Ultra-high field MRI presents great opportunities to acquire images at unprecedented spatial resolutions and contrast. However, the long scan time associated with high resolution, as well as the considerable problem of RF inhomogeneity that leads to unwanted variations in image contrast and SNR [1] are among some of the well-known challenges. Slice specific spokes pulses have previously been shown to remedy the B1+ inhomogeneity for 2D GRE at 9.4T [2]. In this study, we address the acquisition time challenge and present a B1+ homogenised high resolution simultaneous multi-slice (SMS) 2D GRE imaging at 9.4T. Slice and slice-group specific multi-band spokes excitations were designed to optimize the B1+ for each slice location.All experiments were performed on a 9.4T whole body MR scanner (Siemens Medical Solutions, Erlangen, Germany) equipped with head gradients and 16-channel pTX system, using a custom 16TX/31RX array coil [3]. B0 and B1+ maps for the pulse design were acquired with a dual-echo GRE and a T2* compensated transmit phase-encoded DREAM [4,5], respectively. Slice specific 3-spoke pulses were designed in Matlab according to the spatial domain method [6] with magnitude least square optimisation [7,8] including a local SAR regularisation [9] using virtual observation points compressed [10] SAR matrices. The three spokes were placed on the kx-ky plane at (-2.79, -2.79)m^-1, (2.79, 2.79)m^-1 and (0, 0)m^-1. In contrast to a previous study [11], the actual SMS pulses used here were generated on-the-fly by the pulse sequence on the scanner, for each SMS slice-group, using the spokes parameters optimised offline for each individual slice. Pulse peak amplitude over channels was minimized for each SMS-3 pulse using the fast inter-slice phase relaxation approach shown in ref [12]. The flip angles of the SMS spokes pulses were mapped with PreSat-TFL [13] to check against the B1+ predictions. In vivo single band, SMS-2 and SMS-3 T2*-weighted images were acquired using a CAIPIRINHA [14] enabled SMS GRE sequence (TR 400ms, TE 14ms, 17 degrees flip angle, 1 mm slice thickness, 12 slices, in plane resolution 0.28mmx0.28mm, bandwidth 70Hz/Px, slice CAIPIRINHA shift by gradient blips: FOV/2 for SMS-2 and FOV/3 for SMS-3, no in-plane GRAPPA acceleration). The SMS-GRE images were reconstructed in Matlab using slice-GRAPPA [15] with a 7x6 kernel, using a low resolution (matrix 64x64) single-band spokes acquisition as reference. The study was performed according to local IRB rules and after obtaining informed consent. Fig. 1 (a) and (b) show the RF amplitudes and phases, respectively, for each of the transmit channel of an example 3-spokes pulse of one SMS-3 slice group. The monopolar (fly-back) gradient waveforms of the pulse is shown for all three axes in (c) and the corresponding k-space trajectory is shown in (d). Fig. 2 shows flip angle maps of the measured CP mode (a), the 3-spokes MLS prediction (b), as well as the flip angle maps of the actual single-band (c), SMS-2 (d) and SMS-3 (e) excitations as verified with PreSat-TFL. Excellent agreement is observed between predicted and measured B1+ patterns. Fig. 3 shows the in-vivo high resolution SMS-GRE images acquired with 3-spoke single band (a), SMS-2 (b.d) and SMS-3 excitations (c,e). We here improved on the previously shown spokes pulses to further increase the image quality of high resolution GRE at 9.4T [2]. The severe B1+ homogeneity in the CP mode (fig. 2a) necessitates the use of optimised excitations to remove the unwanted variations in SNR and contrast in the imaging plane (fig. 3). Unlike most previous work that uses a single shim or spokes pulse for the entire volume, we employ a slice-by-slice spokes design to ensure that the same high degree of homogeneity is obtained in all slices across the whole brain. Simultaneously we were able to address the issue of long acquisition times by use of SMS acquisition. Importantly, using SMS to accelerate these scans incurs minimal SNR penalty compared to in-plane undersampled single-band scans at the same acceleration factor, since SNR losses are only due to g-factor but not reduced sampling [15]. SMS-GRE is therefore more time efficient in terms of SNR than single-band GRE as long as approximately $$$\text{g-factor}<\sqrt{\text{SMS-factor}}$$$. The g-factor was here minimised by use of CAIPIRINHA sampling [14]. The fast SMS pulse implementation in the sequence, based on spokes shim parameters determined offline, can be easily transferred to other 2D sequences, such as SMS-EPI [16]. In summary, SNR/time efficient B1+ homogenized high resolution GRE imaging at 9.4T was achieved by using SMS-pTX spokes excitations, in conjunction with a fast and flexible scanner implementation for the SMS pulse generation.