** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[enapace]

I don't follow the tech news for a while but from the info it look like we will need a new type of Motherboard for Nvidia NV-Link ? as NV-Link will be up to 5 to 12 times faster than traditional PCIE 3.0

That's not quite true unless you are building a IBM Server PC as they are the only CPUs that will have NVLink it's not a consumer thing yet.

 

[Mauller]

NvLink can work between gpu slots while the cpu communicates to them using PCIE. it is just a matter that NvLink capable PCIe slots will have a different number of pins and keying from what info has been released.

So with that it could be unlikely to come to the consumer space unless you use the special Nvlink versions of the GPU's and have a compatible mobo.

 

[drunkenmaster]

Automotive is very much 1st access/2nd Access customers as that’s how the market works.

Yes, you have priority customers and that is how all markets work, but you're still talking about samples, samples of pre-production cards before they are released are just that. You're acting as if only a few companies will get quantity of the samples and others will have to wait till Q4 to get samples. Nope, anyone that can offer Nvidia real sales come Q4 gets a sample because absolutely no car maker, tier 1/2 or below will plan a car release with a Nvidia product in it for 2017/2018 without having tested the product earlier.

They get samples, then tier 1 guys get their orders filled first when production is in full swing and tier 3 guy gets lower volume or has to wait a couple months. This is no different to EVGA getting priority over another company or, I've forgotten which company went under because Nvidia used to give them tier 1 priority then they decided to make some AMD cards, Nvidia put them into last access to their stuff category and basically starved them of product while causing end users to buy from the likes of EVGA instead.

Well you are wrong again. Some of the rumours did come about from the April Q2 news. It’s not hard to see how it happened. NVidia talk about Pascal in Q2, some people assume that mean start of Q2 so talk about Pascal arriving in April. Other see that talk about Pascal April and think that’s a reference to desktop Pascal and so the rumour news site start posting about Pascal in April. Not that all rumours followed that but some would have.

I literally provided proof of where these rumours came from and you're saying I'm wrong.

There were actually other stupid rumours though most of them additive, when Samsung stated production of HBM2 started then tech websites went with, cards using it must be coming soon. Because they are uninformed idiots who don't know what they are talking about.

However the Drive PX2 announcement specifically stated sampling, not release, they stated production or shipping(forget which already) in Q4, not Q2. It also doesn't have a GP100 on it, nor HBM2 memory involved in the mock up, nor performance indicated by GP100(8TF is currently achievable on one gpu and drive px2 will have 2 16nm gpus on it, they won't even be GP104, but 2 4tf gpus, Gp106 probably).

So GP106.... sampling in a few months, shipping end of year means GP100 is coming in April? yet the stories talking about GP100 for April all say it's about the shipping manifest and maybe throwing in a dumb statement about HBM2 going into production a few months before April to help out.

 

[spot35]

We should rename the thread nvidia pascal rumours thread. Maybe that would calm people down

I love the headline of nvidia planning a surprise pascal demo. It's not much of a surprise if it's 'outed' here, is it

I've started playing with overclocks to see how far I can push my Titans to keep up with the newer games. I feel I'm in for a long wait if I want the big pascal (i.e. 1080ti).

 

[Gregster]

I love a good rumor and love a good click bait article and seeing what is what come the end of them and the actual release date. Wccftech contradict themselves and post up so many different release dates/specs/prices, one of them gets it right

 

[bru]

So DM just to be absolutely clear, are you saying that they will not and cannot show us a Pascal based discrete GPU and tell us that it will be available to buy within about 6 weeks, at GTC. Yes or No?

 

[Simon Belmont]

So DM just to be absolutely clear, are you saying that they will not and cannot show us a Pascal based discrete GPU and tell us that it will be available to buy within about 6 weeks, at GTC. Yes or No?

Why ask him? He knows about the same as the rest of us NOTHING!

 

[Mauller]

So DM just to be absolutely clear, are you saying that they will not and cannot show us a Pascal based discrete GPU and tell us that it will be available to buy within about 6 weeks, at GTC. Yes or No?

