NVIDIA GeForce GTX 675MX vs NVIDIA Tesla C2070 Comparison

The NVIDIA Tesla C2070 and the NVIDIA GeForce GTX 675MX come from different eras and different markets, yet the recorded database scores place them closer than their backgrounds suggest. Both are end-of-life NVIDIA parts built at TSMC, both carry GDDR5 memory, and both land in the lower-middle of the all-GPU percentile table: the Tesla C2070 sits at the 47th percentile with an average benchmark score of 9716, while the GTX 675MX sits at the 43rd percentile with an average of 8427. The only direct head-to-head measurement available, the Geekbench OpenCL test, favors the mobile Kepler card by a clear margin. What follows is a breakdown of where each card wins, where it loses, and what the underlying silicon says about why.

Head-to-Head Benchmarks

There is exactly one common test in the database for both cards: geekbench_opencl. The GTX 675MX scores 10723 against the Tesla C2070's 9716, a win for the 675MX with the Tesla trailing by 9.4 percent. That is a meaningful gap in a compute-oriented test, and it is notable because the C2070 was designed as a professional compute card while the 675MX was a notebook graphics part.

The rest of the recorded results cannot be compared one-to-one, but they still tell a story. The Tesla C2070's only recorded score is that 9716 OpenCL result, which also serves as its average. The GTX 675MX adds a geekbench_metal score of 6131 to its OpenCL result, and its average across the database comes out to 8427, lower than the Tesla's average but computed over more tests.

Context from each card's nearest rivals sharpens the picture. The C2070's 9716 average sits within a hair of the NVIDIA Tesla M10 at 9724 (a -0.1 percent delta), just above the NVIDIA Quadro P4000 at 9665 (+0.5 percent) and the AMD Radeon Pro WX 2100 at 9653 (+0.7 percent), and just below the NVIDIA GeForce GTX 1070 at 9780 (-0.7 percent). The GTX 675MX's 8427 average is nearly identical to the AMD Radeon 880M at 8436 (-0.1 percent), marginally behind the NVIDIA GeForce MX330 at 8458 (-0.4 percent) and the AMD Radeon HD 8870M at 8462 (-0.4 percent), and ahead of the AMD Radeon R9 M375X at 8325 (+1.2 percent). In short: on OpenCL, the card from the newer, mobile-class Kepler generation beats the older Fermi workstation part; on aggregate score, the two occupy overlapping but distinct neighborhoods in the mid-range.

Where Each One Wins

The GTX 675MX takes the only direct contest. Its geekbench_opencl score of 10723 beats the C2070's 9716 by 9.4 percent, and the head-to-head record reflects this with one win for the 675MX and zero for the Tesla. The 675MX also posts the higher theoretical compute throughput: 1255.7 GFLOPS FP32 against the C2070's 1027.7 GFLOPS, and a higher texture fill rate of 52.32 GTexel/s versus 32.14 GTexel/s. With 960 shading units and 80 TMUs against 448 and 56, the Kepler card simply has more of the resources these throughput numbers derive from.

The C2070 wins on everything memory-related. It carries 6 GB of GDDR5 on a 384-bit bus delivering 143.4 GB/s of bandwidth, against 2 GB on a 256-bit bus delivering 115.2 GB/s on the 675MX. Three times the memory capacity and roughly a quarter more bandwidth means the Tesla handles datasets and workloads that simply do not fit on the mobile card. The C2070 also leads in pixel throughput at 16.07 GPixel/s versus 13.08 GPixel/s, thanks to 48 ROPs against 32.

So the split is clean: throughput-per-test and OpenCL performance go to the 675MX; capacity, bandwidth, and raster output go to the C2070.

The Verdict

The data supports a straightforward recommendation. If the workload fits in 2 GB and the goal is raw compute throughput per benchmark run, the GTX 675MX is the stronger card: it wins the shared OpenCL test by 9.4 percent, delivers higher FP32 and texture rates, and does it all on a 100 W TDP MXM module with no power connectors. If the workload is memory-bound or needs capacity, the Tesla C2070 is the only option of the two: 6 GB against 2 GB is not a gap you can engineer around.

