Heh, nope, this card could give me a blowjob every night. I still wouldn't buy it. I"m sucker for fast cards and I am lucky enough to be in a situation where I can afford to buy whatever piece of tech that I want, but 3k for a GPU? naah...
All of you bitching about the price, grow up and learn to read. This isn't a gaming card, per se. It's a PROSUMER card, meaning it's meant for professional and enterprise use. The price tag is par for the course for what it does. AMD did the same thing for the VEGA prosumer card. Come on, now. Most of you are supposed to be intelligent people, being on this forum at any rate. Use the brains and good sense you're supposed to have...
Its main benefits are DP and AI and "NVIDIA TITAN users now have free access to GPU-optimized deep learning software on NVIDIA GPU Cloud." It's aimed at business use for sure and no doubt some wealthy prosumers will buy it for the DP performance but it's priced to keep those at bay.
Actually everyone has free access to the Nvidia GPU Cloud. And while they don't support any architecture except Pascal and Volta, they don't prevent you from running the environment with their other architectures.
So AMD charges you more and gives you a "Pro" driver and that great but nVidia charges you more, give you actual prosumer hardware (Tensor cores) but give you "regular" Geforce drivers and that's not ok.
Exactly. The price tags like that of course are derivatives from monopolistic prices of Intel. The 4B transistor mobile processor costs $20 while 3B Intel's one may cost $2000. The trick is that people think that duopolies kill monopolies while we see that even triopolies or quadropolies often are the same monopolies effing brains of people how costly the factories are
Mrspadge, you must be one extreme capitalist who's supporting price gouging, price fixing, monopoly and consumer fraud. So many of the likes of you here. No wonder corporations run our government and jobs being shipped oversea.
Unfortunately... A5 is right in this instance. This isn't gouging and is actually a very reasonable price for the PURPOSE of this card. This isn't a gaming-centric card, despite it being absurdly capable in that regard.
nVidia is doing what any company in their position would do: maximize profits and focus on markets specific to the strengths of their products.
Indeed, kudos for them for doing so. What a strange world where people think making a profit is evil. Must be the new generation of proto-communists I suppose, the rabid desire for free stuff.
The free market has a wonderful metric for deciding what succeeds and what doesn't, it's called price. If it's too high, it won't sell. If it's too low, demand will spike and they'll fail to meet the demand, so raising the price slows the demand. This is money 101. Those who don't like this basic notion of trade just want the efforts of others without paying for it. No wonder we have a welfare state that's out of control.
What's so funny to me is that even $10K is absolutely *nothing* in some pro markets. Some of the systems I used to work with were in the $500K to $1.5M range, and the companies that used them were able to do things that meant their investment paid for itself in just the first day of use. Hobbyists and gamers really have little idea of the true nature of industrial hardware; nothing wrong with that, they don't need to know, but the context of articles like this ought to be enough to convey that there's a deeper big money world beyond the narrow slice of life called gaming.
It's not "extreme capitalist" to believe in some basic freedom: they are free to offer products and you are free not to buy them, if you don't want to. Do you think that's wrong? Do you also imprison every Porsche or Lamborghini merchant as you drive by, because you think they should offer you their cars for less?
Incorrect analogy, because cars like Porshe and Lamborghinis do not have any real function other than "going fast" and "look good". Aka its a luxury unnecessary that can easily be replaced with multiple brands and for different costs according to your needs.
Er, by that definition, pretty much any use of the computer other than directly supporting your societal functions (i.e. if it's for work, in which case you typically aren't paying for it), is unnecessary. Even if you do pay for your own hardware, faster hardware can quickly offset higher prices, and for any work related tasks, this looks to be at a decent price.
These communist trolls never see the irony of commenting on tech message boards. All they have to do is produce a better product and offer it for the people instead of complaining that nobody else does it.
Jconsumer, that would mean they would have to present their talents and skills to the free market, but of course they don't have any. The foundation of communism: get free stuff, do nothing in return, cause chaos. Well, that and of course the eventual murder of a hell of a lot of people.
It's always the luxury items that bring these basement dweller communists out to complain about "rampant capitalism." In their never-never land that never can exist, nor has ever existed, Willy Wonka offers them free prosumer video cards, evian bottle waters and Masserattis, courtesy of the Red Republic.
Porsche and Lamborghini are luxury items. People wants them but they don't need them. If you need a car, there will be many others that fit your need. Honda or Toyota will do just fine, unlike some computer components you need to speed up your productivity.
You want to base this on "computer components you need to speed up your productivity"? So what exactly are you losing by this new offer? You are still free to buy 4 AMD MI25 instead for inference, or the Tesla V100. For FP64 you can still buy 3 Xeon Phi of Hawaii Fire Pro. If you think that's a better deal - fine, vote with your wallet! Otherwise consider again who's making a bad offer.
And if you want to compare this card based on FP32 performance - well, it can do that, but it's obviously not a suitable tool for that and I'm sure nVidia is totally fine if you go with a GTX1080Ti instead.
This has four times the DL performance of AMD's MI25 and eight times the DP performance, oh yeah, unlike that paper part- you can buy this one the day it was announced. This is a comical level slaughter against team red on a *VALUE* basis alone.
These comment threads are rather comical for people who follow the wider industry. AMD is not competition in this segment, nobody really is. Intel, Google and Facebook are spending billions of dollars trying to go toe to toe with nVidia, but all of those years and dollars sunk into CUDA- which a ton of team red fanboys blasted as being stupid, has resulted in utter domination.
The vocal miser crowd (aka mostly red fanboys) doesn't seem to get the simple logic that the smart companies like NV and Apple knows they can't make money off them and thus can care less about their inane opinions.
In a way what NVIDIA is doing reminds me of the headier days of SGI. There was demand for a gfx product that could make possible levels of performance which would not be available via COTS products for at least 5 years, so SGI built what companies wanted. They were expensive as heck (the Onyx in my garage was original $1M), but the performance was lightyears ahead of its time, ditto the features. I saw an Onyx in 1993 doing 12bits/channel and fully subsampled AA, mipmapped textured 3D at HD at a time when proper 3D simply didn't exist on PCs, and wouldn't for several more years (something SGI itself helped get going). Point being, the commercial demand was there, so they made stuff to satisfy that demand, and they made freakin' fortune in the process (some reseller sales reps became millionaires). SGI died due to numerous mistakes, but it's not as if the core demand for something advanced has ever gone away, especially these days with such heavy burdens placed on intelligence gathering, image analysis, data management, etc. Even today it's hard to put together a PC that can do what the Onyx2 Group Station for Defense Imaging could do almost 20 years ago.
If NVIDIA reckons there's a demand, I'm sure they'll make something that costs 10x as much, and still sell plenty of them, though perhaps these days such products are done on the quiet, unlike SGI which made its tech available openly (more or less; DoD and others received customised versions).
At $3k this might be worth it, if it's a true pro-sumer card like the original Titan. Resale value could be very good. I knew people that sold their original Titans for about $100 less than the original price because it was so good as a workstation card. $100 to rent a bleeding edge card only to replace it with the, equal or better, GTX version will be worth it to some. If it gets even within 20% of the V100 for the same tasks, it'll FLY off the shelves.
Pretty much everything about this card screams compute, not gaming. 1/2 FP64, and 110TF of DL make this a pretty damn nice compute card. Plus, it's based on GV100. As a discounted V100, its price is pretty good.
