Context. Stellar activity mitigation is one of the major challenges in the detection of Earth-like exoplanets in radial velocity measurements. Several promising techniques are being developed to investigate the use of spectral time series and differentiate between stellar and planetary perturbations. In this context, developing a software that can efficiently explore the parameter space of stellar activity at the spectral level is of great importance. Aims. The goal of this paper is to present a new version of the Spot Oscillation And Planet (SOAP) 2.0 code that is able to model stellar activity at the spectral level using graphical processing units (GPUs). Methods. We took advantage of the computational power of GPUs to optimise the computationally expensive algorithms behind the original SOAP 2.0 code. For that purpose, we developed GPU kernels that allow for stellar activity to be modeled on any given wavelength range. In addition to the treatment of stellar activity at the spectral level, SOAP-GPU also includes a change of spectral line bisectors from center to limb. It can also take PHOENIX spectra as input to model the quiet photosphere, spots, and faculae, allowing for simulations of stellar activity for a wider space in terms of stellar properties. Results. Benchmark calculations show that for the same accuracy, this new code improves the computational speed by a factor of 60, as compared to a modified version of SOAP 2.0 that generates spectra, when modeling stellar activity on the full visible spectral range with a resolution of R = 115 000. Although the code now includes the variation of spectral line bisector with center-to-limb angle, the effect on the derived RVs is minimal. We also show that it is not possible to fully separate the flux from the convective blueshift effect when modeling spots, due to their lower temperature and the subsequent appearance of molecular absorption in their spectra. Shown to be rather negligible for the Sun, this degeneracy between the flux and convective blueshift effect becomes more important when we move to cooler stars. However, this issue does not impact the estimation of the total effect (flux plus convection), demonstrating that users can trust this output. Conclusions. The publicly available SOAP-GPU code allows us to efficiently model stellar activity at the spectral level, which is essential for testing further stellar activity mitigation techniques at the level of spectral time series that are not affected by other sources of noise. Besides a huge gain in performance, SOAP-GPU also includes more physics and is able to model different stars than the Sun, from F to K dwarfs, thanks to the use of the PHOENIX spectral library. However, we do note that due to the limited understanding of stellar convection and activity on other stars than the Sun, the more we move away from the solar case and the more carefully we ought to consider the output of the code.