KitGuru KitGuru.net – Tech News | Hardware News | Hardware Reviews | IOS | Mobile | Gaming | Graphics Cards
Some new details surrounding Nvidia's GP100 Pascal-based GPU have been uncovered this week, revealing that it is targeting some pretty high-performance numbers with 4 TFLOPS of double precision compute performance and three times that in single precision performance. This information dates back to a presentation given by Spanish university professor, Manuel Ujaldon in June last year.
These performance targets, if accurate, signal that Nvidia might be focussing once again on getting a 1:3 single to double precision performance ratio, which Nvidia stepped away from with Maxwell, which focussed more on single precision.
As Videocardz points out, this also gives us a hint as to how many CUDA cores the GP100 (‘Big Pascal) GPU will have. Under the assumption that Nvidia will keep the core clock at 1000 MHz, then this GPU could have 6144 CUDA cores, which is twice as much as GM200. This level of performance would likely be reserved for a future professional GPU or a TITAN series card.
Obviously, we don't know how accurate the original presentation was or if it would still remain accurate today almost a year later so this is mostly speculation. Either way, we should learn more in the coming months as we get closer to the official launch of Pascal.
KitGuru Says: A lot of people are waiting on Nvidia's Pascal launch this year so hearing about potential performance is pretty exciting. Will any of you guys be upgrading once Pascal hits?
2nd quarter release in 2016 hopefully. Im wondering at the cost however. With this performance gain, i hope it still stays at the 1 – 1.5k range.
I want Volta, I want it in 2018 first quarter and I want it for 699$ 😀
NV will repeat OG Titan -> 780Ghz / 780Ti (Kepler gen) and Titan X -> 980Ti. Unless one needs DP or 32GB of HBM2, for gaming, it’s way smarter to get the 780Ti/980Ti version or 2 of those (for $1500).
Ever since the Titan brand came out, for pure gaming it has been a pure rip-off.
Doubtful. NV released true flagships every 2 years or so. Since 980Ti/Titan X came out Spring 2015, we won’t see a true successor to them in the consumer space (fully unlocked) until next year. That means flagship Volta is likely a 2019 product. They could sell you a mid-range GV204 style chip (aka 680/980 successor) for $699. That’s actually where the industry is heading with mid-range die already at $500-550 for the last 2 generations. I can see NV bumping it to $599-649 this year with 980’s successor if say it beats 980Ti by 30-40%.
Flagship Volta is mid-2017, coinciding with the new supercomputers at the DOE built with Volta-based Teslas and Power 9 CPUs.
Yep, plan on getting a gpu this year, and will probably ride on that until 10nm parts are available in a few more years. It is very exciting, the world has been stuck on 28nm gpu’s for what? 4 years now? I forget because it’s been so stinking long! I have a 7870HD, this card was basically rebranded like 3 times, the 270 is basically the same card for example.
Unless the reviews for AMD’s cards are hot, I will definitely pay the premium difference to go NV from now on. Every time I try AMD I regret it, mostly because of inferior drivers and weird graphic problems in games and apps, no wonder 75% of people on steam use NV. =
Sorry AMD fans, I am committing a mutiny, I’ll walk ye plank and leave this ship all on me own power tank ya very much! Bye!
BRING ME TEH PASCAL! I want it now =,(
still hesitant to put much faith in this first round of cards with HBM.
Do you really think they gonna kill the GM market and their consumers ?! People are buying GTX 980 Ti for 600-700$-€ right now. Next GP x80 flagshit card will be 20-30% better for 550-600$-€ (launch price) ! Don’t expect something crazy for 2016. HBM will only be for the next TITAN card
❝my .friend’s mate Is getting 98$. HOURLY. on the internet.❞….two days ago new McLaren. F1 bought after earning 18,512$,,,this was my previous month’s paycheck ,and-a little over, 17k$ Last month ..3-5 h/r of work a day ..with extra open doors & weekly. paychecks.. it’s realy the easiest work I have ever Do.. I Joined This 7 months ago and now making over 87$p/h.Learn. More right Here;;/268➤➤➤➤➤ http://GlobalSuperEmploymentVacanciesReportsJobs/GetPaid/98$hourly…. .❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2:❦2::::;;/268.
