NVIDIA GeForce GTX 580M vs NVIDIA GeForce GTX 670M Comparison

The NVIDIA GeForce GTX 670M and NVIDIA GeForce GTX 580M are both end-of-life mobile graphics solutions built on the same GF114 chip and Fermi 2.0 architecture, but they target different points in the GeForce lineup—the 600M and 500M generations respectively. The benchmark data shows a remarkably close contest, with the GTX 670M scoring 6513 in the single Geekbench OpenCL test against 6389 for the GTX 580M, a delta of just 1.9%. This slim margin masks deeper architectural differences in memory configuration and compute resources that shape where each GPU is preferable.

Where Each One Wins

The GTX 670M takes the only benchmark victory in the head-to-head comparison, winning the Geekbench OpenCL test with a score of 6513 against 6389. This 1.9% advantage is narrow, but it places the GTX 670M at the 38th percentile among all GPUs, while the GTX 580M sits just one point lower at the 37th percentile. The GTX 670M’s win suggests it has a slight edge in compute-oriented workloads that stress raw parallel processing, likely due to its higher shading unit count and more efficient memory arrangement for certain operations.

The GTX 580M, despite losing the single benchmark, counters with a substantial memory advantage: it offers 2 GB of GDDR5 on a 256-bit bus, yielding 96.00 GB/s of bandwidth, compared to the GTX 670M’s 1536 MB on a 192-bit bus at 72.00 GB/s. This 33% bandwidth advantage means the GTX 580M is better positioned for texture-heavy or large-frame-buffer scenarios, such as high-resolution gaming or applications that repeatedly access large datasets. The GTX 580M also has more texture mapping units (64 vs 56) and ROPs (32 vs 24), which translates to higher texture and pixel fill rates—39.68 GTexel/s versus 33.49 GTexel/s for texture rate, and 9.920 GPixel/s versus 8.372 GPixel/s for pixel rate.

In practice, the GTX 670M wins on compute throughput and overall benchmark score, while the GTX 580M wins on memory bandwidth and fill-rate-bound tasks. For users running OpenCL-accelerated applications, the data favors the GTX 670M. For those who prioritize memory capacity or bandwidth for specific workloads, the GTX 580M is the stronger choice despite its lower aggregate score.

Architecture Differences

Both GPUs share the same fundamental architecture: Fermi 2.0, built on TSMC’s 40 nm process, with 1,950 million transistors on a 332 mm² die. The transistor density is identical at 5.9M per mm². The chip is the same GF114, but NVIDIA configured it differently across the two mobile parts.

The GTX 580M is the fuller implementation, with 384 shading units, 64 TMUs, and 32 ROPs. The GTX 670M scales these down to 336 shading units, 56 TMUs, and 24 ROPs. This reduction in compute resources is partially offset by the GTX 670M’s higher FP32 throughput of 803.7 GFLOPS, though that is still lower than the GTX 580M’s 952.3 GFLOPS—a 15.6% advantage for the older part. The GTX 580M also leads in pixel rate (9.920 GPixel/s vs 8.372 GPixel/s) and texture rate (39.68 GTexel/s vs 33.49 GTexel/s).

Memory architecture diverges sharply. The GTX 580M uses a 256-bit bus with 2 GB GDDR5, while the GTX 670M uses a 192-bit bus with 1536 MB GDDR5. Both run memory at 750 MHz, yielding 3 Gbps effective, but the wider bus gives the GTX 580M 96.00 GB/s versus 72.00 GB/s. This is the most significant architectural distinction beyond the core configuration.

Power consumption also differs: the GTX 580M is rated at 100 W TDP, while the GTX 670M consumes 75 W. Both use MXM modules with no power connectors and an MXM-B (3.0) bus interface. Display outputs are portable-device-dependent for both. The GTX 670M belongs to the GeForce 600M generation, succeeding the 500M series, while the GTX 580M is from the 500M generation, succeeding the 400M series.

The Verdict

Based strictly on the data, the GTX 670M is the better choice for users who value raw OpenCL compute performance, as it wins the only benchmark test and holds a higher percentile rank (38 vs 37). Its 1.9% lead is modest, but it is a lead nonetheless, and it achieves this with 25% lower TDP (75 W vs 100 W), making it more suitable for thinner or more thermally constrained laptops.

The GTX 580M is preferable for users whose workloads are memory-bandwidth sensitive. Its 96.00 GB/s bandwidth is 33% higher than the GTX 670M’s 72.00 GB/s, and its 2 GB frame buffer exceeds the 1536 MB of the younger part. This makes the GTX 580M better equipped for high-resolution textures or large compute buffers, even though it loses the aggregate benchmark.

For gaming at typical laptop resolutions, the GTX 670M’s higher benchmark score suggests it will edge out the GTX 580M in most titles. However, the GTX 580M’s superior fill rates (9.920 GPixel/s vs 8.372 GPixel/s) could give it an advantage in games that are heavy on pixel shading or anti-aliasing. The choice ultimately hinges on whether the user prioritizes the higher compute score and lower power draw of the GTX 670M, or the memory bandwidth and capacity of the GTX 580M.

