NVIDIA GeForce MX350 vs NVIDIA Tesla K20Xm Comparison

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA Tesla K20Xm has an average benchmark score of 12,625, while the NVIDIA GeForce MX350 scores 10,883. The Tesla K20Xm sits at the 52nd percentile among all GPUs, while the MX350 is at the 49th percentile.

Q: How do the two compare in the only head-to-head benchmark available?

A: In the Geekbench OpenCL test, the Tesla K20Xm scores 17,215 versus 8,689 for the MX350, a difference of 98.1% in favor of the Tesla K20Xm.

Q: What is the memory configuration difference?

A: The Tesla K20Xm has 6 GB of GDDR5 memory on a 384-bit bus with 249.6 GB/s of bandwidth. The MX350 has 2 GB of GDDR5 memory on a 64-bit bus with 56.06 GB/s of bandwidth.

Q: Which GPU has higher FP32 compute throughput?

A: The Tesla K20Xm delivers 3.935 TFLOPS of FP32 performance, more than double the 1.879 TFLOPS of the MX350.

Q: What architecture does each GPU use?

A: The Tesla K20Xm uses the Kepler architecture with the GK110 chip, built on a 28 nm process at TSMC. The MX350 uses the Pascal architecture with the GP107S chip, built on a 14 nm process at Samsung.

Q: What is the power draw difference?

A: The Tesla K20Xm has a TDP of 235 W, while the MX350 has a TDP of 20 W. The Tesla requires a 550 W suggested power supply, while the MX350 needs no power connectors.

Architecture Differences

The two GPUs come from different NVIDIA generations and target entirely different segments. The Tesla K20Xm is a compute-oriented accelerator from the Tesla Kepler (Kxx) generation, while the MX350 is a mobile graphics processor from the GeForce MX (3xx) generation. This fundamental split shows up in nearly every architectural decision.

The manufacturing processes are a clear generational divide. The K20Xm uses a 28 nm process at TSMC with 7,080 million transistors on a 561 mm² die, yielding a transistor density of 12.6 million per square millimeter. The MX350 uses a 14 nm process at Samsung with 3,300 million transistors on a 132 mm² die, resulting in a transistor density of 25.0 million per square millimeter. The MX350 is built on a much denser, more modern process, but the K20Xm packs more than twice the raw transistor count.

Compute resources differ massively. The K20Xm has 2,688 shading units, 224 texture mapping units, and 48 ROPs. The MX350 has 640 shading units, 32 TMUs, and 16 ROPs. Neither GPU has dedicated ray tracing cores or tensor cores, as both predate those additions to NVIDIA's lineup. The K20Xm's larger shader count and wider texture pipeline directly explain its higher throughput figures: 164.0 GTexel/s versus 46.98 GTexel/s, and 40.99 GPixel/s versus 23.49 GPixel/s.

Memory architecture also differs significantly. The K20Xm uses a 384-bit memory bus with 249.6 GB/s of bandwidth, while the MX350 is restricted to a 64-bit bus with 56.06 GB/s. The K20Xm also has three times the memory capacity at 6 GB versus 2 GB. Both use GDDR5 memory, but the MX350 runs at a higher effective data rate of 7 Gbps compared to the K20Xm's 5.2 Gbps, though the K20Xm's much wider bus still gives it a massive bandwidth advantage.

The bus interfaces differ as well. The K20Xm uses PCIe 3.0 x16, while the MX350 uses PCIe 3.0 x4, which is typical for a low-power mobile part. The K20Xm is a dual-slot card with no display outputs, indicating its role as a dedicated compute accelerator. The MX350 has portable device dependent display outputs, reflecting its integration into laptops. The K20Xm also supports DirectX 12 (11_0) while the MX350 supports DirectX 12 (12_1), and Vulkan support differs: the K20Xm supports Vulkan 1.2.175, while the MX350 supports Vulkan 1.4. Both support OpenGL 4.6.

Head-to-Head Benchmarks

The database contains one head-to-head benchmark between these two GPUs: Geekbench OpenCL. The results are heavily lopsided. The Tesla K20Xm scores 17,215, while the MX350 scores 8,689. The K20Xm wins by 98.1%, meaning it nearly doubles the MX350's score in this compute-oriented test. This aligns with the raw specifications: the K20Xm has 2,688 shading units versus 640, and 3.935 TFLOPS of FP32 throughput versus 1.879 TFLOPS.

The K20Xm's average benchmark score of 12,625 places it near rivals such as the AMD Radeon RX 7600M XT (12,710, a 0.7% difference), the NVIDIA GeForce GTX 670 (12,773, 1.2% lower), the NVIDIA GeForce GTX 590 (12,830, 1.6% lower), and the AMD Radeon Pro 455 (12,831, 1.6% lower). The MX350's average score of 10,883 places it near the AMD Radeon Pro 450 (10,804, 0.7% higher), the NVIDIA Quadro K2200 (10,761, 1.1% higher), the NVIDIA GeForce GTX 1650 SUPER (11,047, 1.5% lower), and the AMD Radeon RX 550 (11,075, 1.7% lower). The K20Xm's closest rivals are all desktop-class or high-end mobile parts, while the MX350 competes with entry-level and older workstation GPUs.

