A specialist GPU-accelerated workstation for simulation or rendering workflows
If ever there was ever a need to match hardware to software, this is it. The WS1850 Nvidia Maximus certified 3D graphics workstation has a very well defined role. It is designed specifically to reduce compute times in Computer Aided Engineering (CAE) software and ray trace renderers using powerful GPUs (Graphics Processing Units).
This is commonly referred to as GPU compute and when a workstation features Nvidia GPUs, it is called Nvidia Maximus.
To give a bit more detail, Nvidia Maximus workstations feature multiple GPUs: one dedicated to graphics and one or more dedicated to compute. In our WS1850 workstation there is a Quadro 4000 GPU for 3D graphics and Tesla C2075 GPU for simulation or rendering.
In terms of workflow, the big play for Nvidia Maximus is that it allows engineers to ‘design and simulate’ or ‘design and render’ at the same time, with little to no slow down.
Then, when deadlines are looming, or perhaps during lunch time or overnight, every available GPU and CPU resource can be thrown at a problem to speed things up.
In the WS1850, there is quite a bit of difference between the Quadro 4000 and Tesla C2075, so do not expect too big a speed up when using both GPUs. To express this in numbers: the Quadro 4000 boasts 243 Gigaflops double precision performance while the Tesla C2075 delivers more than twice that with 515 Gigaflops.
The WS1850 is not all about GPUs. It also features a powerful Intel Xeon E5 2670 CPU whose eight CPU cores also happen to be pretty good at number crunching. Between the simulation / rendering software, the Windows operating system and the Nvidia Maximus drivers, the user has pretty good control over where the workstation’s processing resources are deployed.
Firing up our CAD software we made some tweaks to a model and sent it off to render in Bunkspeed Shot using the Tesla C2075 GPU. As the progressive ray trace renderer refined the image, we were still able to carry on designing in our CAD tool quite happily. The software was responsive with no impact whatsoever on 3D performance.
To take things up a notch, we used the Xeon CPU for mental ray rendering in 3ds Max and the Tesla GPU for ray tracing in Bunkspeed Shot. With some tinkering to CPU resources we probably could have left some free for 3D modelling in our CAD application.
Beyond processors, the WS1850 is amply supported with 32GB PC3-1600MHz Quad Channel DDR3 memory. This is particularly important when handling simulation workflows as jobs tend to start and end on the CPU, with certain elements getting served up to the GPU as and when applicable.
The WS1850 does not hesitate when getting data to and from the CPU. The 120GB Intel 520 Series Solid State Drive (SSD) boasts the excellent read and write rates of 550MB/s and 500MB/s respectively. A high performance 7,200RPM 2TB SATA 3 hard drive is provided for longer term storage.
As mentioned at the outset, the WS1850 Maximus workstation needs to be matched to applications. There has to be a simulation / rendering element to your workflow, and while this is limited to certain software, the list of compatible Nvidia Maximus applications is growing.
For simulation, the main ones include Ansys Mechanical, MSC Nastran, MSC Marc, and DS Simulia Abaqus. For design visualisation, there is Bunkspeed Shot, 3ds Max Design, DS Catia Live rendering and RTT DeltaGen.
GPU compute can be quite a hard technology to get your head round and buying into the technology is a serious investment. A good way of thinking about it is to compare the WS1850 Maximus to a more traditional dual processor workstation.
All design and engineering workflows need a single CPU. The second CPU is really used to better support multi tasking or accelerating multi-threaded processes like rendering or simulation. With Maximus that second processor just happens to be a GPU. In certain applications, particularly in simulation, it has been demonstrated to get the job done much quicker.
As a closing comment, we feel it is important to mention that the Quadro 4000 and Tesla C2075 are first generation Maximus products, based on Nvidia’s ‘Fermi’ architecture. Nvidia has just announced ‘second generation’ Maximus, based on its new ‘Kepler’ architecture, and machines with the new Quadro K5000 and Tesla K20 should appear in early 2013. We would expect there to be quite a leap in performance, if you can wait that long.