Nvidia Volta Megathread

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
DENVER, Nov. 13, 2017 (GLOBE NEWSWIRE) -- NVIDIA today announced the world’s most advanced data center GPU — the NVIDIA Tesla V100 GPU based on NVIDIA’s VOLTA ARCHITECTURE — is available through every major computer maker and chosen by every major cloud to deliver artificial intelligence and high performance computing.

Dell EMC, Hewlett Packard Enterprise, Huawei, IBM and Lenovo have all announced Volta-based offerings for their customers. Providers such as Alibaba Cloud, Amazon Web Services, Baidu Cloud, Microsoft Azure, Oracle Cloud and Tencent Cloud have also announced Volta-based cloud services.

Building on this breadth of offerings, NVIDIA has introduced new software and tools on the NVIDIA GPU Cloud (NGC) container registry that make it easy for scientists to deploy NVIDIA’s accelerated computing platform for compute-intensive research.

NVIDIA shared the news at the SC17 supercomputing conference, where dozens of computer makers and cloud service providers launched a wide range of Volta-based products and services.

“Volta is the world’s most powerful platform for AI and HPC, and will allow the world’s top minds in scientific research to push the limit on what’s possible in areas like drug discovery, alternative fuel sources and predicting natural disasters,” said Jensen Huang, founder and chief executive officer of NVIDIA. “With Volta now in data centers and clouds around the world, a new wave of innovation is underway that will have an incredible impact across society.”
 
I don't think there is anything that will stop Nvidia any time soon. The only exception would be a new player to enter the market that can beat the performance of Nvidia's top performing GPU's. Don't expect Intel to have anything for ~2 years, and don't have high expectations of that product. That being said before too long other big companies might want to take a stab at these markets.
 


Well, they are actually.

Like I said previously, somewhere, nVidia is just the one making the most noise about it. IBM has been in the AI business and HPC business for a very long time now. I can't remember which was the other big player in that field. NEC, maybe? I remember them having super computers that were challengers as well to general computing disciplines.

Cheers!
 

Yes maybe in the past but as today no one can match the raw performance and efficiency of NVIDIA GPU's that's why instead of throwing a bunch of billion challenging NVIDIA IBM, LENOVO and others have decided to ride on the same train.
 


I want to say it was Hitachi that made a statement a few months ago that they were going to come up with something 5X faster than what current high end GPU's could perform in 2 years, but with a simple google search I couldn't find it. I think any and all forward thinking companies will realize the power of A.I, and its role in the future of humanity. Eventually, it will contain ~all of the information in the world. It will take over creation of hardware, software, automation, and create things beyond human comprehension. All companies should be extremely interested in getting involved with A.I.
 


Yep, one of them was NEC: https://en.wikipedia.org/wiki/NEC_SX-ACE

I do remember that another of the Japanese big ones was also a big player in the field, but I haven't read anything about them in a long time. It may have been Hitachi, like you say.

And nVidia is just making a lot of noise; that is how good nVidia's PR is. They're getting into the car AI business, so they want you to buy cars with "nVidia inside", much like Intel does with their stickers, haha. This does not mean they're not a good contender, again, even with their "short" time in the fray.

Cheers!
 
I just hope they keep using their market position to continue pushing efficiency. Whichever company lags behind in performance is usually forced to push thermal and electrical limits. But the company out in front hasn't always done much with it, aside from charging a premium and delaying new releases. Pascal was a marked and welcome change from that approach. So was Polaris for that matter, despite some early issues.
 
Nvidia first announced the Volta microarchitecture back in March 2013, presenting it as a successor to the Pascal microarchitecture that can be found in the GTX 1000 series today. Previous roadmaps showed that the first Volta GPUs could hit the market in Q3 2017, but Nvidia stated at the Hot Chips conference this year that Volta products are scheduled for an early 2018 release. Nvidia’s Tesla Chief Platform Architect, Rob Ober, was present at the conference and revealed some new details about the Volta GV100 chips.