Actually, if the claimed FP64 rates are true, I would personally consider it a decent buy, something nvidia hasn't had for a loooong time.
It is basically a tesla V100 with one defective and disabled memory channel. Asking 10k for the full version of the chip - that's truly criminal. But getting almost the same performance and 3/4 of the memory at less than 1/3 of the price is pretty decent.
Now if nvidia were just to brush up their OpenCL support.
Although it is highly likely that the FP64 rates are limited in software. It would be very odd if they offered that much FP64 performance at that price point. Their DPFLOP price us very consistent across their product range. It is very close for say something gaming grade like the 1080 ti and the quadro gp100, the 1080ti being that "affordable" only because its FP64 rate is capped at 1/32 of the SP rate.
The pro equivalent of the 1080ti is the P6000. Those are based on different silicon than the GP100 which is the one that has 1/2-speed DP compute. The P6000 also offers only 1/32 FP64.
No price is "criminal", that's just stupid. The market decides what makes for a viable price. If it's too high, the product won't sell. I know a company that bought a machine for $1.5M; for their particular task, the system paid for itself in less than an hour. Kinda weird saying some product is too expensive when the price in question doesn't even register on the radar of what some types of tech cost in other fields.
This is what happens when AMD is total crap and Nvidia just gets to run wild. They know they got at least a year before they have to sell Volta at reasonable prices. So why not give some whales a $3000 version for now!
If you do professional work (or your time is very valuable), then paying the premium can be very much worth it.
I paid $650 for a Vega 64 then another $100 for the EK-FC waterblock - so $750 all in for a Vega 64.
In the time I've owned it, not mining at all, I estimate it has saved me about $900 in time over the 980ti it replaced. I've also been able to tweak it to only use about 160W without losing much performance (maybe 5% in my specific compute task - which I have now completed, so no more savings to be had for me).
If I was monetizing my work, this card could make me some good money.
If you are doing professional computational work you CANNOT consider AMD unfortunately. Modern compilers just don't support them... This card is a STEAL for people in engineering simulations, AI, and finance.
For gaming maybe. But for AI, simulation, or other professional work you have to compare it to a quadro or tesla. Looks at it that way I don't think it's a terrible deal. It could be cheaper and definitely will be, but with AMD limping behind there's nothing we can do.
If it pays for itself, it's a great price. Who are you to judge? It's all about TCO and ROI; the upfront raw cost is rarely a deciding factor. Otherwise, SGI would never have done what it was able to do.
Fermi was relativity small chip compared to these monsters(529mm²). Consumer/prosumer side biggest was Maxwell gm200 @ 601mm². Pascal GP100 was 610mm², but that was never entered even prosumer side Titan series.
Intel is the "hope" ?!?! What planet are you coming from ?! :)
AMD is "total crap" !??! Like AMD Radeon Fury X beating GF 980 TI in all applications possible in FHD and UHD 4K ?!?
Trust me, VEGA is getting better with every driver release and in Spring it will be faster than 1080 TI with people wondering why they've bought nVIDIA and then calming it is "because of the environment and the planet" ;)
Don't you think all those GF 980 TI buyers would like an AMD Fury X now ?! The same will happen with VEGA.
Yeah it's a bit more power hungry, but it is capable and competitive while the "Intel hope" is truly "total crap" .
Either people's brains are getting less capable of comprehending the world around them or the marketing departments are using a new king of detergent for the washing .... :)
Vega 64 beats the 1080 Ti in one game as far as I can see (Forza 7 - and even then nobody can understand why), and loses handily in a good number of others. AMD shouldn't be taking a year to get the best performance out of its cards. Also, we all know how power hungry Vega 64 LC is, before you go anywhere near overclocking, so "a bit more power hungry" is inaccurate.
Intel do now have the expertise to develop decent GPUs, but that sort of thing takes time.
From my own personal viewpoint, those 980 Ti users are really not going to be crying into their cereal over not getting a Fury X (especially when you hit that 4GB limit), and those 1080 Ti users are needing surgery to stitch up their sides over the inference that Vega is going to blow them all away in a few months.
" AMD shouldn't be taking a year to get the best performance out of its cards."
THIS ^^^^
That's all you really had to say. LOL
AMD needs to get a handle on their damn drivers. There's no reason to leave THAT much performance on the table when the card is released. Minds are usually made up upon product release, not how it's doing a year later. This sad argument is as pointless as saying "Buh buh buh my AMD card runs DX12/VULCAN benchmarks faster than your NVIDIA!" Soon the argument will be Well it can run (insert future API) faster in benchmarks!
No one buys a card for its future possible/maybe performance. They buy cards on how it performs now.
I have two 290X 8GB cards I reflashed to 390X (which did a whole lot of nothing) in my machine. When I bought them they were still behind Nvidias cards by 5-10% but they were almost half the price. Rebate made them less so it was a good price/perf deal. Right now though nothing AMD has is wroth the price theyre asking. even the 56 is a questionable deal compared to NVIDIA.
I think a lot of people who own 1080 tI also bought vega cards just because that's what that demographic does if you're spending that amount of money on a video card you're in it for the hardware and you're going to get both both
In seriousness, I like to grab yesteryear's high-end games and crank them to the max when I build a new system. Makes me feel powerful. I got a giddy rush the first time I booted FEAR and turned everything to high.
You heard - how about you just made up the number and are completely wrong.
The real news, however, is GeForce GTX 1080 Ti, averaging nearly 80 FPS at 4K in Crysis. Ten years on, and there’s exactly one gaming graphics card (we’re not counting Titan cards here) able to do this. Everything else from AMD and Nvidia fails to break the symbolic 60 FPS mark.
The GTX 1080 Ti can play Crysis @ 3840x2160 Very High settings:
Even fewer of Nvidia’s cards cooperated with our 4K screen, so we pulled the plug at GeForce GTX 780 Ti. It was for the best, perhaps, as that 2013-era card averaged just over 30 FPS.
The real news, however, is GeForce GTX 1080 Ti, averaging nearly 80 FPS at 4K in Crysis. Ten years on, and there’s exactly one gaming graphics card (we’re not counting Titan cards here) able to do this. Everything else from AMD and Nvidia fails to break the symbolic 60 FPS mark.
This is a cheaper V100 more than a faster 1080 Ti. It's not a GeForce card. It's a true prosumer card. It's got nothing to do with AMD. The card is only out because NVIDIA think there is a market for it at $3,000. As long as they can charge $8000+ for the V100, there isn't a world where they sell the Titan V for $1,000 or $1,200. They'd be compromising their own data center line.
Of course NVIDIA will happily sell the card to anyone who wants the absolute fastest card for gaming they can get right now. It probably won't be surpassed until the 1180 Ti (or 2080 Ti or whatever they call it) comes out in a year or so. But that's a pretty rich premium to pay for a year of the fastest gaming card on Earth.
Exactly. At work we are thinking in investing into deep learning but cost of tesla card that are much weaker than this one are easily $6000++. So $3000 is a steal for deep learning application assuming there is no undisclosed limitation in speed.
They're going to be expensive anyway. NVIDIA did continue to sell Titan X at a stupid price when the 295 X2 was creaming it for half the price, if memory serves.
I think the 2 series was when AMD abandoned its little core/multi core idea and went back to making big cores. Yup still didn't work. AMDs HW isnt bad but their drivers fail time and time again to actually utilize the GPUs effectively.