The single benchmark result confirms what the compute specifications suggest: the K20Xm is a much more powerful compute device. The MX350 does have a higher effective memory data rate (7 Gbps versus 5.2 Gbps) and a denser process node, but these do not translate into better OpenCL performance. The K20Xm's advantages in shader count, memory bandwidth, and FP32 throughput dominate the comparison.

In terms of the win count, the K20Xm takes the single head-to-head test, while the MX350 wins none. The database records one win for the K20Xm and zero for the MX350. The MX350's only benchmark entries are Geekbench OpenCL at 8,689 and Geekbench Vulkan at 13,077, but there is no Vulkan result for the K20Xm to compare against, so those cannot be used for a direct head-to-head comparison.

Specification Differences

The two GPUs differ in nearly every specification category. The K20Xm uses the GK110 chip with Kepler architecture, while the MX350 uses the GP107S chip with Pascal architecture. The process node differs: 28 nm for the K20Xm versus 14 nm for the MX350. The K20Xm has 7,080 million transistors on a 561 mm² die, while the MX350 has 3,300 million transistors on a 132 mm² die. Transistor density is 12.6 million per square millimeter for the K20Xm and 25.0 million per square millimeter for the MX350.

Clock specifications differ as well. The K20Xm has no listed base or boost clock, while the MX350 has a base clock of 1354 MHz and a boost clock of 1468 MHz. Memory clocks also differ: the K20Xm runs at 1300 MHz (5.2 Gbps effective), while the MX350 runs at 1752 MHz (7 Gbps effective).

Memory capacity is 6 GB for the K20Xm versus 2 GB for the MX350. Both use GDDR5, but the bus width is 384 bits for the K20Xm versus 64 bits for the MX350. Memory bandwidth is 249.6 GB/s for the K20Xm versus 56.06 GB/s for the MX350.

Shading units are 2,688 for the K20Xm versus 640 for the MX350. TMUs are 224 versus 32, and ROPs are 48 versus 16. Pixel rate is 40.99 GPixel/s for the K20Xm versus 23.49 GPixel/s for the MX350. Texture rate is 164.0 GTexel/s versus 46.98 GTexel/s. FP32 compute is 3.935 TFLOPS versus 1.879 TFLOPS. The MX350 has a listed FP16 rate of 29.36 GFLOPS (1:64), while the K20Xm has no FP16 listing.

Power specifications are dramatically different. The K20Xm has a TDP of 235 W and a suggested PSU of 550 W, while the MX350 has a TDP of 20 W and no listed power connectors. The K20Xm is a dual-slot card, while the MX350 has no slot width listed. The K20Xm has no display outputs, while the MX350 has portable device dependent outputs. The bus interface is PCIe 3.0 x16 for the K20Xm and PCIe 3.0 x4 for the MX350.

API support differs in DirectX and Vulkan. The K20Xm supports DirectX 12 (11_0) and Vulkan 1.2.175, while the MX350 supports DirectX 12 (12_1) and Vulkan 1.4. Both support OpenGL 4.6. The K20Xm has a launch MSRP of 7,699 USD, while the MX350 has no launch MSRP listed. The K20Xm is 267 mm (10.5 inches) long, while the MX350 has no dimensions listed. The K20Xm was released in 2012 and has Tesla Fermi as its predecessor and Tesla Maxwell as its successor, while the MX350 was released in 2020 with no predecessor or successor listed. Both are end-of-life products.

The Verdict

The data paints a clear picture for compute workloads. The Tesla K20Xm is the far stronger performer in raw compute, with 98.1% higher OpenCL score than the MX350, more than double the FP32 throughput, and nearly 4.5 times the memory bandwidth. Its 6 GB of memory and 384-bit bus make it suited for memory-intensive compute tasks, and its average benchmark score of 12,625 places it above the MX350's 10,883. The K20Xm also has a much higher transistor count and a wider shading unit array, which directly supports its compute advantage.

The MX350 counters with efficiency and portability. Its 20 W TDP versus the K20Xm's 235 W means it fits in thin laptops without power connectors, and its smaller 132 mm² die on a 14 nm process shows a modern manufacturing approach. Its DirectX 12 (12_1) support is newer than the K20Xm's DirectX 12 (11_0), and its Vulkan 1.4 support is more recent than the K20Xm's Vulkan 1.2.175. For mobile users who need basic acceleration and display output, the MX350 is the practical choice.

For anyone prioritizing compute performance, the K20Xm is the clear pick from the recorded data. Its nearest rivals are all within 1.6% of its average score, meaning it slots into a competitive desktop GPU tier, while the MX350's nearest rivals are lower-tier workstation and entry-level parts. The K20Xm's single head-to-head win and higher percentile ranking make it the better compute investment among these two, despite its much higher TDP and lack of display outputs. The MX350 remains the option for portable systems where power draw and physical footprint matter more than raw throughput.