Not really first round HBM IRS second gen cards with HBM2
Good choice, but at the same time bad. G-Sync monitors are less and more expensive, if you looking for one. Nvidia does not support open standarts and crippling their previous gen gpus, meaning you will have to upgrade…a lot.
Game changer for card sales might be more DX12 mix and match SLI/Crossfire. Just bought a 980Ti last summer and would be cool to add one of these in 2017 and run them both.
I want to like AMD, I would love to support pro-open standard people! But at the end of the day, it’s just a computer, not a food charity or something. I want power, I want solid and consistent performance, I want some of the features that AMD is missing.
Hopefully AMD hits a home run this time, but I seriously doubt it with their ever shrinking RnD budget as they are due for bankruptcy before to long =,(
And I disagree with needing to upgrade a lot, because NV cripple their own cards.
We need to upgrade a lot with either brand, as both companies upgrade to smaller transistors every so often. Which means more powerful cards, and games studios can then put out games with better graphics.
The old 35nm cards are junk not because NV didn’t support them for new stuff properly, they are junk because we have 28nm cards now, they are much faster… This will be the same story for 1614nm.
I find old AMD GPU’s and old NV GPU’s to be about equally bad at playing new games, for the most part anyways… Sound about right or do you disagree?
Fully unlocked flagship Pascal is either late 2016 or early 2017. Even if Volta Teslas are available for supercomputers, my entire comment was regarding consumer products. NV has not launched 2 major flagships just 1 year apart since at least Tesla generation (GTX280 -> refresh 285 –> real flagship was Fermi GTX480/580), which suggests there is no way that NV will deliver flagship Pascal in Q3/Q4 2016 only to launch a 50-70% faster flagship Volta in the consumer space in mid-2017.
Besides, NV almost always has a history of overpromising and underdelivering as far as timelines go. They said Fermi delay would not impact Kepler but it did. March 26, 2010 they launched GTX480, but it took until November 7, 2013 to launch a true successor in the form of the GTX780Ti. Therefore, logic dictates that most likely NV will launch GP104/204 this year (aka 980 successor) with GDDR5X and call it a flagship to milk the consumers, only later to release the real flagship in 2017. The other possibility is NV will launch a cut-down GP100 chip much in the same way they milked the market with the OG cut-down Titan.
I am really in a pickle to tell you the truth. I tried both. You are right for every point you made here. And to be honest, if you look at some of the games released recently, I think it was that game Project Cars – If you see, GTX 960 is actually beating a GTX 780. I just could not believe that. How the hell is this possible? You can look some other game benchmarks and you will be surprised.
I can only blame GameWorks for that and some of the other NV techs like PhysX. And the trend of games really goes like this – “Nvidia doesn’t loose out if it uses AMD’s solutions, but AMD does when they try to use Nvidia’s.” I am ok with that while i am an Nvidia user, but really, some of those moves are really anti competitive. However, when you turn on the GameWorks features…this is what you get – FRUSTRATION!
https://www.youtube.com/watch?v=esm8YzFBPOk
Sorry I know it’s ubisoft, but that’s not really the point here…
https://www.youtube.com/watch?v=ZcF36_qMd8M
Yes. This is what I mean…
I am not a NV nor AMD fanboy. I have a GTX 980. I had a HD7950. I switched cuz i wanted G-Sync…and so they came up with Free-Sync. So I paid that $250 premium… sheeesh! OK it is $250, not a $1000, but still a lot. fine…
You are making a good move there, don’t get me wrong. NV has some really good stuff to offer you! And I’m loving it. Just awesome! But keep in mind that they do sometimes things that are just … can’t describe it…
Anyway, enough of this. Get yourself a Pascal card, and enjoy, or just enjoy with some new AMD card. I know I just kinda went too far showing you those videos, but you should know the truth. I really wanna support AMD as well, but I need these frames coming, better visuals…gosh well, that’s what everyone wants no?