Using the TSMC 12 nm FinFET design and the SXM2 form factor, the GV100 has a massive die measuring 32 inch2, which is home for 21 billion transistors. Flanking the die are four stacks of HBM2 memory chips cumulating 16 GB. The GPU integrates 80 activated SMs - the equivalent of 5,120 CUDA cores. There are 4 additional non-activated SMs that offset any defects in the manufacturing process. The whole package where the die and memory chips reside draws an average of 300 W at 1 V
 
in the past nvidia shares their roadmap for at least two generation ahead. for example we know the existence of maxwell when nvidia starts rolling out fermi to the market. if nvidia stays true to that right now we should know the name the architecture that will succeed volta. but a lot of things can change in four years time so giving the public about early concept of new architecture improvement end up being useless. i still remember when maxwell was first being unveil it was supposed to succeed kepler in both compute and gaming aspect; meaning maxwell top compute chip should offer better FP64 performance than GK110. also for maxwell generation nvidia was supposed to include some kind of ARM CPU inside their GPU. in the end none of that happen. volta was supposed to be 2016 but nvidia slips in pascal in the middle for a reason that only nvidia truly knows.

right now i'm just interested what nvidia will do with their next generation gaming GPU. will they keep their current pace or will they finally decided to take it easy.
 
Pascal was inserted between Maxwell and Volta because NVidia couldn't make the Tensor units perform as designed at the time. Everything else was ready, so they renamed it and released it as Pascal. In the meantime they had ironed out the Tensor core problems and made other improvements to the design, especially in the process portion.
 
79a.jpg


Following the surprise TITAN X Pascal launch slated for 2nd August, it looks like NVIDIA product development cycle is running on steroids, with reports emerging of the company accelerating its next-generation "Volta" architecture debut to May 2017, along the sidelines of next year's GTC. The architecture was originally scheduled to make its debut in 2018.

Much like "Pascal," the "Volta" architecture could first debut with HPC products, before moving on to the consumer graphics segment. NVIDIA could also retain the 16 nm FinFET+ process at TSMC for Volta. Stacked on-package memory such as HBM2 could be more readily available by 2017, and could hit sizable volumes towards the end of the year, making it ripe for implementation in high-volume consumer products.
https://www.techpowerup.com/224413/nvidia-accelerates-volta-to-may-2017
 


AFAIK nvidia never talk about the existence of tensor core inside their GPU until they officially unveil GV100 back in GTC. though we got some hint there might be some new hardware related to AI when nvidia unveil tegra xavier. if i remember correctly pascal was added to the roadmap because nvidia said they could not make certain memory tech on volta on time so they push volta further. pascal was done with the tech that is possible at the time for 2016 time frame.
 
That card doesn't makes sense to have the tag TITAN. Either NVIDIA have lost it or used up all their brains in creating that powerful card that they ran out thinking power to think of a new name for it and ended up naming it TITAN V even though it being a non gaming card.

With TITAN X PASCAL they moved away from naming it under GTX lineup but still it was for gaming. With TITAN V they completely moved away from gaming.
 
No, the name Titan is plenty recognizable (EDIT: and fits perfectly with this card). If there's one thing you can trust from nVidia, it's their marketing department XD

It might not be as long lived as "GeForce", but it does carry it's own weight. Just like the "ti" moniker.

Cheers!
 
But they always have the Titan cards as the top dog until they release the xx80ti version that is either on par or slightly better.

Given the announcement, it is great at gaming AND computing, so...

Cheers!
 
@King Dranzer this is what a titan should be. Even with the original titan the main feature of the card is unlocked FP64 performance. Just that you can also gaming on it since nvidia compute and gaming card share the same architecture. What does not really make sense is the titan series after that. Titan maxwell have nothing extra in it except having more VRAM than 980ti. No specific compute feature being unlocked when the architecture itself was focused 100% for rendering and gaming workload. Titan pascal is a bit better. It still does not have massive FP64 but it has massive INT 8 performance as compute feature that did not exist in GP104/6/7. but titan pascal end up nothing special to it when nvidia release 1080Ti with it's INT 8 performance intact. But for the compute crowd i think they would want GP100 titan more. And their wish has been answered with titan v. Though titan V still significantly more expensive than previous titan at 3k.
 
