Nvidia keeps upping the performance with each new GPU based on the Pascal architecture. Its latest, the Quadro P6000, can deliver 12 teraflops of single-precision performance, a teraflop more than the Titan X unveiled just last week. The P6000 is aimed at the type of powerful workstations used for tasks like editing HD video and applying special effects, as well as the CAD/CAM applications used by engineers. Workstations generally have more powerful components and more memory capacity than regular desktop PCs.
The new GPU was announced Monday at SIGGRAPH, a computer graphics trade show in Anaheim, California. Details about pricing and availability weren’t immediately available. The P6000 has 3,840 cores and 24GB of GDDR5X memory. It supports 432GBps (gigabytes per second) of memory bandwidth. It has four DisplayPort 1.4 slots, supporting resolutions up to 4K at 120Hz and 5K at 60Hz. The GPU has internal pipelines to render faster 3D video. The new Quadro is significantly faster than its predecessor, Nvidia said in a blog post, though it didn’t provide side by side performance figures.
The current top-end Quadro is the M6000, which is based on the Maxwell architecture, and delivers peak single-precision performance of 7 teraflops. The M6000 uses GDDR5 memory and has lower memory bandwidth.
Nvidia also announced the Quadro P5000, which delivers 8.9 teraflops of peak performance. It has 2,560 cores, 16GB of GDDR5X memory and 288GBps of memory bandwidth. Pricing for that part wasn’t immediately available, either.
There’s still significant interest in high-powered workstations, in part for creating VR content for headsets like the Oculus Rift and HTC Vive.
For now, the Quadro P6000 is Nvidia’s fastest GPU, though it’s likely that an even faster chip for supercomputing will follow.
Nvidia’s Quadro chips dominate the workstation market, though rival AMD is expected to announce new competing products later on Monday.
Nvidia says that Pascal is twice as powerful at simultaneous multi-projection – ie outputting two video signals at once – than Maxwell. This is useful when you’re outputting different images for each eye – eg for 3D cinema or VR.
Dynamic load balancing is what the boards use to decide how much of its power to each of the multiple tasks you might ask it to do – as the scene you’re previewing or composite or edit you’re looking at could include 3D elements (including a camera, lights, models and the like) and non-3D elements that the card is also processing (video, 2D effects, etc). The card’s driver works out what needs the processing power most and directs it proportionally.
GDDR5X RAM can pass information in an out twice as fast as the GDDR5 used by Maxwell – avoiding bottlenecks that could slow down performance. The P6000 has 24GB of RAM, up from the M6000’s 12GB. The P5000 has 16GB, up from the M5000’s 8GB.
Both cards support the DisplayPort 1.4 output standard, which means they can output to screens at resolutions up to 7,680×4,320 (aka 33-megapixel or twice the resoltion of 5K screens like that offered by the Apple’s top-of-the-line iMac).
Both Pascal-based Quadro graphics cards will be out in September.