce MX350 Comparison

The NVIDIA GeForce GTX 670 and the NVIDIA GeForce MX350 are products of very different design philosophies, separated by nearly eight years of GPU development. The GTX 670 is a desktop graphics card built on the Kepler architecture, aimed at enthusiasts who wanted serious gaming horsepower in 2012. The MX350 is a laptop-focused part from the Pascal era, prioritizing low power draw over raw throughput. Despite the generational gap, the recorded database scores bring them surprisingly close together, which raises an interesting question: how much does architecture efficiency compensate for a massive difference in hardware resources? The data suggests the answer is "a lot, but not enough to flip the result."

Head-to-Head Benchmarks

The head-to-head record contains two direct comparisons, and the GTX 670 wins both. The margins, however, tell two very different stories.

In the Geekbench OpenCL test, the GTX 670 scored 15341 against the MX350's 8689. That is a 76.6 percent advantage for the Kepler card, a blowout by any measure. OpenCL workloads tend to reward raw parallel compute resources, and this result lines up neatly with the hardware gap: the GTX 670 fields 1344 shading units, 112 texture mapping units, and 32 render output units, while the MX350 offers 640 shading units, 32 TMUs, and 16 ROPs. In compute-heavy scenarios, the older desktop card simply has more than twice the machinery.

The Vulkan test is far more competitive. Here the GTX 670 scored 15553 versus 13077 for the MX350, a gap of 18.9 percent. That the MX350 gets within striking distance at all is remarkable given its hardware deficit, and it points to the gains Pascal made through higher clock speeds and a more modern driver path for Vulkan. The MX350 supports Vulkan 1.4 compared to the GTX 670's Vulkan 1.2.175, and that newer API support may explain part of the resilience.

The aggregate picture reflects this split. The GTX 670's average benchmark score across the database is 12773, placing it in the 52nd percentile of all GPUs. The MX350 averages 10883, sitting at the 49th percentile. A three-point percentile difference between a 170 W desktop card and a 20 W laptop chip is the kind of result that invites a closer look at what each silicon generation actually delivers per unit of hardware.

Where Each One Wins

On pure benchmark outcomes, the GTX 670 sweeps: two head-to-head wins, zero for the MX350. Where does that leave the newer chip?

The GTX 670's territory is anything that scales with parallel throughput. Its OpenCL dominance, texture rate of 109.8 GTexel/s versus the MX350's 46.98 GTexel/s, and pixel fill rate of 27.44 GPixel/s versus 23.49 GPixel/s all favor the desktop card for graphically demanding work. Memory bandwidth is another decisive factor: 192.3 GB/s on a 256-bit bus compared to 56.06 GB/s on a 64-bit bus. Both cards carry the same 2 GB of GDDR5, but the GTX 670 feeds its GPU at more than three times the rate.

The MX350's case rests on efficiency and modernity rather than outright performance. Its 20 W TDP is the standout figure against the GTX 670's 170 W, and it requires no external power connectors at all, while the GTX 670 demands two 6-pin connectors and a suggested 450 W power supply. For laptop contexts where the GTX 670 could never physically exist, the MX350 is the only option of the two. It also wins on feature support, with DirectX 12 at feature level 12_1 versus the GTX 670's 12_1... more precisely, the GTX 670 reports 11_0 feature level, meaning the older card conforms to a lower tier of DirectX 12 hardware support. That difference can matter for software that requires newer feature levels.

In short: the GTX 670 wins when wall power and physical space are unconstrained, and the MX350 wins by default in any scenario where a 38 mm wide, 241 mm long dual-slot card drawing 170 W is a non-starter.

Architecture Differences

The two chips come from different branches of NVIDIA's family tree. The GTX 670 uses the GK104 die, Kepler architecture, part of the GeForce 600 generation, manufactured by TSMC on a 28 nm process. The MX350 uses the GP107S die, Pascal architecture, from the GeForce MX (3xx) generation, manufactured by Samsung on a 14 nm process.

The process node difference shows up clearly in the density figures. The MX350 packs 3,300 million transistors into 132 mm², yielding 25.0M transistors per square millimeter. The GTX 670 carries 3,540 million transistors on a much larger 294 mm² die, at 12.0M per square millimeter. Despite having slightly fewer transistors overall, the MX350 achieves more than double the density, which is how it fits a complete GPU into less than half the silicon area.

Clock behavior is inverted relative to transistor count. The GTX 670 runs at a 915 MHz base clock with a 980 MHz boost, while the MX350 clocks considerably higher at 1354 MHz base and 1468 MHz boost. Memory also runs faster on the newer part, 1752 MHz for 7 Gbps effective against 1502 MHz for 6 Gbps effective, but the narrow 64-bit bus cancels out that frequency advantage in bandwidth terms.

