GTX1060 about to be out?

SEANMCAD

http://hexus.net/tech/news/graphics/94127-leaked-nvidia-geforce-gtx-1060-presentation-slides-published/

In July?

Quizzical

Out in the sense that the GTX 1080 is out today, or in the sense that the GTX 980 is out today?

I'd give right around zero credibility to claims that it's as fast as a GTX 980 outside of weird corner cases.  The architecture is nearly a straight die shrink of Maxwell, but if the slides are accurate, 87% of the compute performance, 86% of the memory bandwidth, and 70% of the performance for the fixed-function graphics stuff that is part of the GPCs.

Malabooga

Nah, it will follow same path as 1070/1080, it will be clocked high on stock with almost no overclocked headroom and it will perform close to 980, but when you crunch the numbers (if those spec numbers are actually true) and put all that OC potential Maxwell has (and Pascal doesnt) it comes out as GTX970 performance OC-OC and that fits very well with 1070 comparison where 970 is 30-40% slower than 1070.

1280/1920 -> 1060 is 0,66 of 1070 or 1070 has 1,5x cuda cores.

13lake

There's gonna be plenty of them in stock, it has 48 ROPs, so that means that it's gonna fused off 1080/1070 that failed binning.

 And Because this a new process node, with high margin of out of spec chips on the wafer, and considering they've been holding off 1070s just in case, they're gonna have plenty of binned chips, and plenty of proper full working 1070s to convert to 1060 if the need arises due to for example amd surprising them with a 485 or something, or flooding the market with heavy out of the box oc-ed custom 480s.

You guys remember how 980 Ti launched at the same price point as Fury X, to ruin the Fury X sales, as in the fury X with the expensive new HBM and water-cooling which needed higher than $650 for AMD

Nvidia launched  the 980 Ti at the lower price point to stunt the sales of AMD, they can afford to do it again and again and again.

Malabooga

Doubt it. Fiji was very complicated chip and "prototype consumer HBM card". NVidia designed 1060 to conform with their philosophy. But AMD doesnt care much about NVidias philosophy.

And if NVidia starts to sell half of GP104 as "1060" it will be final proof that yields on TSMC 16nm performance node are terrible as 1070/1080 supply is still very low.

13lake

Your first paragraph is pure nonsense, please explain how it has any connection to anything in my post above ?

As for the second paragraph, you've just confirmed that laws of physics indeed still cannot be broken ,...
As in the first products on a new node, which is so small that it has more issues than nodes before (due to macrophysical and quantum constraints) is not producing a lot of proper chips ?

Final proof as in proof that you do not understand how binning works ?

Next you're gonna tell us, AMD is immune to these problems ?

ah screw it ok here's the logic :

Not enough GDDR5X chips + lots of moderately broken 1080 chips > more 1070s > not enough 1070 chips that can meet the 1070 requirements > more 1060s

So the less 1080s and 1070s we see in the shops, and the worse their supply is, the more 1060s Nvidia can make, ... it's that simple.

And on top of that nvidia can choose to gimp perfectly good 1080s and 1070s to make more 1060s. 

Quizzical

A part that needs to disable half of the shaders on the chip either is going to be very low volume or is indicative of terrible yields. It would make more sense for it to be a different die than a salvage part of GP104.

13lake

Quizzical said:

A part that needs to disable half of the shaders on the chip either is going to be very low volume or is indicative of terrible yields. It would make more sense for it to be a different die than a salvage part of GP104.

More sense ? wow, as in it would make more sense to throw away or recycle all the half disabled chips ?
Whether it would make more sense is your personal opinion.
Considering Nvidia has done the opposite for parts of 600 series, the whole 700 series, and half of the 900 series, that makes it an opinion they laughed at and ignored completely, ...

And considering that no matter how many mistakes they do, nor how much opposite to what you consider the correct path they go against, they suffer no consequences, i'm inclined to believe that they will continue to do it.

Loke666

Quizzical said:

Out in the sense that the GTX 1080 is out today, or in the sense that the GTX 980 is out today?

