Nvidia unveils the Volta-based Quadro GV100

by Mark Tyson on 28 March 2018, 09:31

Tags: NVIDIA (NASDAQ:NVDA)

Quick Link: HEXUS.net/qadr55

Add to My Vault: x

During his GTC 2018 keynote yesterday evening, Nvidia CEO Jensen Huang unveiled the company's latest and greatest professional graphics card. The Volta GPU based Quadro GV100 was built to satisfy the demands imposed by the recently announced real-time ray-tracing technology, in a workstation. In a nutshell this add in card is equipped with; 5120 CUDA cores, 640 Tensor cores, 32GB of HBM2, and 4x DP 1.4.

The graphics computing power outlined above can deliver 14.8 TFLOPS of single-precision compute, 7.4 TFLOPS of double-precision compute power, and 118.5 TFLOPS of deep learning performance. This 4.4-inch (11.2cm) tall and 10.5-inch (26.7cm) long workstation PCI Express 3.0 x 16 graphics card boasts a max power consumption of 250W.

Nvidia's Quadro GV100 also uses a new interconnect technology called NVLink 2 which "extends the programming and memory model out of our GPU to a second one". In practical terms that means you can connect two Quadro GV100 cards and they will act as one GPU. The linked entity provides 10,240 CUDA cores, 64GB of HBM2 memory, and 236 TFLOPS of tensor-core power.

In his keynote speech Nvidia CEO Jensen Huang claims that virtually everyone is adopting RTX technology. Referring to RTX as "the single most important advance in computer graphics in 15 years," Huang shares a slide listing three dozen partners in industries spanning gaming, design, film, architecture - all of whom are onboard with RTX. Quadro GV100 will be able to leverage Nvidia's RTX technology via Microsoft's DirectX Raytracing (DXR) API, the Nvidia Optix API, or an upcoming version of the Vulkan API.

The Nvidia Quadro GV100 is up for sale now, directly from Nvidia for $8,999 a piece with free shipping. There's a buy limit of five pieces per customer.



HEXUS Forums :: 13 Comments

Login with Forum Account

Don't have an account? Register today!
EEC memory?
lumireleon
EEC memory?
Samsungs HBM2 supports ecc as standard so it's possible it's there.

All I can say is, AMD better get on with producing something comparable performance wise because Nvidia prices are getting stupid now, the last quadros topped out around £/$5000, so if the upwards price trend goes on to consumer cards they're going to be stupidly expensiv, in essence I can't see them ‘removing’ the premium that mining added to gpu prices.
LSG501
lumireleon
EEC memory?
Samsungs HBM2 supports ecc as standard so it's possible it's there.

All I can say is, AMD better get on with producing something comparable performance wise because Nvidia prices are getting stupid now, the last quadros topped out around £/$5000, so if the upwards price trend goes on to consumer cards they're going to be stupidly expensiv, in essence I can't see them ‘removing’ the premium that mining added to gpu prices.

Getting stupid? They've been stupid since the Titan first came out!
…so some article stated that Titan V was giving compute errors when a scientific simulation was done twice and the researchers blamed it on lack of EEC memory. So what is Titan V used for?
https://wccftech.com/nvidia-titan-v-error/
lumireleon
…so some article stated that Titan V was giving compute errors when a scientific simulation was done twice and the researchers blamed it on lack of EEC memory. So what is Titan V used for?
https://wccftech.com/nvidia-titan-v-error/

Titan is the consumer chip. Quadro is workstation/scientific etc.