Neither Crossfire or SLI are particularly great. When they work, yeah, but more often than not one card is just sitting there. I'll probably take my second 290 and put it in another rig so the wife can have her own DT and then I'll get her an oculus so she can fill up her SSD with crap instead of mine :'D
I think AMD have gone from having drivers that were often buggy and not able to expose the full potential of their cards, to having drivers that are more stable... and still not able to expose the full potential of their cards. Is Vega's Draw Stream Binning Rasterizer supported yet?
lolol funny stuff for sure 3K really Nvidia then again there will be more than enough tards that will pay that much. Which is pretty sad because if Nvidia feels they sold enough of them at insane price then they will just raise the prices on the lesser cards in the rest of the line up. So I guess we all should thank all the tards in advance for allowing Nvidia to fleece our wallets in the near future.
You;re not thinking about this practically. Imagine this could be used for a self driving car. Wouldn't that be worth an extra $3K? Or used for a home care giver or house maid? Compared to the cost of human labor, this could significantly save costs. Obviously its not for running Doom 4
A quick look at the specs point toward a failed V100 with a defective HBM stack. I wonder if there will be four stacks under the heat sink.
This confirms that the GV100 has video output circuitry as suspected. At 815 mm^2, it was widely thought that nVidia cut that logic to free up as much die area as possible.
You can see that there are some meaningful differences between the two cards that could explain the difference. One is the lower memory bandwidth of the Titan V.
It seems like you'd be much better off with these, and they'd probably be worth the price premium to you for that purpose. They have dedicated logic for deep learning.
You can get 2 of these for the cost of 4x Titan Xp and you'll get 5x the performance for ML applications (assuming NV APIs, etc). It's kind of a no-brainer.
Unlikely; I'd be surprised if NVIDIA launched anything significantly new before May 2018, and when they do it'll probably be whatever replaces the 1080 in the current stack, that's where they usually started each round of new releases.
A bold but logical move. A card finally worth the name "Titan" again, in the sense that it offers significant upgrades over the biggest Geforce. If they sell every Tesla with GV100 they can make right now, why not charge as much for this card? It's going to be worth it to anyone able to use its special features (tensor cores & FP64). It's obviously not for anyone else, but even nVidia doesn't claim so.
You don't know what you are talking about, Nvidia pushed like crazy to get the V100 out to gain market share in datacenter (where they have a LOT of competition; in fact they only have < 10% of market), and it only landed a few months ago
There were several big contracts nVidia signed years ago for super computer parts. They had contractual obligations to get Volta out on time. I see this reason as the main driving factor as to why GV100 has a die size of 815 mm^2. The sane response would be to wait for a new process node to keep yields in check but nVidia simply couldn't without breaking those lucrative contracts.
All of you bitching about the price, grow up and learn to read. This isn't a gaming card, per se. It's a PROSUMER card, meaning it's meant for professional and enterprise use. The price tag is par for the course for what it does. AMD did the same thing for the VEGA prosumer card. Come on, now. Most of you are supposed to be intelligent people, being on this forum at any rate. Use the brains and good sense you're supposed to have...
Prosumer is a level lower than Professional/Enterprise. It's more like for individuals who want the same level as the Professional without the same level as support/warranty.
Such non-entirely-pro cards also have other internal compromises such as smaller cache chunks (not the ones mentioned in the article), slower PCIe return path and non-ECC RAM.
In terms of performance this should be well ahead of the Vega; it's nearly twice the GPU with twice the memory bandwidth. However both are chasing similar markets.
I know at least two people doing design work on Titans of various generations, and no gaming whatsoever. But please continue to demonstrate your hindquarters discomfort.
99% of the comments are about how overpriced this is. Nobody is considering the benefits of this card in processing and mining bitcoin at current prices. A farm of these become a bargain if the bubble continues.
As somebody who has multiple Tesla systems for virtualized environments I am very interested to see how the Titan V works in Citrix as a GPU resource. I have a number of smaller locations with older Tesla cards that these could possibly replace at a fraction of the cost.
Except that professional products are rarely painted gold... Nvidia marketing is squarely aiming at a certain demographic. Compared to people who have more money than brain, the real users who would benefit from such a product are just a minority.
You can expect the vast majority of those cards to end up in gaming rigs for e-peen waving.
StrangerGuy, indeed, it's a peculiar contradiction that people with less money must therefore think they're more intelligent. :D But if they were, wouldn't they be the ones with the money? Really it's just a different spin on jealousy, and the desire to have what others have without the need to expend effort in exchange.
Those are NVIDIA's official numbers. I do agree they seem a bit off, but NVIDIA seems to be basing their official numbers on a slightly lower clockspeed (~1350MHz).
Brian Smith wrote "In fact in some respects I'm surprised NVIDIA is selling a GV100 card for so little; these are GV100 salvage parts that don't make the cut for Tesla - so the alternative would be throwing them away". Rotfl Did you or better someone else independently calculate the production cost of this card? And to remove the troll about huge factories and research cost how about Apple vs Intel processor cost comparison, why the first one being of latest technology still costs 100x cheaper then generation older server chips of intel? :)
Did you Rotfl and calculate the R&D cost of this chip & card?
The onyl price comparison which makes sense right now is to the Tesla V100, and they're selling all of these as fast as they can produce them. So obviously there's a market for something "almost as powerful" for 30% the price, although its majority is not going to be gamers.
Lerianis - congratulation on being the 101st person to make that comment, without reading the article carefully enough to understand this is a professional (not GeForce aka gaming) card.
I'd give you a door prize, but alas the first hundred people who made the same comment took them all.
:D The evolution of English in the US is a curious tale. In some cases the settlers took with them certain verbal forms that stayed as they were in the US, while they changed in Britain. In other cases, the spellings changed to match the rise of new accents and so on. Sometimes the differences can be so subtle it's often hard for either side to realise what's really going on, eg.:
Anyway, this Titan V is priced exactly where it should be; if it weren't, it would priced differently, based on NVIDIA's own estimates of the target market and what it will tolerate; sales experience will determine how this evolves over time, based on supply & demand, just as it should.
I don’t think anyone buys the V100 for $10k. In HPC and large volumes $5k are more realistic and therefore the $3k for a salvage part a good fit. NVlink is only 2-way on PCIe anyway and perhaps an acceptable loss, even if I’d have loved it to be retained, obviously.
I wonder if ECC is also disabled on the HBM: It must obviously be there physically, but they could be mean enough to disable the error correction for market segmentation.
I hope to have some V100 to play with at work these weeks (with 2Way NVLink), then I’ll decide if I want to kneel down in front of my wife and beg for permission to buy one of these instead of a fur coat for her.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
159 Comments
Back to Article
eek2121 - Friday, December 8, 2017 - link
Heh, nope, this card could give me a blowjob every night. I still wouldn't buy it. I"m sucker for fast cards and I am lucky enough to be in a situation where I can afford to buy whatever piece of tech that I want, but 3k for a GPU? naah...Averant - Friday, December 8, 2017 - link
All of you bitching about the price, grow up and learn to read. This isn't a gaming card, per se. It's a PROSUMER card, meaning it's meant for professional and enterprise use. The price tag is par for the course for what it does. AMD did the same thing for the VEGA prosumer card. Come on, now. Most of you are supposed to be intelligent people, being on this forum at any rate. Use the brains and good sense you're supposed to have...Averant - Friday, December 8, 2017 - link
Apologies. Posted to someone's comment. It was not meant to be there.smilingcrow - Friday, December 8, 2017 - link
It's more of a true Professional card than Prosumer with those specs and pricing.Geranium - Friday, December 8, 2017 - link
Well AMD's Vega FE comes with Pro Driver. AFAIK Titan V comes with Geforce driver, not with Pro driver.smilingcrow - Friday, December 8, 2017 - link
Its main benefits are DP and AI and "NVIDIA TITAN users now have free access to GPU-optimized deep learning software on NVIDIA GPU Cloud."It's aimed at business use for sure and no doubt some wealthy prosumers will buy it for the DP performance but it's priced to keep those at bay.