Both GPUs predate dedicated ray tracing and tensor hardware, so neither lists RT cores or tensor cores. The MX350 does expose FP16 capability at 29.36 GFLOPS with a 1:64 ratio, a mode absent from the GTX 670's recorded data. Interface and connectivity differ sharply as well: the GTX 670 connects over PCIe 3.0 x16 with fixed display outputs (2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2), whereas the MX350 uses PCIe 3.0 x4 and leaves display output as "Portable Device Dependent." Both are end-of-life products.

FAQ

Q: Which GPU is faster overall?

A: The GTX 670. It won both head-to-head benchmarks, leading by 76.6 percent in Geekbench OpenCL and 18.9 percent in Geekbench Vulkan, and holds a higher average benchmark score (12773 versus 10883).

Q: How close are they in database ranking?

A: Very close. The GTX 670 sits at the 52nd percentile of all GPUs; the MX350 sits at the 49th percentile, a gap of just three percentile points.

Q: How do their rivals compare?

A: The GTX 670's nearest rivals include the GTX 590 (average score 12830, about 0.4 percent apart), the Radeon Pro 455 (12831), the RX 7600M XT (12710), and the FirePro W5100 (12847). The MX350 trades blows with the Radeon Pro 450 (10804), the Quadro K2200 (10761), the GTX 1650 SUPER (11047), and the RX 550 (11075), all within roughly 2 percent of its average.

Q: Which supports newer graphics APIs?

A: The MX350. It supports Vulkan 1.4 and DirectX 12 with 12_1 feature level, versus Vulkan 1.2.175 and DirectX 12 with 11_0 feature level on the GTX 670. Both support OpenGL 4.6.

Q: Do they have the same amount of memory?

A: Yes, both ship with 2 GB of GDDR5. However, the GTX 670 delivers 192.3 GB/s of bandwidth over a 256-bit bus, while the MX350 manages 56.06 GB/s over 64 bits.

Q: How large is the power difference?

A: The GTX 670 is rated at 170 W TDP and needs two 6-pin power connectors with a suggested 450 W power supply. The MX350 is rated at 20 W TDP and uses no power connectors.

The Verdict

The data supports a straightforward performance verdict with an important caveat. If the goal is maximum benchmark throughput, the GTX 670 is the clear choice: it wins every recorded head-to-head test, leads by 76.6 percent in the most compute-oriented benchmark, and offers triple the memory bandwidth. Its rival set (GTX 590, Radeon Pro 455, FirePro W5100) confirms it operates in a slightly higher performance bracket than the MX350's neighborhood.

The caveat is context. The MX350 exists in a form factor and power envelope the GTX 670 cannot touch, and its Vulkan result, within 18.9 percent of a card with more than double the shading units, demonstrates how much efficiency Pascal extracted from a small die. Users who need newer API support (Vulkan 1.4, DirectX 12_1 feature level) in a low-power package will find the MX350 is the only one of the two that qualifies. Users with a desktop chassis and adequate power delivery who want the faster card should pick the GTX 670. Neither product remains in production, so both are legacy options today.

Specification Differences

• Architecture: Kepler (GTX 670) vs Pascal (MX350)
• Generation: GeForce 600 vs GeForce MX (3xx)
• Chip: GK104 vs GP107S
• Process node: 28 nm (TSMC) vs 14 nm (Samsung)
• Transistors: 3,540 million vs 3,300 million
• Die size: 294 mm² vs 132 mm²
• Transistor density: 12.0M/mm² vs 25.0M/mm²
• Base clock: 915 MHz vs 1354 MHz
• Boost clock: 980 MHz vs 1468 MHz
• Memory clock: 1502 MHz (6 Gbps effective) vs 1752 MHz (7 Gbps effective)
• Bus width: 256 bit vs 64 bit
• Bandwidth: 192.3 GB/s vs 56.06 GB/s
• Shading units: 1344 vs 640
• TMUs: 112 vs 32
• ROPs: 32 vs 16
• Pixel rate: 27.44 GPixel/s vs 23.49 GPixel/s
• Texture rate: 109.8 GTexel/s vs 46.98 GTexel/s
• FP32: 2.634 TFLOPS vs 1.879 TFLOPS
• FP16: not listed vs 29.36 GFLOPS (1:64)
• TDP: 170 W vs 20 W
• Power connectors: 2x 6-pin vs none
• Suggested PSU: 450 W vs not listed
• Bus interface: PCIe 3.0 x16 vs PCIe 3.0 x4
• Display outputs: 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 vs portable-device dependent
• DirectX: 12 (11_0) vs 12 (12_1)
• Vulkan: 1.2.175 vs 1.4
• Slot width: Dual-slot vs not listed
• Dimensions: 241 mm x 111 mm x 38 mm vs not listed
• Release date: May 2012 vs February 2020
• Predecessor/successor: GeForce 500 / GeForce 700 vs none listed
• Launch MSRP: 399 USD (GTX 670) vs not listed for MX350