There are practical realities recorded in the database that shape the decision as much as the scores. The C2070 is a dual-slot card, 248 mm long, drawing 238 W through one 6-pin and one 8-pin connector, with a suggested PSU of 550 W and a PCIe 2.0 x16 interface. The 675MX is an MXM-B (3.0) module destined for portable devices with device-dependent display outputs. These two cards do not compete for the same socket, so in practice the comparison is academic: workstation or server slot versus notebook upgrade path. Within each context, both sit near the middle of the historical GPU field, at the 47th and 43rd percentiles respectively.

FAQ

Q: Which card is faster in the shared benchmark?

A: The GeForce GTX 675MX. It scores 10723 in geekbench_opencl versus 9716 for the Tesla C2070, winning by 9.4 percent.

Q: Which card has more memory?

A: The Tesla C2070, with 6 GB of GDDR5 on a 384-bit bus (143.4 GB/s). The GTX 675MX has 2 GB on a 256-bit bus (115.2 GB/s).

Q: How do the two compare against all GPUs in the database?

A: The Tesla C2070 ranks at the 47th percentile; the GTX 675MX ranks at the 43rd percentile. Both are mid-pack parts.

Q: Which card has higher theoretical compute performance?

A: The GTX 675MX, at 1255.7 GFLOPS FP32 versus 1027.7 GFLOPS for the C2070. The C2070 counters with a higher pixel rate, 16.07 GPixel/s against 13.08 GPixel/s.

Q: What interfaces do these cards use?

A: The Tesla C2070 uses PCIe 2.0 x16 in a dual-slot desktop form factor. The GTX 675MX uses MXM-B (3.0) as a notebook module with no power connectors.

Q: Are these cards still in production?

A: No. Both are listed as end-of-life. The C2070 was released in July 2011 and the GTX 675MX in September 2012.

Architecture Differences

These cards represent two consecutive NVIDIA architectures. The Tesla C2070 is built on the GF100 chip, the Fermi architecture, on a 40 nm TSMC process. The GTX 675MX uses the GK104 chip, the Kepler architecture, on a 28 nm TSMC process. The generational shift shows up clearly in the physical data. Despite the Kepler card packing more transistors (3,540 million versus 3,100 million), its die is far smaller at 294 mm² versus 529 mm², giving it more than double the transistor density: 12.0M per mm² against 5.9M per mm².

The resource philosophy differs too. Fermi in the C2070 pairs 448 shading units with 56 TMUs and a heavy 48 ROP back end plus the wide 384-bit memory interface. Kepler in the 675MX goes wide on shaders instead: 960 shading units and 80 TMUs, but a leaner 32 ROPs and a 256-bit bus. That maps directly onto the benchmark story, with the 675MX winning shader- and texture-heavy throughput and the C2070 holding the memory and rasterization advantages.

Lineage is recorded for both. The C2070 succeeds the Tesla line and is followed by Tesla Kepler; the 675MX sits in the GeForce 600M generation, successor to the GeForce 500M and predecessor to the GeForce 700M.

Specification Differences

• Form factor and power: The C2070 is a dual-slot card with 1x 6-pin + 1x 8-pin connectors, a 238 W TDP, and a suggested 550 W PSU. The 675MX is an MXM module with no power connectors and a 100 W TDP.
• Memory: 6 GB GDDR5, 384-bit, 143.4 GB/s (C2070) versus 2 GB GDDR5, 256-bit, 115.2 GB/s (675MX). Memory clocks are 747 MHz (3 Gbps effective) versus 900 MHz (3.6 Gbps effective).
• Process and die: 40 nm, 529 mm², 3,100 million transistors (C2070) versus 28 nm, 294 mm², 3,540 million transistors (675MX).
• Shader and fill resources: 448 shading units, 56 TMUs, 48 ROPs (C2070) versus 960 shading units, 80 TMUs, 32 ROPs (675MX).
• Rates: FP32 of 1027.7 GFLOPS and pixel rate of 16.07 GPixel/s (C2070) versus 1255.7 GFLOPS and 13.08 GPixel/s (675MX); texture rate of 32.14 GTexel/s versus 52.32 GTexel/s.
• Interface and outputs: PCIe 2.0 x16 with 1x DVI (C2070) versus MXM-B (3.0) with portable-device-dependent outputs (675MX).
• API support: Both list DirectX 12 (11_0) and OpenGL 4.6; only the 675MX records Vulkan support, at version 1.2.175.
• Physical dimensions: The C2070 measures 248 mm long; no dimensions are recorded for the MXM module.
• Release timing: July 2011 for the C2070, September 2012 for the 675MX; both are end-of-life with no launch MSRP recorded in the database.