Freakie - Friday, December 8, 2017 - link
Actually everyone has free access to the Nvidia GPU Cloud. And while they don't support any architecture except Pascal and Volta, they don't prevent you from running the environment with their other architectures.Gasaraki88 - Thursday, December 14, 2017 - link
So AMD charges you more and gives you a "Pro" driver and that great but nVidia charges you more, give you actual prosumer hardware (Tensor cores) but give you "regular" Geforce drivers and that's not ok.Whatever Geranium.
Lerianis - Saturday, December 9, 2017 - link
Oh.... okay, Averant.... that makes things more clear. If it is meant for something like 3D movie making for Pixar then it is much more reasonable.Railgun - Sunday, December 10, 2017 - link
The more things change, the more they stay the same.Cellar Door - Friday, December 8, 2017 - link
If you had a professional workload - $3K would be perfectly reasonable as compared to a $10K price tag for a Tesla.In just a few poorly worded sentences, you demonstrated being full of shit and probably a teeny booper looking for attention.
tipoo - Friday, December 8, 2017 - link
It's a bargain for academics who can't swing the 10K one for deep learning with the boss.It's not for you, gaming, or mining.
valinor89 - Saturday, December 9, 2017 - link
But its GOLD! What else do you need? /sBy the way, for that price that gold better be real 24k gold plated! /s /s
rsandru - Sunday, December 10, 2017 - link
Agreed, the gold finish and the price tag just says everything about what's going on in Nvidia's mind.jjj - Friday, December 8, 2017 - link
3k$, these people belong in jail, for rehabilitation.Frenetic Pony - Friday, December 8, 2017 - link
This is what you get without competition. Hopefully with Koduri's firing someone competent will take over the GPU side at AMD.peevee - Friday, December 8, 2017 - link
Let's hope.SanX - Saturday, December 9, 2017 - link
Exactly. The price tags like that of course are derivatives from monopolistic prices of Intel. The 4B transistor mobile processor costs $20 while 3B Intel's one may cost $2000. The trick is that people think that duopolies kill monopolies while we see that even triopolies or quadropolies often are the same monopolies effing brains of people how costly the factories areMrSpadge - Friday, December 8, 2017 - link
I'm glad you're apparently not a judge.sonny73n - Friday, December 8, 2017 - link
Mrspadge, you must be one extreme capitalist who's supporting price gouging, price fixing, monopoly and consumer fraud.So many of the likes of you here. No wonder corporations run our government and jobs being shipped oversea.
A5 - Friday, December 8, 2017 - link
"Charging more than I want them to" != "price fixing" or "price gouging".Words have meaning, and you apparently don't know them.
sonny73n - Friday, December 8, 2017 - link
A5, your argument about meanings of words is just lame. I will not write an essay on every topic so imbecile like you could understandLordSojar - Friday, December 8, 2017 - link
Unfortunately... A5 is right in this instance. This isn't gouging and is actually a very reasonable price for the PURPOSE of this card. This isn't a gaming-centric card, despite it being absurdly capable in that regard.nVidia is doing what any company in their position would do: maximize profits and focus on markets specific to the strengths of their products.
mapesdhs - Monday, December 11, 2017 - link
Indeed, kudos for them for doing so. What a strange world where people think making a profit is evil. Must be the new generation of proto-communists I suppose, the rabid desire for free stuff.The free market has a wonderful metric for deciding what succeeds and what doesn't, it's called price. If it's too high, it won't sell. If it's too low, demand will spike and they'll fail to meet the demand, so raising the price slows the demand. This is money 101. Those who don't like this basic notion of trade just want the efforts of others without paying for it. No wonder we have a welfare state that's out of control.
What's so funny to me is that even $10K is absolutely *nothing* in some pro markets. Some of the systems I used to work with were in the $500K to $1.5M range, and the companies that used them were able to do things that meant their investment paid for itself in just the first day of use. Hobbyists and gamers really have little idea of the true nature of industrial hardware; nothing wrong with that, they don't need to know, but the context of articles like this ought to be enough to convey that there's a deeper big money world beyond the narrow slice of life called gaming.
MrSpadge - Friday, December 8, 2017 - link
It's not "extreme capitalist" to believe in some basic freedom: they are free to offer products and you are free not to buy them, if you don't want to. Do you think that's wrong? Do you also imprison every Porsche or Lamborghini merchant as you drive by, because you think they should offer you their cars for less?tamalero - Friday, December 8, 2017 - link
Incorrect analogy, because cars like Porshe and Lamborghinis do not have any real function other than "going fast" and "look good". Aka its a luxury unnecessary that can easily be replaced with multiple brands and for different costs according to your needs.bji - Friday, December 8, 2017 - link
This is a really dumb comment.Drumsticks - Friday, December 8, 2017 - link
Er, by that definition, pretty much any use of the computer other than directly supporting your societal functions (i.e. if it's for work, in which case you typically aren't paying for it), is unnecessary. Even if you do pay for your own hardware, faster hardware can quickly offset higher prices, and for any work related tasks, this looks to be at a decent price.Jconsumer - Friday, December 8, 2017 - link
or a 1950's American Capitalist Automobile like they do in Cuba. Viva La Communism!!!peevee - Friday, December 8, 2017 - link
I bet he does. Where do you think bolshevicks and maoists come from?Jconsumer - Friday, December 8, 2017 - link
These communist trolls never see the irony of commenting on tech message boards. All they have to do is produce a better product and offer it for the people instead of complaining that nobody else does it.sonny73n - Friday, December 8, 2017 - link
Jconsumer, you sound like a hateful troll from the '60s but buying everything that made in China today.mapesdhs - Monday, December 11, 2017 - link
Jconsumer, that would mean they would have to present their talents and skills to the free market, but of course they don't have any. The foundation of communism: get free stuff, do nothing in return, cause chaos. Well, that and of course the eventual murder of a hell of a lot of people.Jconsumer - Friday, December 8, 2017 - link
It's always the luxury items that bring these basement dweller communists out to complain about "rampant capitalism." In their never-never land that never can exist, nor has ever existed, Willy Wonka offers them free prosumer video cards, evian bottle waters and Masserattis, courtesy of the Red Republic.sonny73n - Friday, December 8, 2017 - link
Porsche and Lamborghini are luxury items. People wants them but they don't need them. If you need a car, there will be many others that fit your need. Honda or Toyota will do just fine, unlike some computer components you need to speed up your productivity.MrSpadge - Friday, December 8, 2017 - link
You want to base this on "computer components you need to speed up your productivity"? So what exactly are you losing by this new offer? You are still free to buy 4 AMD MI25 instead for inference, or the Tesla V100. For FP64 you can still buy 3 Xeon Phi of Hawaii Fire Pro. If you think that's a better deal - fine, vote with your wallet! Otherwise consider again who's making a bad offer.And if you want to compare this card based on FP32 performance - well, it can do that, but it's obviously not a suitable tool for that and I'm sure nVidia is totally fine if you go with a GTX1080Ti instead.
