Two days after the announcement of the Geforce Titan X graphics card, Nvidia is back again with GM200 and is set to launch it in their new professional graphics counterpart, the Quadro M6000. The company also refreshed its Iray Visual Computing Appliance (VCA) network-attached render server. The Quadro M6000 is packing a fully enabled GM200 GPU, with all of its 3072 CUDA cores are enabled. With a maximum clockspeed of 1.14GHz the card is capable of pushing 7 TFLOPs of single precision performance. Coupled with the card is GM200?s double-sized ROP clusters, giving M6000 96 ROPs and better than 2x the pixel throughput of the outgoing K6000.

The M6000 supports 10 different boost states, the fastest of which is the 1.14GHz state that gives the card its 7 TFLOPS of performance. GPU Boost allows NVIDIA to raise their shipping clockspeeds for better performance without violating the card?s cooling or power delivery restrictions.

Paired with the GM200 is 12GB of GDDR5 memory clocked at 6.6GHz, which is good for 317GB/sec of memory bandwidth.

On the overall performance front, Quadro M6000 is expected to offer a significant performance boost over K6000. Along with the greater clockspeed and the slight increase in the number of CUDA cores, M6000 brings with it the Maxwell 2 family architecture and its efficiency improvements.

In addition, the M6000 is the first Quadro capable of driving four 4K displays thanks to the updated display controller. Meanwhile Quadro also gains the latest NVENC video encoder, which though unlikely to be used at this early stage, opens the door up to real-time HEVC encoding on Quadro.

M6000's TDP is 250W, up from 225W on K6000. Interestingly despite this increase, M6000 still only requires 1 8-pin PCIe power connector.

NVIDIA is aiming the M6000 at the traditional pro visualization market.

The Quadro M6000 will be available soon in complete systems through the company's OEM partners, and as individual cards via the typical retail channels. Expect to see it launch at $5000+, as has been the case with past Quadro 6000 series cards.

NVIDIA also announced a refresh of its Iray Visual Computing Appliance (VCA), the company's high-end network-attached render server. The VCA specializes in high performance remote rendering jobs, packing in multiple GPUs into a single server box, with further scale-out capabilities to multiple VCA boxes via 10GigE and Infiniband.

The Quadro VCA packs in 8 of NVIDIA?s high-end M6000 Quadro cards, clocked slightly differently. The cards have 12GB per GPU and are fully enabled, giving the entire VCA some 96GB of VRAM and 24,576 CUDA cores.

Driving the Quadro cards will be a pair of 10-core Xeon processors, 256GB of system memory, and 2TB of solid state storage.

The new Quadro VCA runs CentOS 6.6 and will also come with Iray 2015 and Chaos's V-Ray RT pre-installed to make setup easier.

The Quadro VCA will be available through NVIDIA's VCA partners for $50,000.

Quadro VCA specs:

• GPUs: 8 NVIDIA High-End GPUs
• GPU Memory: 12 GB per GPU
• NVIDIA CUDA Cores: 24,576
• CPU: Xeon E5 (2.8 GHz)
• CPU Cores:
  • 20 Physical
  • 40 with Hyper Threading
• System Memory: 256 GB
• Storage: 2TB SSD
• Network:
  • 2 x 1GigE,
  • 2 x 10GigE (SFP+),
  • 1 x InfiniBand
• Installed Software:
  • Linux CentOS 6.6
  • VCA Manager
  • Iray 2014.3.4 or newer
  • V-Ray 3.0 or newer