He is saying that if they show Pascal then I wont be GP100, which makes sense with how new the node is. And as he said, 8Tflops is already capable, with Fiji, so for 10Tflop you only need two small GP104-6 cards. Etc. But it will more than likely not be available as soon as Polaris will be.

 

[Dicehunter]

Why ask him? He knows about the same as the rest of us NOTHING!

I forgot who it is but isn't there a chap who always says he "knows someone" and when asked about it says "I can't say anymore about it" ?

 

[Simon Belmont]

I forgot who it is but isn't there a chap who always says he "knows someone" and when asked about it says "I can't say anymore about it" ?

It was DP. He was saying that a lot during CES

 

[Dicehunter]

It was DP. He was saying that a lot during CES

Gotcha, Funny because I called him out on it around the time of Maxwell V2's launch and he said he would message me via trust with some proper details that he couldn't openly talk about etc....

Never materialized, Wonder why.........

 

[Simon Belmont]

Gotcha, Funny because I called him out on it around the time of Maxwell V2's launch and he said he would message me via trust with some proper details that he couldn't openly talk about etc....

Never materialized, Wonder why.........

Quite a few members here think they know everything. All they do is jibber jabber. Those people are on ignore. blah, blah, blah. Too annoying to deal with tbh

 

[Orangey]

Gotcha, Funny because I called him out on it around the time of Maxwell V2's launch and he said he would message me via trust with some proper details that he couldn't openly talk about etc....

Never materialized, Wonder why.........

His dad got the boot for breaking NDA and went back to work at Nintendo.

 

[Rroff]

Why ask him? He knows about the same as the rest of us NOTHING!

DM generally knows his stuff to be fair - though he seems to have got stuck in a loop of late :S i.e. before Fermi launch he was dead right on a number of aspects that I only managed to verify once stock was building up for launch.

 

[drunkenmaster]

So DM just to be absolutely clear, are you saying that they will not and cannot show us a Pascal based discrete GPU and tell us that it will be available to buy within about 6 weeks, at GTC. Yes or No?

Where did I say that or anything like it?

It's the same stuff all the time with you, Gregster. There is a rumour, the origins of THAT rumour can be dissected, you can sometimes see or discuss if a rumour is true.... this has NOTHING to do with anything else. Saying that a specific rumour about an April release is false has no connection to if said product releases in April, none at all. These are two entirely unconnected things, something you and Gregster seem to have trouble understanding.

For instance, I'm going to claim, because a magic duck told me, that GP100 is launching in April. Now, if Pascal launches in April... does that make my made up rumour correct? No it doesn't.

Disproving a rumour is just that, disproving that rumour, it proves nothing else. I've never claimed it does and in fact it's you/Gregster who keep coming up with these "so what you're saying is...." inferences which I haven't made.

I don't believe GP 100 has a shot in hell of coming out in April, it might, that rumour based off a shipping manifest that absolutely did not say what these idiot websites claimed it said will be incorrect regardless of when Pascal comes out.


Ignoring completely and utterly ignorant made up rumours, there really is absolutely nothing to indicate GP100 is coming out this year, nor big Polaris.

There is everything to indicate small/medium Polaris will launch before or by June. The general impression I get is Nvidia will talk about Pascal, maybe the architecture as opposed to a specific card much like AMD did at CES at some point before AMD launches medium Polaris, that could be GTC. In May/June timeframe I think they'll show some medium Pascal cards and talk about it, maybe paper launch it, with availability to come maybe August/Sept.

He is saying that if they show Pascal then I wont be GP100, which makes sense with how new the node is. And as he said, 8Tflops is already capable, with Fiji, so for 10Tflop you only need two small GP104-6 cards. Etc. But it will more than likely not be available as soon as Polaris will be.

Titan X is 7tf, Drive PX2 is claimed to be 8TF. I'd expect big Pascal to be around the 13-15tf range, GP104 to be about the 7-8tf range and so if the PX2 has 8TF total(from two discrete gpus and two tegra socs with a gpu in each) then each discrete gpu will be pulling around 3.5-4TF. So I'd expect GP106 to be the gpu used in it.

