NVIDIAs Quadro P6000 features a whopping 3840 shaders, 24 Gigabyte of VRAM but yet no High Bandwidth Memory. The company's new flagship for the professional market is based on the GP102 chip.

So far the Quadro P6000 is the most powerful NVIDIA Pascal based graphics card announced. The Titan X, which was announced last week, features "only" 3584 shader units and 12 Gigabyte of GDDR5X memory and therefore the P6000 is the first Pascal based card to host 24 Gigabyte of VRAM. In the professional sector this lots of VRAM can drastically improve performance, which is why the AMD FirePro S9170 for instance even features 32 Gigabyte.

Unfortunately there is no detail information available on these cards yet. Clock speeds for example have not yet been disclosed and therefore customers will have to wait until the cards are actually available to find out more about the performance. Nevertheless one can guess that frequencies might be similar to the Titan X since, the TDP is limited to 250 Watt. When it comes to memory throughput we're looking at 480 Gigabyte per second which tranlsates into GDDR5X memory clocking at 10 GHz effective clock speeds.

Next to the P6000 NVIDIA also announced the Quadro P5000, which is based on the GP104 and features 16 Gigabyte of GDDR5X memory. Apart from that there are 2560 shader units at work and memory bandwidth is 320 Gigabyte per second. Unfortunately there is no information on the TDP.

Somewhat special with the GP102 chip from NVIDIA is the fact, that compared to the GP100 these cards don't support fast FP16 and FP64 calculations. In the case of double precision the GP102 is 32 times slower than the GP100, which is basically the same as with the GP104 found in todays gaming graphics cards. On the other hand - compared to the GP104 - the GP102 is capable of accelerating INT8 calculations, delivering four times the performance level seen at FP32.






Source: NVIDIA