First Volta GPU "TITAN V" announced

Costs a rough $3000 USD

https://www.nvidia.com/en-us/titan/titan-v/

3,072 bit memory bus(652.8 GB/s) and 5020 cores, ~1.5GHz gpu, 12gig vram

$3000… wonder what guarantee it has? If it's run flat out 24/7 mining etherium or doing high-end GPU compute…. I am completely out of touch with current GPUs, still have a GTX980 which is used solely for gaming and it's pretty OK. I have no plans to upgrade to whatever consumer version they release anytime soon - the only reason I got my 980 was because my previous card died. Of course , I don't game in 4K or stereo so I probably don't need super high-end stuff.
 
Joined
Aug 23, 2007
Messages
2,137
Location
Cape Town, South Africa
Seems to be highly focused on compute, and the Titan line is not really intended as a high end gaming card anymore.
 
Joined
Nov 8, 2014
Messages
12,085
Seems to be highly focused on compute, and the Titan line is not really intended as a high end gaming card anymore.

True that. It's why they intentionally removed the "GTX" from the Titan product line to make it clear that they are not primarily intended for gaming.

Still can not wait to see gaming benchmarks of this puppy tho :) .
 
Joined
Oct 18, 2006
Messages
3,201
It's a card for people who don't know why they have all that money.

In practical terms, it's probably not going to represent value for money in any conceivable rational manner.
 
Can be achieved much cheaper if that's the kind of GPU you're looking for. Usually, GPU is not for "machine learning".
 
For machine learning and other compute tasks?

Yes, it's more of an entry level professional compute card, rather than a high end consumer card.
 
Joined
Nov 8, 2014
Messages
12,085
Can be achieved much cheaper if that's the kind of GPU you're looking for. Usually, GPU is not for "machine learning".

It is. It's actually the main topic of nVidias main Shows and their fastest growing segment. I remember when I watched it last year, hoping that the GTX 1080/1070 gets announced and they were only talking about deep learning, which while disappointing was also quite interesting.

GPUs are extremely specialized which makes them good for mining and other very specific tasks, while CPUs are multi purpose tools.
 
Joined
Jun 2, 2012
Messages
4,691
GPUs are becoming the main processing units of many supercomputers, replacing CPUs for their operations. Maybe this one could be a component in those behemoths.

pibbur
 
It is. It's actually the main topic of nVidias main Shows and their fastest growing segment. I remember when I watched it last year, hoping that the GTX 1080/1070 gets announced and they were only talking about deep learning, which while disappointing was also quite interesting.

GPUs are extremely specialized which makes them good for mining and other very specific tasks, while CPUs are multi purpose tools.

I guess you could be right, though I still imagine similar power could be achieved much cheaper. The price sounds pretty ridiculous - just like their other Titan cards - which have only been marginally faster than much, much cheaper cards.
 
The thing is, although these cards are referred to as GPUs, they have features that are more relevant to compute tasks than gaming performance; though the gaming performance might not be great enough to justify the price, the compute performance might blow the consumer gaming cards out of the water.
 
Joined
Nov 8, 2014
Messages
12,085
One example is the Titan supercomputer (ranged as the 5th fastest in the world) which conatains 18,688 CPUs and 18,688 GPUs. The GPU involved is the Nvidia Tesla K20X GPU. This is a so-called General Purpose GPU (GPGPU). It doesn't contain a display output (later Tesla units do).

As far as I know (correct me if I'm wrong), the GPUs operate by transforming data into "image data".

Titan is used for scientific computing, which involve high-performance and very demanding simulations. You can read about it here: https://en.wikipedia.org/wiki/Titan_(supercomputer).

pibbur who, having millions of image processors, thinks (hopes) that he still outperforms supercomputers when it comes to general image processing (especially segmentation and recognition).
 
One example is the Titan supercomputer (ranged as the 5th fastest in the world) which conatains 18,688 CPUs and 18,688 GPUs. The GPU involved is the Nvidia Tesla K20X GPU. This is a so-called General Purpose GPU (GPGPU). It doesn't contain a display output (later Tesla units do).

As far as I know (correct me if I'm wrong), the GPUs operate by transforming data into "image data".

Titan is used for scientific computing, which involve high-performance and very demanding simulations. You can read about it here: https://en.wikipedia.org/wiki/Titan_(supercomputer).

pibbur who, having millions of image processors, thinks (hopes) that he still outperforms supercomputers when it comes to general image processing (especially segmentation and recognition).

Yeah, I guess I haven't been following recent developments.

Previous Titan cards haven't had anything near this level of performance for these "deep learning" purposes. At least, this is the first time I've heard them refer to that specific purpose in terms of marketing.

Should be interesting to see how this evolves.
 
Joined
Apr 12, 2009
Messages
23,459
Back
Top Bottom