AMD Greenland Vega 10 said to have 4096 stream processors

[kalniel]

But whether they did or didn't wave around a real Pascal GPU on stage, or whether you're shown an orange instead of an apple, none of that changes what the company, rumors, or speculations indicate, it hasn't changed anything, sure there could be many reason why they done it but those reasons would be worse than guesswork.

You're absolutely right - it doesn't change what the company is making. But it does change the trust in what the company says about what it's making. Whether that's 4GB VRAM or specific features and performance levels. It reduces one's belief in what comes out of a company to only that which can be tested infront of one's own eyes.

[Corky34]

It reduces one's belief in what comes out of a company to only that which can be tested infront of one's own eyes.

Call me a skeptic then as i don't trust anything that a company says until i see it with my own eyes.

[scaryjim]

... But that lie hasn't changed the rumored May release date has it.

No, it hasn't changed the rumour. It has changed the credibility of that rumour. If nvidia had shown working Pascal silicon at CES, the May launch date would have been credible. The fact that nvidia didn't show any Pascal silicon at CES makes the May release date questionable. The fact that they lied about having Pascal silicon suggests they have something to hide, which tanks the credibility even further.

Of course, none of that means the rumour is wrong, but the weight of evidence is very much on the side of nvidia not having a retail card on sale by the end of May.

[DanceswithUnix]

Well the rumors seem to say both will hit the market with HBM2 cards around the same time.

True, but then the rumours also say that as AMD had a hand in developing HBM and Nvidia are just a customer, AMD get preferential supply. Now, if HBM2 is plentiful, then that doesn't matter. If HBM2 is in tight supply, then that means AMD makes as much as they want and Nvidia goes on allocation.

But, as HBM2 only currently makes sense for really high end stuff and there aren't that many people who can afford a £400 graphics card (Fury Nano is currently about £380) then I don't see HBM2 supply being much of an issue.

Perhaps HBM2 could help make a single width half heigh R9-380 class card for ITX machines, in which case there could be a healthy market for it. For most people, cost is important and rumour says DDR5 is still cheaper and quite fast enough for the mass market.

Edit to add: There is a chance that Vega will work fine with HBM1 memory, and even Polaris.

[scaryjim]

... Well based on the link i posted on the other page it probably is going to be a paper launch as in the article they say...

Can't see AIB partners doing much in the way of shipping products in the space of 4 days.

To be fair, I hadn't read the original article you linked to, as I was more interested in your curious response to Pascal-gate, so I hadn't realised that when you said Pascal was going to "show its face" in May you meant it was going to be a product announcement, rather than an actual product launch. Seems like you agree that ...

... the weight of evidence is very much on the side of nvidia not having a retail card on sale by the end of May.

Although now I have read the article it appears I need to buy another box of Maldon, as the concept of launching a GDDR5X card only two months after GDDR5X began sampling is a whole new level of technological oneupmanship. Micron reckon it won't even enter mass production until the summer. I can't see a range of partners having AIBs ready with GDDR5X by the end of May. I wonder if we'll see a high clock rate GDDR5 implementation first, followed by a GDDR5X refresh a few months later...? Be an odd way to go about it, but otherwise I doubt we'll see cards in the shops before September...

EDIT: although last time I researched mass-manufacturing to product launch it apparently happened within a month, so perhaps I need to take my own comments with a pinch of salt

EDIT2: Also worth noting from that link (to my own comment ) that Samsung are already mass-producing HBM2, which means that HBM2 cards this summer are also not impossible. A little food for thought there....

[kalniel]

Call me a skeptic then as i don't trust anything that a company says until i see it with my own eyes.

But you seem to, when you say for example:

Nvidia have already confirmed Pascal (their next GPU) is going to using HBM2

or

When i bought my copy of Windows 7 Microsoft promised me they would provide security related updates until January 2020

There isn't much scepticism in the word 'confirmed' and 'promised' But I applaud your sceptical approach as shown with regards other products and company releases - it's definitely a good approach to take regarding unreleased products in particular.

I measure my scepticism rather than outpour it - in fact I'm generally quite the optimist until I have reason to distrust the source. That means I'm quite often wrong, but that's no problem to admit to either

[Corky34]

No, it hasn't changed the rumour. It has changed the credibility of that rumour. If nvidia had shown working Pascal silicon at CES, the May launch date would have been credible. The fact that nvidia didn't show any Pascal silicon at CES makes the May release date questionable. The fact that they lied about having Pascal silicon suggests they have something to hide, which tanks the credibility even further.