 

[drunkenmaster]

DM generally knows his stuff to be fair - though he seems to have got stuck in a loop of late :S i.e. before Fermi launch he was dead right on a number of aspects that I only managed to verify once stock was building up for launch.

A loop only because people here can't understand the difference between disproving one rumour and assuming I'm saying the opposite is true.

I say this specific rumour is BS, this rumour is that GP100 is coming out in April including a justification for their assumption. I say this rumour is completely unfounded, nothing more or less. They take that to mean I'm saying GP100 isn't coming out in April.

The basis for the April rumours is categorically false, that is it. I don't know for a fact nor claim it as fact that GP100 isn't coming out in April, but that rumour is not just the date, but the justification for the date... it IS false without question.


PS, I was also the only, well first one on here telling everyone we wouldn't have GPU's on 20nm because it offers nowhere near the power advantage and that finfet would bring us the overall node drop we require for new GPUs.

Next year could be interesting, there is the possibility we won't see a big Polaris at all. Early 2017 is encroaching into 10nm full production territory where a medium Polaris + 1 on 10nm makes more sense than a big Polaris on 16mn. TSMC seemingly being 3-6 months behind I can see a lot of ways it plays out. Maybe Nvidia actually launch big Pascal first as AMD wait an extra 3-6 months to do a medium 10nm chip instead. Maybe Nvidia scraps plans for big Pascal and waits a few months more than AMD and goes the same route, skipping big 16nm chips. We're in a very strange period. 28nm stuck around for so long as everyone had trouble(Intel also) with double patterning.

Foundry guys struggled to do double patterning and finfet on the same node(Intel rather tick tocked that with finfet on single patterning then only dealing with double patterning for 14nm). But 10nm is finfet and double patterning still... as such most of the biggest problems have been overcome and 10nm is coming extremely quickly compared to other nodes.

For 7nm we could see huge problems again, though kinda depends, we either have current tech plus triple patterning which will mean extremely long and expensive design/tape out stages, worse yields, more complex manufacturing. Or EUV matures and we get less masks, shorter design/tape out and better yields.

 

[Gregster]

Where did I say that or anything like it?

It's the same stuff all the time with you, Gregster.

I know you dislike me but you should at least check who you are quoting before giving me a hard time

 

[Ayahuasca]

The comma before hand suggests he's referring to you both, made evident at the end of the same paragraph.

 

[Rroff]

Next year could be interesting, there is the possibility we won't see a big Polaris at all. Early 2017 is encroaching into 10nm full production territory where a medium Polaris + 1 on 10nm makes more sense than a big Polaris on 16mn. TSMC seemingly being 3-6 months behind I can see a lot of ways it plays out. Maybe Nvidia actually launch big Pascal first as AMD wait an extra 3-6 months to do a medium 10nm chip instead. Maybe Nvidia scraps plans for big Pascal and waits a few months more than AMD and goes the same route, skipping big 16nm chips. We're in a very strange period. 28nm stuck around for so long as everyone had trouble(Intel also) with double patterning.

Its quite likely to be the case - for why look at Volta.

 

[Mauller]

.Next year could be interesting, there is the possibility we won't see a big Polaris at all. Early 2017 is encroaching into 10nm full production territory where a medium Polaris + 1 on 10nm makes more sense than a big Polaris on 16mn. TSMC seemingly being 3-6 months behind I can see a lot of ways it plays out. Maybe Nvidia actually launch big Pascal first as AMD wait an extra 3-6 months to do a medium 10nm chip instead. Maybe Nvidia scraps plans for big Pascal and waits a few months more than AMD and goes the same route, skipping big 16nm chips. We're in a very strange period. 28nm stuck around for so long as everyone had trouble(Intel also) with double patterning.

I don't see them using 10nm anytime soon for large monolithic dies. Late 2018 at the latest when the second 10nm processes drop will be more likely. Considering the first gen 10nm will be small SOC's for mobiles etc.