Big Pascal will be here before the end of the year with HBM 2 in tow. Mid-size chips may show up in the June timeframe.
Costly HBM1 wasn’t worth it with its measly 4GB and its bandwidth was wasted on 28nm GPUs but HBM2 is whole lot better, no hesitation here, we’re all in on Pascal.
There is no evidence to support your first sentence. You are posting pure conjectuve without any 3rd party source backing it. NV generally takes 2 years to refresh their flagship in the consumer space. This is because in the 1st half of a generation they can sell mid-range as a flagship (i.e., 30% faster than last gen’s flagship), and then release the true flagship towards the 2nd half of a generation. They have utilized this strategy since GTX680. They could release the Big Pascal but it will be a cut-down chip like 780 style to milk the market more.
It’s not conjecture. HPC clients get to see these chips and test them out a year in advance since there’s so much money on the table when it comes to accelerators. The benefits of working with the guy who runs my university’s supercomputer are particularly applicable for this.
The big die is ready for production. It’s only a matter of playing chicken and striking first.
You aren’t addressing any of the points I made with facts. Your earlier statement that flagship Volta is a mid-2017 GPU is flat out just something you made up. Whether HPC clients receive their required allocation of GP100 does not give us any confirmation that the consumer GP100 is slated to launch shortly thereafter. When NV tapped out Fermi, all initial allocations of the OG Titan went directly to Oak Ridge.
BitsandChips — http://www.bitsandchips.it/9-hardware/6621-rumor-scaletta-di-presentazione-delle-gpu-pascal
“Our source has informed us of the possible presentation of graphics cards lineup based on the various facets of the architecture of NVIDIA Pascal.
April 2016 – GTC: Tesla (GP100)
June 2016 – Computex: GTX 1080 (GP104) & GTX 1070 (GP104)
4Q 2016 – GP106 & GP107
4Q 2016 or 1Q 2017 – Titan (GP100)
1Q / 2Q 2017 – GP108 (GM108 rebrand ?)
It begins in April, with the Tesla board, based on GPU GP100, and dedicated to the HPC market. ‘S Oak Ridge National Laboratory and Lawrence Livermore National Laboratory have already announced that they will make extensive use. The first dedicated to the consumer market cards, will be those with GPU GP104, always produced with FF16 node ++ TSMC. They should be submitted to COMPUTEX in Taipei, but the market should start only from 3Q 2016. Later will be presented the lower-end cards, and the Titan range top.”
Next gen’s Titan is slated for Q4 2016 at best, most likely Q1 2017. Not a chance that a Volta flagship (Big die) is due Mid-2017.
Face it, you got called out on making up statements that you tried to pass off as facts. This isn’t WCCFtech.
No, the product launches once it’s shipped to customers. Otherwise it’s a paper launch. You merely play semantics. You’re not fooling anyone.
And Teslas traditionally show up AFTER mainstream users get the chips, binning and all that. It was the same with Kepler and Maxwell. They are facts, and no, it’s not, but you’re of equally low quality.
And Volta has to be in Oak Ridge by the end of Q1 2017 by contract, so I’m sorry but the facts are not on your side.
/
“NV released true flagships every 2 years or so.”
actually it is around 500 days between 1st gpu’s in the generations (*80 models)
WTF does “loose out” mean?
Loose out means get diminishing performance. When you try to use PhysX with AMD GPU, instead of using your GPU, it uses your CPU. You therefore get frame rate drops.
It was that way because there was no architecture developed from the ground up to take advantage of the new memory type, you can have the tranfers rates, but if the GPU itself can’t process the data due to architectural boundaries, of course it seems a waste…stay calm new architectures are coming soon to take advantage of that available juice in data tranfer
Do you mean “lose out”?