StrangerGuy - Monday, December 11, 2017 - link
Ah yes, "productivity", aka a pathetic attempt to excuse your entitlement syndrome.Jconsumer - Friday, December 8, 2017 - link
You must also be an extreme capitalist being on here. No communist government lacky ever made a video card the likes of NVIDIAmapesdhs - Monday, December 11, 2017 - link
Here's a perfect summary of the kind of world sonny73n would have us all live in:https://www.youtube.com/watch?v=zw-FF6CPmvs
BenSkywalker - Friday, December 8, 2017 - link
This has four times the DL performance of AMD's MI25 and eight times the DP performance, oh yeah, unlike that paper part- you can buy this one the day it was announced. This is a comical level slaughter against team red on a *VALUE* basis alone.These comment threads are rather comical for people who follow the wider industry. AMD is not competition in this segment, nobody really is. Intel, Google and Facebook are spending billions of dollars trying to go toe to toe with nVidia, but all of those years and dollars sunk into CUDA- which a ton of team red fanboys blasted as being stupid, has resulted in utter domination.
StrangerGuy - Monday, December 11, 2017 - link
The vocal miser crowd (aka mostly red fanboys) doesn't seem to get the simple logic that the smart companies like NV and Apple knows they can't make money off them and thus can care less about their inane opinions.mapesdhs - Monday, December 11, 2017 - link
In a way what NVIDIA is doing reminds me of the headier days of SGI. There was demand for a gfx product that could make possible levels of performance which would not be available via COTS products for at least 5 years, so SGI built what companies wanted. They were expensive as heck (the Onyx in my garage was original $1M), but the performance was lightyears ahead of its time, ditto the features. I saw an Onyx in 1993 doing 12bits/channel and fully subsampled AA, mipmapped textured 3D at HD at a time when proper 3D simply didn't exist on PCs, and wouldn't for several more years (something SGI itself helped get going). Point being, the commercial demand was there, so they made stuff to satisfy that demand, and they made freakin' fortune in the process (some reseller sales reps became millionaires). SGI died due to numerous mistakes, but it's not as if the core demand for something advanced has ever gone away, especially these days with such heavy burdens placed on intelligence gathering, image analysis, data management, etc. Even today it's hard to put together a PC that can do what the Onyx2 Group Station for Defense Imaging could do almost 20 years ago.If NVIDIA reckons there's a demand, I'm sure they'll make something that costs 10x as much, and still sell plenty of them, though perhaps these days such products are done on the quiet, unlike SGI which made its tech available openly (more or less; DoD and others received customised versions).
smilingcrow - Friday, December 8, 2017 - link
How dare they release a card for business users where the price is not an issue as the return on investment makes it good value.mapesdhs - Monday, December 11, 2017 - link
(ok so that's a far briefer and better way of putting it. :D)dagnamit - Friday, December 8, 2017 - link
At $3k this might be worth it, if it's a true pro-sumer card like the original Titan. Resale value could be very good. I knew people that sold their original Titans for about $100 less than the original price because it was so good as a workstation card. $100 to rent a bleeding edge card only to replace it with the, equal or better, GTX version will be worth it to some. If it gets even within 20% of the V100 for the same tasks, it'll FLY off the shelves.Drumsticks - Friday, December 8, 2017 - link
Pretty much everything about this card screams compute, not gaming. 1/2 FP64, and 110TF of DL make this a pretty damn nice compute card. Plus, it's based on GV100. As a discounted V100, its price is pretty good.ddriver - Friday, December 8, 2017 - link
Actually, if the claimed FP64 rates are true, I would personally consider it a decent buy, something nvidia hasn't had for a loooong time.It is basically a tesla V100 with one defective and disabled memory channel. Asking 10k for the full version of the chip - that's truly criminal. But getting almost the same performance and 3/4 of the memory at less than 1/3 of the price is pretty decent.
Now if nvidia were just to brush up their OpenCL support.
Although it is highly likely that the FP64 rates are limited in software. It would be very odd if they offered that much FP64 performance at that price point. Their DPFLOP price us very consistent across their product range. It is very close for say something gaming grade like the 1080 ti and the quadro gp100, the 1080ti being that "affordable" only because its FP64 rate is capped at 1/32 of the SP rate.
nivedita - Friday, December 8, 2017 - link
The pro equivalent of the 1080ti is the P6000. Those are based on different silicon than the GP100 which is the one that has 1/2-speed DP compute. The P6000 also offers only 1/32 FP64.mapesdhs - Monday, December 11, 2017 - link
No price is "criminal", that's just stupid. The market decides what makes for a viable price. If it's too high, the product won't sell. I know a company that bought a machine for $1.5M; for their particular task, the system paid for itself in less than an hour. Kinda weird saying some product is too expensive when the price in question doesn't even register on the radar of what some types of tech cost in other fields.Sttm - Friday, December 8, 2017 - link
This is what happens when AMD is total crap and Nvidia just gets to run wild. They know they got at least a year before they have to sell Volta at reasonable prices. So why not give some whales a $3000 version for now!INTEL, YOU ARE ONLY HOPE!
PixyMisa - Friday, December 8, 2017 - link
It's an 815mm^2 chip. That's never going to be cheap.Still, for the professional users who need it (particularly if you need the DP performance) that's not a bad price.
coogerjr - Friday, December 8, 2017 - link
That's a really bad price when you consider the fact that the 2080 is half a year away and will be 85% as fastlooncraz - Friday, December 8, 2017 - link
If you do professional work (or your time is very valuable), then paying the premium can be very much worth it.I paid $650 for a Vega 64 then another $100 for the EK-FC waterblock - so $750 all in for a Vega 64.
In the time I've owned it, not mining at all, I estimate it has saved me about $900 in time over the 980ti it replaced. I've also been able to tweak it to only use about 160W without losing much performance (maybe 5% in my specific compute task - which I have now completed, so no more savings to be had for me).
If I was monetizing my work, this card could make me some good money.
LithiumFirefly - Friday, December 8, 2017 - link
If you're doing professional worth then you should be looking at AMDQyu - Friday, December 8, 2017 - link
If you are doing professional computational work you CANNOT consider AMD unfortunately. Modern compilers just don't support them... This card is a STEAL for people in engineering simulations, AI, and finance.Demiurge - Monday, December 11, 2017 - link
That doesn't sounds right, Qyu. By the way, what compiler are you using to compile CUDA code on?SanX - Saturday, December 9, 2017 - link
Ask somebody else who does not believe you to doublecheck your calculations. You will find the opposite conclusions.Zanor - Friday, December 8, 2017 - link
For gaming maybe. But for AI, simulation, or other professional work you have to compare it to a quadro or tesla. Looks at it that way I don't think it's a terrible deal. It could be cheaper and definitely will be, but with AMD limping behind there's nothing we can do.nerd1 - Friday, December 8, 2017 - link
Literally EVERY machine learning researchers I know use either 1080 or 1080 ti.nevcairiel - Friday, December 8, 2017 - link
Possibly because they had no relatively affordable option with Tensor Cores yet. The Tesla V100 was significantly more expensive then this.peevee - Friday, December 8, 2017 - link
In games maybe, but they will cut DP, like, by 16 times.LordSojar - Sunday, December 10, 2017 - link
This. Is. Not. A. Gaming. Card. Irrelevant comparison.mapesdhs - Monday, December 11, 2017 - link
If it pays for itself, it's a great price. Who are you to judge? It's all about TCO and ROI; the upfront raw cost is rarely a deciding factor. Otherwise, SGI would never have done what it was able to do.mapesdhs - Monday, December 11, 2017 - link
(NB: I was replying to coogerjr; this reply mechanism is weird...)beginner99 - Friday, December 8, 2017 - link
Yeah I consider it cheap especially for deep learning if you compare it to existing tesla cards. Of course it's retarded as a gaming card.IGTrading - Friday, December 8, 2017 - link
It's a bit o the high side.What was the previous "largest die size" GPU ?