I'd give right around zero credibility to claims that it's as fast as a GTX 980 outside of weird corner cases.  The architecture is nearly a straight die shrink of Maxwell, but if the slides are accurate, 87% of the compute performance, 86% of the memory bandwidth, and 70% of the performance for the fixed-function graphics stuff that is part of the GPCs.

As fast as the 970 I could believe but not the 980. Generally speaking is usually a card of a new generation one step better then one of the last generation. That is simplifying things a lot but it usually is pretty close to reality anyways.

Quizzical

When is the last time Nvidia or AMD released a part that disabled at least half of the shaders of a chip with more than four compute units?  The only one that comes to mind is that AMD had a weird, low volume FirePro part that disabled 2/3 of the shaders of Cayman while still having the full memory bus.  Even Fermi, which had demonstrably awful yields, didn't have to go that far, as the bottom desktop GF100 part had 11/16 compute units enabled, and the bottom GF104 part had 6/8.

Yes, salvage parts are common.  That's why the GTX 1070 exists.  But you don't have very many big chips where you need to disable more than 1/4 of the compute units unless something has gone horribly awry.  And you do expect to throw away some fraction of the chips that come back from the fab because it's cheaper to toss out the handful where the PCI Express controller or video decode block or whatever was defective than to bloat the die by having two of everything.

Quizzical

Loke666 said:

Quizzical said:

Out in the sense that the GTX 1080 is out today, or in the sense that the GTX 980 is out today?

I'd give right around zero credibility to claims that it's as fast as a GTX 980 outside of weird corner cases.  The architecture is nearly a straight die shrink of Maxwell, but if the slides are accurate, 87% of the compute performance, 86% of the memory bandwidth, and 70% of the performance for the fixed-function graphics stuff that is part of the GPCs.

As fast as the 970 I could believe but not the 980. Generally speaking is usually a card of a new generation one step better then one of the last generation. That is simplifying things a lot but it usually is pretty close to reality anyways.

I could certainly believe that a GTX 1060 with the claimed specs is competitive with a GTX 970 and, for that matter, a Radeon RX 480.  That's about what you'd expect from the specs.

Malabooga

13lake said:

Your first paragraph is pure nonsense, please explain how it has any connection to anything in my post above ?

As for the second paragraph, you've just confirmed that laws of physics indeed still cannot be broken ,...
As in the first products on a new node, which is so small that it has more issues than nodes before (due to macrophysical and quantum constraints) is not producing a lot of proper chips ?

Final proof as in proof that you do not understand how binning works ?

Next you're gonna tell us, AMD is immune to these problems ?

ah screw it ok here's the logic :

Not enough GDDR5X chips + lots of moderately broken 1080 chips > more 1070s > not enough 1070 chips that can meet the 1070 requirements > more 1060s

So the less 1080s and 1070s we see in the shops, and the worse their supply is, the more 1060s Nvidia can make, ... it's that simple.

And on top of that nvidia can choose to gimp perfectly good 1080s and 1070s to make more 1060s. 

It has much in connection, seem you have very low understanding of how things work, NVidia inteded to lauch 980ti. And binning has nothing to do with this, half of the chip is just not done unless you have big problems with yields

You seen to not understand how economics works and how chip production works, sorry, but your post is just nonsense.

SEANMCAD

Quizzical said:

Out in the sense that the GTX 1080 is out today, or in the sense that the GTX 980 is out today?

I'd give right around zero credibility to claims that it's as fast as a GTX 980 outside of weird corner cases.  The architecture is nearly a straight die shrink of Maxwell, but if the slides are accurate, 87% of the compute performance, 86% of the memory bandwidth, and 70% of the performance for the fixed-function graphics stuff that is part of the GPCs.

no in the sense that I can walk into a best buy in a small town on a random day and walk out with a 1070 in my hand that I pulled from the shelf kind of 'out'

ianicus

Any comparisons to aMD are moot. Cards are useless without decent display drivers :pleased:

13lake

Malabooga said:

It has much in connection, seem you have very low understanding of how things work, NVidia inteded to lauch 980ti. And binning has nothing to do with this, half of the chip is just not done unless you have big problems with yields

You seen to not understand how economics works and how chip production works, sorry, but your post is just nonsense.

Nope.

