This page separates the VBR performance story from the bitrate-curve story. The plain 3.100 / 3.101b3 path is slow, while the enhanced Visual Studio 2026 3.101b3 builds recover and exceed 3.99.5 throughput using SIMD path selection and OpenMP multi-thread support.
The enhanced 3.101b3 VS builds are fastest overall. They win every album/setting combination in the current test set.
3.99.5 is the strongest official/plain speed reference. It wins 13 of 15 official/plain comparisons.
3.100 matches 3.99.5 output size but is roughly twice as slow. That points to an implementation or plain-build performance regression, not a different VBR decision.
These charts use LAME 3.99.5 as the baseline. A value of 1.0x means equal speed; larger values mean faster encoding on the test system. Output-size differences near 0% mean the file size was effectively unchanged for that setting.
These charts compare output size and encoding time only. They do not claim audible quality differences.
The Optimized Visual Studio 2026 builds are not merely recompiled copies of the MinGW binaries. They include runtime SIMD path selection for SSE, AVX2 and AVX-512 capable CPUs, and OpenMP multi-thread support for parallel work where the encoder path can use it.
For that reason, the performance result should be read as Optimized 3.101b3 target builds compared with a controlled plain MinGW baseline, not as a simple compiler brand comparison.
Lower is faster. Times are average elapsed seconds per album at each VBR level. All non-Optimized rows are MinGW plain builds; Optimized rows are Visual Studio 2026 builds.
| Version / build | Toolchain | -V0 | -V1 | -V2 | -V3 | -V4 |
|---|---|---|---|---|---|---|
| 3.95.1 | MinGW plain | 68.64 s | 69.98 s | 70.79 s | 85.27 s | 83.02 s |
| 3.96.1 | MinGW plain | 69.39 s | 71.23 s | 71.83 s | 86.14 s | 84.83 s |
| 3.97 | MinGW plain | 63.25 s | 63.86 s | 64.84 s | 75.34 s | 74.84 s |
| 3.98.4 | MinGW plain | 36.46 s | 35.53 s | 34.73 s | 33.42 s | 33.05 s |
| 3.99.5 | MinGW plain | 36.28 s | 34.89 s | 34.07 s | 33.01 s | 32.23 s |
| 3.100 | MinGW plain | 68.86 s | 67.69 s | 67.48 s | 66.88 s | 65.67 s |
| 3.101b3 cVBR | MinGW plain | 64.98 s | 62.20 s | 61.17 s | 59.43 s | 58.58 s |
| 3.101b3 cVBR Optimized | Visual Studio 2026 | 27.06 s | 26.45 s | 26.10 s | 25.85 s | 25.57 s |
| 3.101b3 cVBRb Optimized | Visual Studio 2026 | 26.99 s | 26.39 s | 26.84 s | 25.68 s | 25.57 s |
There are three albums and five VBR settings, giving 15 album/setting combinations. The table now includes zero-win rows, so 3.95.1 and 3.100 are visible as part of the tested set rather than silently omitted.
| Version / build | Toolchain | Smallest output wins | Fastest official/plain wins | Fastest overall wins |
|---|---|---|---|---|
| 3.95.1 | MinGW plain | 0 | 0 | 0 |
| 3.96.1 | MinGW plain | 6 | 0 | 0 |
| 3.97 | MinGW plain | 4 | 0 | 0 |
| 3.98.4 | MinGW plain | 3 | 2 | 0 |
| 3.99.5 | MinGW plain | 1 | 13 | 0 |
| 3.100 | MinGW plain | 0 | 0 | 0 |
| 3.101b3 cVBR | MinGW plain | 1 | 0 | 0 |
| 3.101b3 cVBR Optimized | Visual Studio 2026 | 0 | 0 | 4 |
| 3.101b3 cVBRb Optimized | Visual Studio 2026 | 0 | 0 | 11 |
The Optimized builds preserve the same modern output-size family while reducing elapsed time. Overall, the best Optimized result is approximately 30.4% faster than 3.99.5 across the five VBR settings.
| Setting | 3.99.5 avg time | Best Optimized 3.101b3 time | Speed gain | Output-size difference |
|---|---|---|---|---|
-V0 |
36.28 s | 26.99 s | +34.4% | -0.017% |
-V1 |
34.89 s | 26.39 s | +32.2% | -0.023% |
-V2 |
34.07 s | 26.10 s | +30.5% | -0.029% |
-V3 |
33.01 s | 25.68 s | +28.5% | +0.004% |
-V4 |
32.23 s | 25.57 s | +26.0% | +0.005% |
3.100 produced the same average output size as 3.99.5 in this dataset, but with a large timing penalty.
| Setting | 3.99.5 avg time | 3.100 avg time | 3.100 slower by | Output size |
|---|---|---|---|---|
-V0 |
36.28 s | 68.86 s | +89.8% | matches 3.99.5 |
-V1 |
34.89 s | 67.69 s | +94.0% | matches 3.99.5 |
-V2 |
34.07 s | 67.48 s | +98.1% | matches 3.99.5 |
-V3 |
33.01 s | 66.88 s | +102.6% | matches 3.99.5 |
-V4 |
32.23 s | 65.67 s | +103.7% | matches 3.99.5 |
The VS builds were made with Visual Studio 2026 and are intentionally performance-enhanced. They use runtime SIMD path selection for SSE, AVX2 and AVX-512 capable CPUs, plus OpenMP multi-thread support. They should be described as optimized target builds rather than as a plain MSVC-versus-MinGW comparison.
Based on listening preference and the matching modern VBR profile, 3.99.5 is the subjective reference point for the forward work.
The enhanced 3.101b3 builds retain the 3.99.5-style output profile while delivering substantially better throughput through the added SIMD and OpenMP execution paths.
The unenhanced 3.100 / 3.101b3 path is much slower even when the output-size profile matches 3.99.5.