FAQ

Q: Which GPU has a higher benchmark score?

A: The GTX 670M scores 6513 in Geekbench OpenCL, which is 1.9% higher than the GTX 580M’s 6389.

Q: How do their memory configurations compare?

A: The GTX 580M has 2 GB GDDR5 on a 256-bit bus with 96.00 GB/s bandwidth, while the GTX 670M has 1536 MB GDDR5 on a 192-bit bus with 72.00 GB/s.

Q: Which GPU has more shading units?

A: The GTX 580M has 384 shading units, compared to 336 on the GTX 670M.

Q: Do they use the same chip?

A: Yes, both use the GF114 chip on the Fermi 2.0 architecture, built on TSMC’s 40 nm process with 1,950 million transistors.

Q: What is the power draw difference?

A: The GTX 580M is rated at 100 W TDP, while the GTX 670M is rated at 75 W TDP.

Q: Which GPU has a higher pixel fill rate?

A: The GTX 580M has a pixel rate of 9.920 GPixel/s, which is higher than the GTX 670M’s 8.372 GPixel/s.

Head-to-Head Benchmarks

The only benchmark test in the head-to-head comparison is Geekbench OpenCL, and the GTX 670M wins with 6513 points against 6389 for the GTX 580M. The delta is 1.9%, which is a narrow margin but consistent with the percentile rankings: the GTX 670M sits at the 38th percentile while the GTX 580M is at the 37th.

The GTX 670M’s nearest rivals in the database include the GT 555M (6493, 0.3% behind) and the Quadro M5000M (6481, 0.5% behind), while the Intel UHD Graphics P750 scores 6554, which is 0.6% ahead. This shows the GTX 670M is clustered tightly with mid-range mobile parts from both NVIDIA and AMD, including the Radeon Vega 10 Mobile (6476, 0.6% behind).

The GTX 580M’s closest competitor is the GTX 460 SE, which scores exactly 6389 with a 0% delta. The RTX PRO 5000 72 GB Blackwell scores 6407, which is 0.3% higher, while the Radeon Pro WX 4100 (6330) and Radeon R7 M350 (6327) trail by 0.9% and 1% respectively. This placement indicates the GTX 580M, despite being an older generation, still competes with a range of professional and consumer parts.

Breaking down the benchmark result, the GTX 670M’s advantage likely comes from its 336 shading units being more efficient per clock, or from memory access patterns that favor its 192-bit bus in this specific test. The GTX 580M’s 384 shading units and higher FP32 throughput (952.3 GFLOPS vs 803.7 GFLOPS) should theoretically give it an edge in raw compute, but the data shows otherwise. This suggests the OpenCL workload is not purely ALU-bound, or that the GTX 670M’s newer generation includes optimizations that improve real-world performance beyond what raw specifications predict.

The 1.9% delta is within the range of run-to-run variance for many benchmarks, but the GTX 670M’s win is recorded as a clear victory in the data. Users should interpret this as a near-tie, with the GTX 670M holding a slight edge in this particular test.

Specification Differences

The two GPUs differ in several specification fields, despite sharing the same chip and architecture.

Shading Units: The GTX 580M has 384 shading units, while the GTX 670M has 336—a difference of 48 units.

Texture Mapping Units: The GTX 580M has 64 TMUs, compared to 56 on the GTX 670M.

ROP Units: The GTX 580M has 32 ROPs, whereas the GTX 670M has 24.

Memory Size: The GTX 580M has 2 GB GDDR5, while the GTX 670M has 1536 MB GDDR5.

Memory Bus Width: The GTX 580M uses a 256-bit bus; the GTX 670M uses a 192-bit bus.

Memory Bandwidth: The GTX 580M delivers 96.00 GB/s, versus 72.00 GB/s for the GTX 670M—a 33% higher figure.

Pixel Rate: The GTX 580M achieves 9.920 GPixel/s, compared to 8.372 GPixel/s on the GTX 670M.

Texture Rate: The GTX 580M reaches 39.68 GTexel/s, while the GTX 670M hits 33.49 GTexel/s.

FP32 Performance: The GTX 580M outputs 952.3 GFLOPS, while the GTX 670M outputs 803.7 GFLOPS—an 18.5% gap.

TDP: The GTX 580M is rated at 100 W, while the GTX 670M is rated at 75 W.

Generation: The GTX 580M is from the GeForce 500M series, while the GTX 670M is from the GeForce 600M series.

Release Date: The GTX 580M launched on 2011-06-27, while the GTX 670M launched on 2012-03-21.

Predecessor/Successor: The GTX 580M succeeds the GeForce 400M and is succeeded by the GeForce 600M; the GTX 670M succeeds the GeForce 500M and is succeeded by the GeForce 700M.

All other fields—process node (40 nm), foundry (TSMC), transistors (1,950 million), die size (332 mm²), memory type (GDDR5), memory clock (750 MHz / 3 Gbps effective), slot width (MXM Module), power connectors (None), bus interface (MXM-B 3.0), display outputs (Portable Device Dependent), and API support (DirectX 12 (11_0), OpenGL 4.6)—are identical between the two.