NVIDIA Launches The TITAN V Graphics Card At $2,999 USD

[phoronix]

Phoronix: NVIDIA Launches The TITAN V Graphics Card At $2,999 USD

NVIDIA has launched their latest TITAN ultra high-end graphics card, the TITAN V, which they say is the most powerful PC GPU ever created. The TITAN V is based upon their Volta architecture and will set you back $2,999 USD...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-TITAN-V-2999

[turboNOMAD]

In Star Trek canon, Titan V is the name of a rocket out of which the first warp-capable spaceship was built on Earth. What Star Trek technology can you build out of a graphics card, DATA robot maybe?

[tildearrow]

AMD, please do something similar for Navi... I'm getting so scared...

(edit: AMD definitely needs to improve double-precision performance for Navi)

[Brisse]

and offers 110 TeraFLOPS for deep learning

Whatever that means... I dunno...
Anyway, it's specced as 15 TFLOPs FP32 if anyone is wondering.

[eydee]

Titan Vista is here. The waiting for the Titan 7 begins.

[starshipeleven]

In Star Trek canon, Titan V is the name of a rocket out of which the first warp-capable spaceship was built on Earth.

In the canon it was a re-purposed a near-future ICBM missile (missile to launch nukes pretty much anywhere in like 3-5 minutes).
They borrowed the name from real-life Titan ICBM missiles https://en.wikipedia.org/wiki/Titan_(rocket_family) that phased out of ICBM role somewhat soon, and production stopped at Titan IV iteration in 2005 due to being a pretty ancient 1960s design and the use of toxic fuels that made any spill a PITA to deal with.
After mid-80s' Titans were not used as ICBM anymore and the remaining ones were pressed into service to launch satellites, probes and the first manned space missions (Gemini) for the Air force and NASA, so Roddenberry just jumped on the same bandwagon in his story.

What Star Trek technology can you build out of a graphics card, DATA robot maybe?

Out of this card, maybe not much, but at least it might not suck too bad if compared to ST tech.

It can be the a part of the back-end used by any simulations they run when testing their ideas with the ship's computer or holodeck (basically what it does also in real life, computing jobs for science or profit) or employed in holodeck as rendering node (details on how that works are fuzzy, but something will have to calculate how to angle the "emitters" to project stuff around, and for a lifelike holodek it's not exactly trivial), or running the AI code animating hologram NPCs in the simulations (which is not as basic as it would seem, it's still a fuckton of image-processing, language-processing, and whatnot just to act naturally, even if the AI of the NPC is dumb as a brick and can only follow pre-programmed decision-making).
Also useful to process voice commands in their ship's computer. Which is not as easy as it may seem.

If for "graphic card" you mean a generic massively multicore system where each core is somewhat weak, then yeah, most obvious application is computing and neural networks, so AI code at any level. From relatively simple programs or automation systems that need to "see", "listen" or operate complex sensors like camera, or more complex semi-autonomous AI systems like self-driving cars, and of course fully autonomous generalist intelligence systems, like Data.

A GPU-like system is currently the hardware most similar to a human/animal brain, but it is not hardwired to do just one job in each area, so in theory it can be used to do anything a brain does, using nearly 100% of its full processing power (unlike a human brain that can't as it's hardwired, each area does only a specific job and you can't just re-purpose it on the fly if at all).

There are more specialized systems for neural networks, but they are still following the same general theme of massively multicore system with very weak cores. IBM's "neurosynaptic chips" for example.

[theriddick2018]

Whatever that means... I dunno...
Anyway, it's specced as 15 TFLOPs FP32 if anyone is wondering.

Yeah FP16 is a nice idea but not many are using it for gains. There is a few games that have FP16 implementation for a couple shaders here and there, but the net gains are not much because you really need to use allot of FP16 compute to start seeing gains. IMO this card is a marketing gimmick, most people would not understand it, some rich kids are going to think they actually got a 110 TFLOP card for gaming or something.. chuckle.

[pinguinpc]

Very powerfull card i dont expect nvidia launch volta now (and less 5120 shaders model) but in this time is excellent because many have more resources in this part of year

Sadly amd dont have anything similar in short time (maybe when appears navi, 2H 2018?)

Michael

When you have this card for test ?

[theriddick2018]

micheal, my theriddick account is broken, doesn't let me login even with correct password.

Had to spin another account, no idea how to fix my other one, reset password works but login still fails. Wondering how many people have broken accounts like mine?¿

[theriddick]

micheal I managed to fix it. Seems Vivaldi is having some trouble with this site as cleaning ALL private data fixed the problem for now. I was just a bit confused at first because theriddick2018 worked just fine so why wouldn't theriddick login work. Must be corrupted cookies or something, shrug.