Totally - Friday, December 8, 2017 - link
I believe it was Fermijabbadap - Friday, December 8, 2017 - link
Fermi was relativity small chip compared to these monsters(529mm²). Consumer/prosumer side biggest was Maxwell gm200 @ 601mm². Pascal GP100 was 610mm², but that was never entered even prosumer side Titan series.IGTrading - Friday, December 8, 2017 - link
Intel is the "hope" ?!?! What planet are you coming from ?! :)AMD is "total crap" !??! Like AMD Radeon Fury X beating GF 980 TI in all applications possible in FHD and UHD 4K ?!?
Trust me, VEGA is getting better with every driver release and in Spring it will be faster than 1080 TI with people wondering why they've bought nVIDIA and then calming it is "because of the environment and the planet" ;)
Don't you think all those GF 980 TI buyers would like an AMD Fury X now ?! The same will happen with VEGA.
Yeah it's a bit more power hungry, but it is capable and competitive while the "Intel hope" is truly "total crap" .
Either people's brains are getting less capable of comprehending the world around them or the marketing departments are using a new king of detergent for the washing .... :)
silverblue - Friday, December 8, 2017 - link
That's a lot of hope you've got there. Here is a video that's worth watching concerning the current performance of the 980 Ti and Fury X:https://www.youtube.com/watch?v=IeUyvEVB984
Vega 64 beats the 1080 Ti in one game as far as I can see (Forza 7 - and even then nobody can understand why), and loses handily in a good number of others. AMD shouldn't be taking a year to get the best performance out of its cards. Also, we all know how power hungry Vega 64 LC is, before you go anywhere near overclocking, so "a bit more power hungry" is inaccurate.
Intel do now have the expertise to develop decent GPUs, but that sort of thing takes time.
From my own personal viewpoint, those 980 Ti users are really not going to be crying into their cereal over not getting a Fury X (especially when you hit that 4GB limit), and those 1080 Ti users are needing surgery to stitch up their sides over the inference that Vega is going to blow them all away in a few months.
mkaibear - Friday, December 8, 2017 - link
There's not a lot of point arguing with IGTrading. He's fanboytastic.Manch - Friday, December 8, 2017 - link
" AMD shouldn't be taking a year to get the best performance out of its cards."THIS ^^^^
That's all you really had to say. LOL
AMD needs to get a handle on their damn drivers. There's no reason to leave THAT much performance on the table when the card is released. Minds are usually made up upon product release, not how it's doing a year later. This sad argument is as pointless as saying "Buh buh buh my AMD card runs DX12/VULCAN benchmarks faster than your NVIDIA!" Soon the argument will be Well it can run (insert future API) faster in benchmarks!
No one buys a card for its future possible/maybe performance. They buy cards on how it performs now.
I have two 290X 8GB cards I reflashed to 390X (which did a whole lot of nothing) in my machine. When I bought them they were still behind Nvidias cards by 5-10% but they were almost half the price. Rebate made them less so it was a good price/perf deal. Right now though nothing AMD has is wroth the price theyre asking. even the 56 is a questionable deal compared to NVIDIA.
LithiumFirefly - Friday, December 8, 2017 - link
I think a lot of people who own 1080 tI also bought vega cards just because that's what that demographic does if you're spending that amount of money on a video card you're in it for the hardware and you're going to get both bothLithiumFirefly - Friday, December 8, 2017 - link
Intel just signed a contract with AMD to use their gpus as an SOC on mobile platformsLordSojar - Sunday, December 10, 2017 - link
Because we all know how efficient AMD's GPU architectures are... oh... wait... Sorry, I confused AMD with NVIDIA. My bad.haukionkannel - Friday, December 8, 2017 - link
Good, good!Now we Are talking real money!
jfallen - Friday, December 8, 2017 - link
But can it play Crysis?(sorry someone had to do it)
Lord of the Bored - Friday, December 8, 2017 - link
I heard it will get about 45FPS.In seriousness, I like to grab yesteryear's high-end games and crank them to the max when I build a new system. Makes me feel powerful. I got a giddy rush the first time I booted FEAR and turned everything to high.
HighTech4US - Friday, December 8, 2017 - link
You heard - how about you just made up the number and are completely wrong.The real news, however, is GeForce GTX 1080 Ti, averaging nearly 80 FPS at 4K in Crysis. Ten years on, and there’s exactly one gaming graphics card (we’re not counting Titan cards here) able to do this. Everything else from AMD and Nvidia fails to break the symbolic 60 FPS mark.
http://www.tomshardware.com/reviews/crysis-10-year...
The Titan V is more powerful than the 1080 Ti it will do even better.
PeachNCream - Friday, December 8, 2017 - link
Check the batteries in your joke detector.Holliday75 - Friday, December 8, 2017 - link
We all know its dead.HighTech4US - Friday, December 8, 2017 - link
Yes it can.The GTX 1080 Ti can play Crysis @ 3840x2160 Very High settings:
Even fewer of Nvidia’s cards cooperated with our 4K screen, so we pulled the plug at GeForce GTX 780 Ti. It was for the best, perhaps, as that 2013-era card averaged just over 30 FPS.
The real news, however, is GeForce GTX 1080 Ti, averaging nearly 80 FPS at 4K in Crysis. Ten years on, and there’s exactly one gaming graphics card (we’re not counting Titan cards here) able to do this. Everything else from AMD and Nvidia fails to break the symbolic 60 FPS mark.
http://www.tomshardware.com/reviews/crysis-10-year...
Since the Titan V is more powerful than the 1080 Ti it would do even better.
Gc - Saturday, December 9, 2017 - link
But can it play Crysis as well as a young human?If so, are gamers obsolete? /s
(Or choose your own autonomous task.)
Yojimbo - Friday, December 8, 2017 - link
This is a cheaper V100 more than a faster 1080 Ti. It's not a GeForce card. It's a true prosumer card. It's got nothing to do with AMD. The card is only out because NVIDIA think there is a market for it at $3,000. As long as they can charge $8000+ for the V100, there isn't a world where they sell the Titan V for $1,000 or $1,200. They'd be compromising their own data center line.Of course NVIDIA will happily sell the card to anyone who wants the absolute fastest card for gaming they can get right now. It probably won't be surpassed until the 1180 Ti (or 2080 Ti or whatever they call it) comes out in a year or so. But that's a pretty rich premium to pay for a year of the fastest gaming card on Earth.
beginner99 - Friday, December 8, 2017 - link
Exactly. At work we are thinking in investing into deep learning but cost of tesla card that are much weaker than this one are easily $6000++. So $3000 is a steal for deep learning application assuming there is no undisclosed limitation in speed.WB312 - Friday, December 8, 2017 - link
Oh boy Volta GPUs are gonna be expensive with no competition from AMD in the high end.silverblue - Friday, December 8, 2017 - link
They're going to be expensive anyway. NVIDIA did continue to sell Titan X at a stupid price when the 295 X2 was creaming it for half the price, if memory serves.Klimax - Friday, December 8, 2017 - link
If you were lucky to have Crossfire to work. (SLI was bit better, but it wasn't "universal" solution either)Heh, AMD needed two chips to beat one...
