Without reading the last 50 pages or so of the pascal and polaris threads is there a consensus on when these chips are gonna launch (at least the full fat versions)?
Thanks.
Nope, everything is a guess and everyone has a different opinion
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Without reading the last 50 pages or so of the pascal and polaris threads is there a consensus on when these chips are gonna launch (at least the full fat versions)?
Thanks.
Without reading the last 50 pages or so of the pascal and polaris threads is there a consensus on when these chips are gonna launch (at least the full fat versions)?
Thanks.
I know, but any bit of news I see I put it up.
Definition: News
newly received or noteworthy information, especially about recent events.
information not previously known to
Still the same recirculated nonsense though, such poor reporting when they cite a site that cited a different site rather than going back to the original source.
What about this is noteworthy information? Nothing in it is even semi accurate, everything in it is incorrect, nothing in it is previously unknown information either
It is the biggest pile of rubbish, it isn't news.
It's not news, it literally gets everything wrong, has a pop at AMD and makes up that Nvidia is going to be ahead. Putting up every link to everything that ever uses the word Nvidia is worthless, to the thread, for the forum, to everybody.
First off.... christiantoday... really?
Second it says HBM2 will be at least three times as fast as HBM1, nope, it's twice as fast, no more, no less, it also says it will have twice the memory capacity when in fact it will have up to 8 times the capacity(1GB per stack, to a potential 8GB per stack).
Then it goes on to say how this will be bad for AMD because they only have HBM1 and act like this will continue into the future, implying Nvidia will have HBM2 while AMD are stuck on HBM1, completely wrong.
It then says the currently available AMD Fiji is a dual gpu design.... of 4GB each, for 8GB per GPU, while Nvidia on the other hand can provide 32GB... literally nothing in that paragraph is accurate. Fiji is a single gpu that only provides 4GB per GPU and has an currently unavailable dual gpu card that is still 4GB per GPU, Nvidia currently has no HBM2 cards out and so can't provide 32GB per card.
What about this is noteworthy information? Nothing in it is even slightly accurate, everything in it is incorrect, nothing in it is previously unknown information either
It is the biggest pile of rubbish, it isn't news.
It's not even the same recirculated nonsense, it's taken existing information... repeated it entirely incorrectly and added nothing remotely new except an incredibly warped idea of how badly behind AMD will be because AMD's last gen cards would have the same memory type as Nvidia's new gen.
What about this is noteworthy information? Nothing in it is even semi accurate, everything in it is incorrect, nothing in it is previously unknown information either
It is the biggest pile of rubbish, it isn't news.
For Polaris AMD stated they it will come before the back to school season, for Pascal there is no info yet outside of rumors.
But only low end Polaris, so that is good for AMD and will help them gain market share and contracts, building faith with the likes of Dell etc, it is not really relevant to anyone here.
AMD claims their small Polaris has twice the efficiency of their old architecture and claimed 70% of that is due to the new process with a 30% improvement from architectural efficiency improvements. Seems decent but we have no idea about large Polaris release dates.
Nvidia are shipping Drive PX2 units to partners in April, these are mid-sized Pascal chips.Volume production is not until Q4 but that is because these aren't needed before 2018 model year cars so volume production is not needed yet and nvidia are prioritizing their TSMC production for gaming and compute parts.
Nvidia were originally schedule to release a large pascal chip for compute purposes around Q2 this year. there has been absolutely zero news on whether this is delayed or on time.
there is no news at all about small Pascal or a mid-sized pascal for gaming. that doesn't tell you much because Nvidia rarely leak much information, it only damages current sales. AMD has veyr low sellthrough for low end chips so they did some marketing on their upcoming Polaris low-end to try to stear contracts away form nviida.
If they go to partners in April, then that's probably the first good test silicon they'll have.
There's no information because NVIDIA categorically have not had (or still do not have) any silicon. It isn't just Charlie that has been hinting that they're miles behind on this node.
First Polaris chips will indeed be the laptop / low power chip that was fully demonstrated at CES. I expect them to actually be in products by May - Apple products.
The desktop version and slightly larger chip (but still smallish and not HBM) will no doubt be ready to go into desktop products, but I think they'll wait to reveal them until E3, for fear of cannibalising sales of existing inventory. I think the bigger HBM chip(s, though probably no bigger than 350-375mm) will be debuted at E3, but volume may be some time later due to HBM2 supply, or wanting to diminish Fiji stocks.
I wouldn't take silence from anyone as a sign of delay. Keeping the cards close to the chest is a good strategy ahead of a node shrink, and a good way to avoid cannibalisation of existing stock.
If a summer date for Polaris or Pascal was announced, who in their right mind would upgrade to a 28nm GPU between now and then?
They already have test silicon, the partners are getting pre-production units, which is final silicon but in a device designed for easier testing and integration.
