14 DAY TRIAL // Although nobody can deny the fact that specially built graphics processing cards, such as Nvidia's Tesla or AMD's FireStream lines, offer the best performance level in HPC applications these solutions are usually too expensive for most companies to consider, so RenderStream developed a HPC system that drops Tesla in favor of more commonplace GTX 580 graphics cards managing to deliver an impressive performance level without braking the bank.
Built on the same graphics core as the Tesla units that Nvidia sells with a hefty price premium, the GTX 580 can deliver similar performance, especially if used in applications that don't require large amounts of GPU memory or double precision calculations.
Furthermore, although the GTX 580 is not built to withstand the rigors of an HPC environment, RenderStream states that they have customers running GTX cards at fairly high load with no failures thus far.
As a result, RenderStream has built the VDACTr8 systems that can utilize up to eight Nvidia GTX 580 cards packing up to 4096 CUDA cores and 12 TFLOPs of compute power in a single workstation or 4U rack mounted system that can prove to be a great alternative for certain types of HPC and GPU rendering applications.
Compared to a similarly configured system that makes use of eight C2050 GPGPU boards, the VDACTr8 almost cuts the price in half, going from $28,295 to $14,225.
For this much money users will get a dual X5650 2.67GHz 6-core machine that comes packed with 12GB of memory, a 80 GB X25-5 SSD boot drive and a 500 GB HDD data drive that can deliver an impressive 12 TFLOPs of compute power.
To further reiterate the power of their new solutions, RenderStream states that eight card GTX 580 system far surpass, in terms of compute power, their discontinued 16-GPU solution that used 8 GTX 295 cards.
As far as availability is concerned, RenderStream states that due to the current high demand of the GTX 580, lead time for these systems could be up to 45 days from time of purchase, so you have some waiting to do before entering the HPC world.