Manch - Friday, December 8, 2017 - link
I think the 2 series was when AMD abandoned its little core/multi core idea and went back to making big cores. Yup still didn't work. AMDs HW isnt bad but their drivers fail time and time again to actually utilize the GPUs effectively.Neither Crossfire or SLI are particularly great. When they work, yeah, but more often than not one card is just sitting there. I'll probably take my second 290 and put it in another rig so the wife can have her own DT and then I'll get her an oculus so she can fill up her SSD with crap instead of mine :'D
silverblue - Friday, December 8, 2017 - link
I think AMD have gone from having drivers that were often buggy and not able to expose the full potential of their cards, to having drivers that are more stable... and still not able to expose the full potential of their cards. Is Vega's Draw Stream Binning Rasterizer supported yet?silverblue - Friday, December 8, 2017 - link
Actually, after some checking, there's the possibility that Vega was shipped without it due to a fault, but I can't confirm anything.rocky12345 - Friday, December 8, 2017 - link
lolol funny stuff for sure 3K really Nvidia then again there will be more than enough tards that will pay that much. Which is pretty sad because if Nvidia feels they sold enough of them at insane price then they will just raise the prices on the lesser cards in the rest of the line up. So I guess we all should thank all the tards in advance for allowing Nvidia to fleece our wallets in the near future.webdoctors - Friday, December 8, 2017 - link
You;re not thinking about this practically. Imagine this could be used for a self driving car. Wouldn't that be worth an extra $3K? Or used for a home care giver or house maid? Compared to the cost of human labor, this could significantly save costs. Obviously its not for running Doom 4Devo2007 - Friday, December 8, 2017 - link
NVidia's page for the TITAN V does specify NVLINK, so it is indeed enabled for this card.Nate Oh - Friday, December 8, 2017 - link
Where on the page do you see it? I can't seem to locate itDevo2007 - Friday, December 8, 2017 - link
Ok, my bad. I clicked the link for more info on Volta and it brought me to the page that talks about everything including NVLINK.https://www.nvidia.com/en-us/data-center/volta-gpu...
It wasn't specific to the TITAN V
Kevin G - Friday, December 8, 2017 - link
A quick look at the specs point toward a failed V100 with a defective HBM stack. I wonder if there will be four stacks under the heat sink.This confirms that the GV100 has video output circuitry as suspected. At 815 mm^2, it was widely thought that nVidia cut that logic to free up as much die area as possible.
palladium - Friday, December 8, 2017 - link
hmm, why does the Titan V have higher SP/DP performance than the Tesla, but lower Deep Learning perf? Is it some form of driver limitation?arayoflight - Friday, December 8, 2017 - link
Faster cores, slower memoryDrumsticks - Friday, December 8, 2017 - link
Deep Learning performance is derived in part by the Tensor Cores. They might be running at a lower frequency than on the V100.twtech - Friday, December 8, 2017 - link
You can see that there are some meaningful differences between the two cards that could explain the difference. One is the lower memory bandwidth of the Titan V.nivedita - Friday, December 8, 2017 - link
It doesn't?versesuvius - Friday, December 8, 2017 - link
So far, the best mining card ever!nerd1 - Friday, December 8, 2017 - link
I was planning to build 4x Titan XP deep learning rig... How does this thing compare to titan xp?Drumsticks - Friday, December 8, 2017 - link
One of these should be faster than a quadruple Titan Xp or Titan X (Pascal) for training, I think.twtech - Friday, December 8, 2017 - link
It seems like you'd be much better off with these, and they'd probably be worth the price premium to you for that purpose. They have dedicated logic for deep learning.arayoflight - Friday, December 8, 2017 - link
110 TFLOPS(Titan V) vs 48.4(assuming perfect scaling) on 4x titan Xp.Yeah, one of these is better for deep learning.A5 - Friday, December 8, 2017 - link
You can get 2 of these for the cost of 4x Titan Xp and you'll get 5x the performance for ML applications (assuming NV APIs, etc). It's kind of a no-brainer.colonelclaw - Friday, December 8, 2017 - link
The best thing about this card is that it means the GeForce 11 series can't be too far away. I'm hoping before Easter 2018.mapesdhs - Monday, December 11, 2017 - link
Unlikely; I'd be surprised if NVIDIA launched anything significantly new before May 2018, and when they do it'll probably be whatever replaces the 1080 in the current stack, that's where they usually started each round of new releases.piroroadkill - Friday, December 8, 2017 - link
12GB seems a bit cheap for that money, seeing as you can get 12GB cards from NVIDIA for much less...piroroadkill - Friday, December 8, 2017 - link
I get that it's much more expensive HBM2, but still... For a card that costs so much, it still only has the same amount of VRAM as the older Titans.mapesdhs - Monday, December 11, 2017 - link
Which suggests NVIDIA is aware that for the target market they don't need more than 12GB atm.MrSpadge - Friday, December 8, 2017 - link
A bold but logical move. A card finally worth the name "Titan" again, in the sense that it offers significant upgrades over the biggest Geforce. If they sell every Tesla with GV100 they can make right now, why not charge as much for this card? It's going to be worth it to anyone able to use its special features (tensor cores & FP64). It's obviously not for anyone else, but even nVidia doesn't claim so.yhselp - Friday, December 8, 2017 - link
Coating inspired by President Trump.mapesdhs - Monday, December 11, 2017 - link
Good, NVIDIA can help make America great again. 8)artk2219 - Tuesday, December 12, 2017 - link
Lol, but i believe you mean rob the American public blind, and screw them over for a generation.LithiumFirefly - Friday, December 8, 2017 - link
Breakneck Pace on a video card line that's been sandbagged for over a year because they have no competitionMadManMark - Sunday, December 10, 2017 - link
You don't know what you are talking about, Nvidia pushed like crazy to get the V100 out to gain market share in datacenter (where they have a LOT of competition; in fact they only have < 10% of market), and it only landed a few months agoKevin G - Monday, December 11, 2017 - link
There were several big contracts nVidia signed years ago for super computer parts. They had contractual obligations to get Volta out on time. I see this reason as the main driving factor as to why GV100 has a die size of 815 mm^2. The sane response would be to wait for a new process node to keep yields in check but nVidia simply couldn't without breaking those lucrative contracts.Averant - Friday, December 8, 2017 - link
All of you bitching about the price, grow up and learn to read. This isn't a gaming card, per se. It's a PROSUMER card, meaning it's meant for professional and enterprise use. The price tag is par for the course for what it does. AMD did the same thing for the VEGA prosumer card. Come on, now. Most of you are supposed to be intelligent people, being on this forum at any rate. Use the brains and good sense you're supposed to have...mr_tawan - Friday, December 8, 2017 - link
Prosumer is a level lower than Professional/Enterprise. It's more like for individuals who want the same level as the Professional without the same level as support/warranty.mapesdhs - Monday, December 11, 2017 - link
Such non-entirely-pro cards also have other internal compromises such as smaller cache chunks (not the ones mentioned in the article), slower PCIe return path and non-ECC RAM.bananaforscale - Friday, December 8, 2017 - link
What's next? Saturn V?Yojimbo - Friday, December 8, 2017 - link
Saturn V is the name of their supercomputer.Hxx - Friday, December 8, 2017 - link
Hopefully this means that the consumer graphics cards are not too far awayUltraWide - Friday, December 8, 2017 - link
If you can't afford it, just move on. Don't waste you life complaining about it. ;)beglarm - Friday, December 8, 2017 - link
Is this card comparable to Vega Founders Edition?Ryan Smith - Saturday, December 9, 2017 - link
Yes and no.In terms of performance this should be well ahead of the Vega; it's nearly twice the GPU with twice the memory bandwidth. However both are chasing similar markets.