Nvidia do have silicon and have tested it and have demonstrated to people under NDA, Charlie just makes up rubbish as usual (he says tape-out for big pascal happened in November without providing any evidence what so ever, just as guilty as the other websites he lambasts, his distorted opinion isn't worth a dime). There are real benchmark figures available for the drive PX units.
Historically nVidia has never released much information before a launch. In fat it would probably be a bad sign if we did start getting solid rumors because it probably points to there being a delay or initial problems so they are trying to divert atention form AMD. If everything is on track Nvidia will keep quiet and mak a big launch event, maybe at GTC. If things are really bad we'll get a paper launch at some point.
Pascal likely taped out in early November and wasn’t back from the fabs when Jen-Hsun gave his CES speech, at least that is what the data shows.
Assuming NVidia are anything at all like the other major GPU company’s then perhaps not as sometimes companies show the silicon behind closed doors for various reasons. A few times I have seen GPU silicon months before it’s publicly shown at shows like CES. Other times there have been dev units sent out a very long time before public shows.“Absolute rubbish. You know as well as everyone else does that if they had had test silicon for ages as you claim (and claimed at CES) that they would have shown test silicon, not Maxwell and lied about it being Pascal.”
Why not, if it was test silicon that timeline fits. IMG new GPU was show publicly for the first time at the same time as Pascal. Before anyone says this is off topic IMG first test silicon tapped out last summer in June 2015. Was privately shown behind closed doors and boards sent out to devs without any leaks until December then demoed at CES at the same time as Pascal. Based on that was is unreasonable about Pascal first test silicon taping out last summer? If Pascal tapped out a month ish after IMG new GPU it might have just missed CES by a matter of weeks but still have tapped out last summer.“You've also repeatedly claimed that they taped out last summer, as per the dubious (putting it nicely) articles. No-one thinks this is credible anymore ... most with any idea thought it was total bowlocks back then. “
Assuming NVidia are anything at all like the other major GPU company’s then perhaps not as sometimes companies show the silicon behind closed doors for various reasons. A few times I have seen GPU silicon months before it’s publicly shown at shows like CES. Other times there have been dev units sent out a very long time before public shows.
I can say with 100% confidence then some GPU companies have silicon for around 6 months before they publicly show anything.
Why not, if it was test silicon that timeline fits. IMG new GPU was show publicly for the first time at the same time as Pascal. Before anyone says this is off topic IMG first test silicon tapped out last summer in June 2015. Was privately shown behind closed doors and boards sent out to devs without any leaks until December then demoed at CES at the same time as Pascal. Based on that was is unreasonable about Pascal first test silicon taping out last summer? If Pascal tapped out a month ish after IMG new GPU it might have just missed CES by a matter of weeks but still have tapped out last summer.
DM it isn’t a different situation in this case. What you said is correct for Arm style IP only products but in this case IMG produced a PC GPU following a NVidia style route designing a board, putting it in silicon wrote drivers e.c.t the timelines should be very similar to a NVidia GPU. The reason I bring IMG up is they had working silicon for around 6 months before it was publicly shown and even then they almost waited a little longer. Knowing the reasons why they did that I can fully believe NVidia have working silicon and chose not to show it.Very different situation, you're talking mostly about a company that sells it's IP. Arm show say a A72 1-2 years before they end up in a shipping product because that is how it works. Nvidia use their own IP to make chips to sell where as primarily PowerVR business is making IP to sell for other companies to incorporate into chips which they then sell.
Arm type IP has been used in the test process of a new process for, 3-4 processes, and we're talking about LONG before production of the process is ready to go into full scale production, but test lab runs of wafers as they prepare a new process before it even gets into the final fabs it will be used in.
Again though, there is very little reason to hold up something, claim it's the new product but it's actually not. Lying has very little benefit and only generally one good reason to do so... you didn't have real parts back to stick on the device.
It's entirely fine to hold up a dev kit and say this is Maxwell but basically same form factor and half the performance we expect from the Pascal version due later, everyone gets that. Holding it up and saying it's Pascal, the only reason to do so is to make people believe you have silicon when you don't. If you wanted to show Pascal silicon, you would if you had it. If you didn't want to show Pascal and you do have silicon for it, you just state you want to keep secrets and show up a Maxwell version, completely understandable and fine. Holding up Maxwell and claiming it's Pascal repeatedly, as before they have done this once before and it's when they were absolutely miles behind where they wanted everyone to think they were.
Yep and I also know the first test silicon isn’t always on the same node as the mass volume products. The first silicon can be on a higher node at a slower speed to work out bugs and start first work on drivers.Aside from NVIDIA's supposed problems, you do realise that TSMC 16nmFF+ initial production probably only began in the very last days of November, right? Samsung (and perhaps GF) will have been doing test chips on LPP since Q1 or Q2 '15.