Of course, none of that means the rumour is wrong, but the weight of evidence is very much on the side of nvidia not having a retail card on sale by the end of May.

Firstly, for me personally, that would depend on who the source of the rumor was/is and knowing why they didn't show any Pascal silicon at CES, I'm to lazy to go searching for who the source was for the rumored May "launch" and we'll probably never know the precise reason for the using a fake/mockup at CES and without that information (IMO) the rumored May "launch" is no more or less credible than any other rumor.

True, but then the rumours also say that as AMD had a hand in developing HBM and Nvidia are just a customer, AMD get preferential supply. Now, if HBM2 is plentiful, then that doesn't matter. If HBM2 is in tight supply, then that means AMD makes as much as they want and Nvidia goes on allocation.

Without knowing whose GPU is going to be faster though it's almost impossible to tell how that preferential supply agreement is going to effect anything, it would be foolhardy for AMD to make more cards than they could ship just to restrict supply to Nvidia, and even if AMD cards turn out to be better they're fighting a strong headwind of public perception.

EDIT2: Also worth noting from that link (to my own comment ) that Samsung are already mass-producing HBM2, which means that HBM2 cards this summer are also not impossible. A little food for thought there....

It was one of the reasons i thought (incorrectly) that we would see AMD first to market with a HBM2 card, they had/have a preferential supply agreement so with the normal low supply of new silicon like RAM i assumed that AMD would be using everything Samsung could produce while they ramped up production.

But you seem to, when you say for example:

When i said "Nvidia have already confirmed Pascal (their next GPU) is going to using HBM2" I was fairly certain (wrongly it later turned out) that Nvidia had *announced that they were, i maybe skeptical but if a company makes an official announcement then i would consider that to be a fact, not unempirical knowledge, opinions, or beliefs, that being what skepticism is, a questioning attitude towards unempirical knowledge, opinions, or beliefs.

There isn't much scepticism in the word 'confirmed' and 'promised' But I applaud your sceptical approach as shown with regards other products and company releases - it's definitely a good approach to take regarding unreleased products in particular.

I measure my scepticism rather than outpour it - in fact I'm generally quite the optimist until I have reason to distrust the source. That means I'm quite often wrong, but that's no problem to admit to either

It's even more so with the Windows 7 quote as Microsoft put that in writing in the EULA, it was a stated legally binding (afaik) fact, it's something that's directly attributable to Microsoft, an official company statement if you will.

*I think i may have assumed that when i read that article and similar others that "we know that Nvidia will pair up its Pascal GPU with HBM2 (called 3D memory in its roadmap slide)" that meant it would be when Pascal first hit the market, and not as it now seems some 6+ months afterwards.

[DanceswithUnix]

Without knowing whose GPU is going to be faster though it's almost impossible to tell how that preferential supply agreement is going to effect anything, it would be foolhardy for AMD to make more cards than they could ship just to restrict supply to Nvidia, and even if AMD cards turn out to be better they're fighting a strong headwind of public perception.

Depends how bad the restrictions are. If there is more demand than supply, then whoever gets the chips gets revenue. OFC, it could be that Apple tie up all the 16nm production making silicon for phones and no-one can sell anything else.

AMD can use that memory on their upcoming APUs as well.

When i said "Nvidia have already confirmed Pascal (their next GPU) is going to using HBM2" I was fairly certain (wrongly it later turned out) that Nvidia had *announced that they were, i maybe skeptical but if a company makes an official announcement then i would consider that to be a fact, not unempirical knowledge, opinions, or beliefs, that being what skepticism is, a questioning attitude towards unempirical knowledge, opinions, or beliefs.

If just one of the Pascal range use HBM2, then "Pascal is going to use HBM2" is a true statement. It would be spin, and possibly deceptive, but it would be true.

[lumireleon]

let us just accept GDDR5 and GDDR5X will soon disappear, SAMSUNG and the rest are heavily investing in next gen memory that's why all of a sudden DDR4 is becoming standard. HBM2 and the next generation of memory will mean thinner and lesser complex motherboards for laptops as the memory is part of the GPU. Intel for the past few years has been fighting back with better IGPU's with the intro of iris pro and soon HBM will come inbuilt in some high end CPU's a strategy to eat more of AMD market share.