peevee - Friday, December 8, 2017 - link
A decent cheap upgrade for an outdated low-end PC.Sanityisreal - Friday, December 8, 2017 - link
Lol, many of the rubes saying these are "prosumer" cards, such a sexy term that was created to take place of the shame that is bitcoin.These cards Nvidia are pumping out are nothing more than expensive mining cards regardless of the political talk otherwise.
"Prosumer".....too funny. Stupid people deserve their fate.
Hurr Durr - Wednesday, December 13, 2017 - link
I know at least two people doing design work on Titans of various generations, and no gaming whatsoever. But please continue to demonstrate your hindquarters discomfort.ColtCooper - Friday, December 8, 2017 - link
99% of the comments are about how overpriced this is. Nobody is considering the benefits of this card in processing and mining bitcoin at current prices. A farm of these become a bargain if the bubble continues.nevcairiel - Saturday, December 9, 2017 - link
No GPU is really effective in mining Bitcoin anymore, and the cost/performance of such an expensive card definitely doesn't make this a good choice.If you want to mine Bitcoin, you need to go ASIC.
mapesdhs - Monday, December 11, 2017 - link
Indeed, from what I read at least a year ago the rise of ASIC/FPGA solutions had completely pushed aside ordinary GPUs from BC mining.Lakados - Friday, December 8, 2017 - link
As somebody who has multiple Tesla systems for virtualized environments I am very interested to see how the Titan V works in Citrix as a GPU resource. I have a number of smaller locations with older Tesla cards that these could possibly replace at a fraction of the cost.rsandru - Sunday, December 10, 2017 - link
Except that professional products are rarely painted gold... Nvidia marketing is squarely aiming at a certain demographic. Compared to people who have more money than brain, the real users who would benefit from such a product are just a minority.You can expect the vast majority of those cards to end up in gaming rigs for e-peen waving.
StrangerGuy - Monday, December 11, 2017 - link
A big lol to you, because NV will be too busy raking in the dough to give a hoot to the noisy misers like you.Do you also happen to whine about how other luxury stuff like BMW 7s etc are overpriced?
mapesdhs - Monday, December 11, 2017 - link
StrangerGuy, indeed, it's a peculiar contradiction that people with less money must therefore think they're more intelligent. :D But if they were, wouldn't they be the ones with the money? Really it's just a different spin on jealousy, and the desire to have what others have without the need to expend effort in exchange.Dr. Swag - Friday, December 8, 2017 - link
The FP32 TFLOP numbers are off. You list it as lower than the V100 yet the clock speeds are higher...Ryan Smith - Saturday, December 9, 2017 - link
Those are NVIDIA's official numbers. I do agree they seem a bit off, but NVIDIA seems to be basing their official numbers on a slightly lower clockspeed (~1350MHz).SanX - Saturday, December 9, 2017 - link
Brian Smith wrote "In fact in some respects I'm surprised NVIDIA is selling a GV100 card for so little; these are GV100 salvage parts that don't make the cut for Tesla - so the alternative would be throwing them away".Rotfl Did you or better someone else independently calculate the production cost of this card? And to remove the troll about huge factories and research cost how about Apple vs Intel processor cost comparison, why the first one being of latest technology still costs 100x cheaper then generation older server chips of intel? :)
MrSpadge - Monday, December 11, 2017 - link
Did you Rotfl and calculate the R&D cost of this chip & card?The onyl price comparison which makes sense right now is to the Tesla V100, and they're selling all of these as fast as they can produce them. So obviously there's a market for something "almost as powerful" for 30% the price, although its majority is not going to be gamers.
oranos - Saturday, December 9, 2017 - link
lol...nedjinski - Saturday, December 9, 2017 - link
ed: "For Sale Now", not "On Sale Now".Railgun - Sunday, December 10, 2017 - link
Don’t be pedantic. Both are correct. This is an international site where they’re correct depending on one’s location.Lerianis - Saturday, December 9, 2017 - link
WHOA! 3K for ONE video card? Uh uh.... too expensive unless it is like.... 200% faster than a 600 dollar video card.MadManMark - Sunday, December 10, 2017 - link
Lerianis - congratulation on being the 101st person to make that comment, without reading the article carefully enough to understand this is a professional (not GeForce aka gaming) card.I'd give you a door prize, but alas the first hundred people who made the same comment took them all.
Pinn - Sunday, December 10, 2017 - link
Maybe a consolation prize for messing up math?Agent Smith - Monday, December 11, 2017 - link
Its called Maths and i won’t ever Americanise that word ever. Just sounds so wrong like the re-naming of planetoids to asteroids.mapesdhs - Monday, December 11, 2017 - link
:D The evolution of English in the US is a curious tale. In some cases the settlers took with them certain verbal forms that stayed as they were in the US, while they changed in Britain. In other cases, the spellings changed to match the rise of new accents and so on. Sometimes the differences can be so subtle it's often hard for either side to realise what's really going on, eg.:https://www.youtube.com/watch?v=UpX8NZMxp9Q
Anyway, this Titan V is priced exactly where it should be; if it weren't, it would priced differently, based on NVIDIA's own estimates of the target market and what it will tolerate; sales experience will determine how this evolves over time, based on supply & demand, just as it should.
nukunukoo - Monday, December 11, 2017 - link
Yes, great, but can it do Crysis?oranos - Monday, December 11, 2017 - link
its annoying because you know that tryhards like k1ngpin will buy 4 of these, overclock them with nitrogen, just to put up some bench numbersNotmyusualid - Thursday, December 14, 2017 - link
Eth: 82MH/s, slight overclock, 270W. Wow.abufrejoval - Thursday, December 14, 2017 - link
I don’t think anyone buys the V100 for $10k. In HPC and large volumes $5k are more realistic and therefore the $3k for a salvage part a good fit. NVlink is only 2-way on PCIe anyway and perhaps an acceptable loss, even if I’d have loved it to be retained, obviously.I wonder if ECC is also disabled on the HBM: It must obviously be there physically, but they could be mean enough to disable the error correction for market segmentation.
I hope to have some V100 to play with at work these weeks (with 2Way NVLink), then I’ll decide if I want to kneel down in front of my wife and beg for permission to buy one of these instead of a fur coat for her.
BillF - Friday, January 26, 2018 - link
Hi - I need to buy a LOT of these ASAP, and advice who has supply? :-)craigfundings - Monday, May 28, 2018 - link
We bring you a genuine and certified credit offer at a low interest rate of 4%. Contact us for more details if you are honestly interested please.