Not so easy with 16nm FF+ - could use planar but there is a fair amount of design difference between 28nm/20nm due to different lithography - unlike the past where a straight up optical shrink was relatively easy.
-
Competitor rules
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.
** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **
- Thread starter Boomstick777
- Start date
Not so easy with 16nm FF+ - could use planar but there is a fair amount of design difference between 28nm/20nm due to different lithography - unlike the past where a straight up optical shrink was relatively easy.
IMG seem to have done that. The first test silicon was 6 months ago on 28nm but as far as I am aware the mass volume products are due on 16nm and smaller. NIvida first silicon needs liquid cooling which might suggest they are running on a older node for first silicon. Granted I am only guessing here but if one GPU company has done it, then its reasonable for NVidia to do it. The first NVidia Pascal chips are not going to be the same as the ones we get in the desktop cards.
I'm afraid it is you who are speaking rubbish.
I never said they had test silicon for ages, I said they had test silcon, and probably in limited quantities which is why they didn't show it, as well as the fact they don't want to show the public the actual chip just like AMD had it hidden in a black box.
They had test silicon, they ran tests on it, they reported said test summary to the public and more detail results to select partners.
No one said they have.
No, it was just before CES.
And how do you know that?
Nope.
Its not been proven one way or the other. Big Pascal for compute was scheduled for H12016 so it makes since the original rumored tape-out was accurate, no one has disproven that. What it might have been was a tape-out of the Drive PX2 pascal units.
We'll find out soon enough.
What higher node? Intel 22nm? There is no TSMC, UMC or Samsung/GF FinFet process above 16/14.
The first silicon cannot have been before end of december / beginning of january on 16nmFF+. Vanilla (non +) isn't suitable for their GPUs.
Also, aside from FinFet, if we ignore that ... how exactly do you test a massively more dense chip design on a process that's massively less dense? LOL.
Last edited:
You are missing the word VOLUME from production. TSMC doesn't have to be in volume production to make test chips. What Samsung does is irrelevant.
The only important factor is TSMC is at volume production before Pascal is suspected of being released, that is the case, so it is irrelevant that they were delayed getting there.
What complete junk as always.
Volume production is absolutely meaningless in terms of taping out per-production engineering samples.
TSMC had tape-outs for products for most of their major customers by the end of the year so would be a bit odd if nVidia wasn't included.
Don't bother D.P, certain people are absolutely convinced that they are correct and there is no way that they can be wrong, no amount of discussion/argument will sway them to even the possibility of doubt. Best if we just wait and see what is said at GTC.
No I did not miss volume out of what I said. I meant what I said. It's not worth arguing anymore.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
AMD had a working card in a black box, they didn't want anyone to see the card/cooler. This is 99% likely down to not wanting people to get in a huff about a preproduction cooler, which could be anything down to a cpu cooler ziptied to the card.
Nvidia DID want to show Pascal.... you can tell because Jensen got up on stage and said "hey guys, here is Pascal, we're awesome, woo, LOOK AT THIS SILICON". Problem was, he lied and it wasn't Pascal. But very few people go "hey, lookey over here and this Pascal chip" without wanting to show Pascal.
It was a non working card without any cooling and even if it had dodgy cooling it's not a commercial card that end users would get huffy about the cooler looking bad or if the cooler looked big and non quiet... because it's supposed to go under the hood of a car not in a end users computer to be silent or sleek or whatever.
Not enough supply, again, you paraded that excuse out repeatedly and it's completely bogus. Chance of Nvidia having 100% yield on any chip they've ever made, none, there is no chance of that. If they had a single wafer of chips back, they would have a dozen or more dead chips perfect for parading around on stage that have zero other use.... because they are dead. Not showing a demo is one thing if you have 50 working chips and they need to be under full load testing, testing for reliability, testing with drivers, sent to partners.... but a bare board on a non working card to be held aloft. They'll have several dozen if not several hundred dead chips sitting around waiting to be made into keyrings for xmas presents or some other pointless use. Limited supply for such a show of a card is a completely invalid excuse. If they have silicon back they have spare silicon to stick on such a card.
Last edited:
How do you know that AMD showed 14nm for certain DM? Nobody got to see it, as it wasn't shown, so could well have been trickery going on inside the "closed case polaris system". Why are you so adamant that AMD are correct and Nvidia are telling lies? Are you perhaps letting bias cloud your judgement? For me personally I hope they do have working polaris chips, the same as I hope nVidia have working Pascal chips but I don't put too much stock into either having the GPU that I want near ready.
Get a clue before posting information as fact.
Quote from the Anandtech article on the Polaris demonstration by Ryan Smith (Anandtech employee who attended the demonstation):
Source: http://anandtech.com/show/9886/amd-reveals-polaris-gpu-architecture
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
lol, you've used almost that exact post before.
AMD's Polaris demo could physically not have been anything but 14nm, pretending it might not have been is simply embarrassing for you and everyone else who makes the argument.
As usual you're trying to make it like there is an opinion here and not fact. Nvidia DID show Maxwell on stage while claiming it was Pascal and AMD did show 14nm Polaris.
I'm not adamant of anything, the industry as a whole says they were shown Polaris, can see it's massive power advantage and they all watched a demo that couldn't be anything but a gpu on a smaller more power efficient process before being shown more in an NDA meeting with numerous journalists confirming they had seen it. The industry at large knows that Polaris is a 14nm and was shown. Trying to pretend it might not have been is nothing short of ridiculous. Respected tech website (wccfnonsense and co are not remotely respected) journalists have confirmed they've seen 14nm Polaris chips, as the post above you links to.
LIkewise there are plenty of pictures of what Nvidia showed on stage, a core that happened to have identical size, components, SMC on the package, package size, memory layout, pcb layout as existing GM204 MXM modules. CHances of making a literally identical chip on 16nm, 0%. Different voltage range for finfet, different power characteristics, there is 0% chance they would have the same SMC on the module.
Honestly, making that claim again after both the things you're arguing have been completely proven to be true(that Nvidia showed MAxwell while claiming it was Pascal and AMD showing 14nm). Arguing they aren't and that there can be opinion on the facts is just ridiculous... and you have the nerve to bring up bias after spending the past couple of days constantly attacking my posts with complete drivel.
Last edited:
One thing before reading the lines below, I do think it was a genuine Polaris chip that Raja held up and I do think that it was a genuine Polaris chip that was inside the closed system running the demo. No trickery involved.
But to say that it physically could not have been anything else is just laughable, with some time and a 3D printer I reckon I could make a GPU chip that would be indistinguishable at a few meters away, stick it on a board and you would never know.
The power meters showing the numbers, well they could be hooked up to anything, if the cables just went down behind the table it was all sitting on. Get a couple of normal cards to run the two demos and just tell everyone in the room what they want to hear.
Just to be clear again. I am not saying that is what did happen, I'm just indicating that it is physically possible, to fool a room full of people if you really wanted too, so stating that it couldn't possible be anything other than what they said it is just daft.
Your forgetting the fact that there were expert independent journalists and other industry experts present.
If you believe that they would be fooled by a 3D printed GPU, then you've probably one of those people who believe the Earth is flat and that the Moon landings were fake too, in other words there's no hope for you.