TITAN should be made and priced practically for gaming crowd because there is already Quadro lineup dedicated to provide workstation and compute performance. Limiting TITAN to what a Quadro already does is meaningless and not worth it. Why would one buy a TITAN over Quadro if he has to use it for dedicated workstation and not for gaming. At present the TITAN V is senseless as it is not for gaming and people who want dedicated workstation card and are ready to spend high amount would go for Quadro over TITAN. TITAN card should handle both gaming and workstation tasks equally good if not the best. That is where TITAN card should stand to makes most sense of it.
 
Ok, wait. You're talking like if you've seen what this card is capable of in games already.

You might not agree with the 3K price tag or the name of it, but I will give nVidia de benefit of the doubt when they claim this card is bonkers. And it is for "prosumers", not gamers really. Gamers usually go up to xxx80ti, unless they have money to burn or they actually have a usage for the card. Check the Steam survey, most gamers are around the $300 price range, so going by your token, the 1080ti and anything above or near a 1080 is not a gamers card.

And finally, Quadros based on Volta are in the 10K price range, so the Titan is peanuts compared to it, even with the 3K price tag.

Cheers!
 
Unless the gaming performance is crazily pumped up the price to performance is not good to be considered as gaming card. Yes if this card performs lot better than TITAN Xp SLI then definitely it is worth it. But can it perform that well we have to wait a day or two to get the results.

NVIDIA's efforts to crate a monster like this is commendable but taking the TITAN further away from gamer community by claiming it to be made for computing and machine learning is what I dint like.

Probably NVIDIA will launch TITAN Xv which will be have same processing power but with DDR6 Memory and priced around $1500 which will be more suitable for gaming.
 
You're definitely not the target audience of this card. I'm 100% sure nVidia is not marketing the Titan lineup as a "price to performance" ratio king. It's just the Veyron of the video cards. This card is "something" that sits between a Quadro and the top dog of the consumer line up. The price gap between them was (before this card) 8K. Titan XP at ~$1500 and the new Quadros at ~9K+. You don't need to double the performance of the current Titan XP to charge a premium for something better if there's nothing next to it. That is the sad reality of things, like it or not.

Cheers!
 
TITAN should be made and priced practically for gaming crowd because there is already Quadro lineup dedicated to provide workstation and compute performance. Limiting TITAN to what a Quadro already does is meaningless and not worth it. Why would one buy a TITAN over Quadro if he has to use it for dedicated workstation and not for gaming. At present the TITAN V is senseless as it is not for gaming and people who want dedicated workstation card and are ready to spend high amount would go for Quadro over TITAN. TITAN card should handle both gaming and workstation tasks equally good if not the best. That is where TITAN card should stand to makes most sense of it.
Click to expand...

so you did not understand what the real target market for titan. first of all the price alone make the titan is not meant for gamer. look at titan black and 780ti for example. out right identical in gaming performance. the titan black sold for $1000 while 780ti sold for $700. looking at it this way it was never make sense for gamer at all. and then we have tesla that costing 4k to 6k a piece. so why the titan exist? originally nvidia push titans for those that want to adopt GPGPU computing but does not have the money to get real tesla. but once they ready to roll out the real deal nvidia hope they will replace those titans with real tesla. this is what the real target market for titan originally.

so yes while true professional will use quadro or tesla there are others that simply cannot afford it due to the high cost. this is part of nvidia strategy to lower the barrier to GPGPU computing for starter or researchers that want to adopt GPGPU. for those that already use GPGPU for a long time they have quadro and tesla with full support (with full priced packaged) for them.

NVIDIA's efforts to crate a monster like this is commendable but taking the TITAN further away from gamer community by claiming it to be made for computing and machine learning is what I dint like.
Click to expand...

nvidia did not take the titan away from gamer. it was never really meant for gamer since the very beginning.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom