We are building a new system at the LCSE aimed at interactive volume visualization of multiterabyte data sets on the PowerWall. This poster will report our progress toward this goal. The ten panels of our PowerWall each are attached to a Dell PC workstation with an Nvidia Quadro 4400 rendering engine. These PCs are interconnected by an Infiniband 4X network with each other and with a 32 processor Itanium-2-based SMP from Unisys. Our goal is to have these PCs volume render their separate viewing frusta from single images in parallel or to cooperate in the rendering of smaller frusta of images to be displayed on a single panel of the PowerWall. A key feature of the system design is the ability to replicate up to 2 TB of data on the nodes, so that each PC can stream the subset of this data that it needs from locally attached disks at full disk bandwidth without any possibility of contention. Each node has 12 SATA disks of 400 GB capacity each, which are striped to deliver over 400 MB/sec data bandwidth (giving over 4 GB/sec aggregate system bandwidth). The idea is to enable interactive exploration of a 2 TB data set that is replicated at every system node. By thoroughly pipelining cooperative volume rendering on 10 PCs, we hope to achieve at least half of the theoretically possible 4 GVoxel/sec rendering rate. This would give us 2 frames per second performance for exploration of a large set of billion voxel cubes consisting of a time series of such cubes for each of several different variables. Integration of the Unisys machine's large computational capacity into this system opens up the possibility of generating new voxel cubes on the fly for variables selected dynamically by the user.