Nvidia intended to launch 980 Ti, exactly after Fury X launch date got announced, they were waiting to counter. they followed tic for tac with the product announcement and then price announcement to ensure that AMD doesn't recover any market share.

Nvidia hoped to price 980 Ti much higher, AMD also hoped to price Fury X higher, however due to the game of cat and mouse, they both went for less starting msrp, the only difference being that Nvidia could afford to sell at loss or lower than wanted because of immense lead in "everything" over AMD, while AMD couldn't afford as in it hurt AMD more than it did Nvidia.

Binning has everything to do with this, half of the chip is being done, exactly because A yields are a huge problem, B the quantum effects + leakage is getting to ridiculous levels.

It is you who seems to not understand what Nvidia can and cannot afford to do, and is willing or not willing to do economics wise, and what they calculate is acceptable sacrifice with regards to chip production to hold onto the market share.

13lake

ianicus said:

Cards are useless without decent display drivers :pleased:

Exactly, AMD has gained a clear lead with driver quality and driver software fidelity over nvidia in past ~2 years.

SEANMCAD said:

no in the sense that I can walk into a best buy in a small town on a random day and walk out with a 1070 in my hand that I pulled from the shelf kind of 'out'

Snapchat or lies !

Malabooga

13lake said:

Malabooga said:

It has much in connection, seem you have very low understanding of how things work, NVidia inteded to lauch 980ti. And binning has nothing to do with this, half of the chip is just not done unless you have big problems with yields

You seen to not understand how economics works and how chip production works, sorry, but your post is just nonsense.

Nope.

Nvidia intended to launch 980 Ti, exactly after Fury X launch date got announced, they were waiting to counter. they followed tic for tac with the product announcement and then price announcement to ensure that AMD doesn't recover any market share.

Nvidia hoped to price 980 Ti much higher, AMD also hoped to price Fury X higher, however due to the game of cat and mouse, they both went for less starting msrp, the only difference being that Nvidia could afford to sell at loss or lower than wanted because of immense lead in "everything" over AMD, while AMD couldn't afford as in it hurt AMD more than it did Nvidia.

Binning has everything to do with this, half of the chip is being done, exactly because A yields are a huge problem, B the quantum effects + leakage is getting to ridiculous levels.

It is you who seems to not understand what Nvidia can and cannot afford to do, and is willing or not willing to do economics wise, and what they calculate is acceptable sacrifice with regards to chip production to hold onto the market share.

NVidia cannot afford to sell cards below the price due to their monopolistic status as they will be fined for violating anti-trust laws same as Intel.

980ti moght have been priced a bit highr but not by much (half way between 550$ 980 and 1000$ Titan) so 759$. The problem NVidia has now is that they havent cleared the channels of 960/970/980 and they will be competing with their own cards, and tere are piles of Maxwell cards still out there.

You think that NVidia can "afford" anything, but thats not true, if their margins drop (and most of revenue for NVidia comes from "gaming") it will have impacts. Thats bad business.

13lake

Malabooga said:

NVidia cannot afford to sell cards below the price due to their monopolistic status as they will be fined for violating anti-trust laws same as Intel.

Nope. pure speculation on your part. giv proof.

Malabooga said:

The problem NVidia has now is that they havent cleared the channels of 960/970/980 and they will be competing with their own cards, and tere are piles of Maxwell cards still out there.

Nope. completely wrong. stock/channel clearance is heavily under way, 960/970/980 are selling for dirt cheap atm, the quite small self-compete time frame is not going to impact nvidia in the slightest.

980 is selling for $315/330 euros, just above yesterdays 970 price. How long do you think that channel will take to clear, haha

970 is selling for ~$235/240 euros haha, 960 selling for $155 muhahaha

Malabooga said:

You think that NVidia can "afford" anything, but thats not true, if their margins drop (and most of revenue for NVidia comes from "gaming") it will have impacts. Thats bad business.

Nope. completely wrong. What Nvidia can afford is only second to Apple. the fanfare around Nvidia is so huge,  they literally need dozens of multiple consecutive high profile fails before it starts to affect them in any way. 

Compared to how little AMD can afford to do wrong, Nvidia is light-years away from any risk.

Ridelynn

Last rumor I saw said 1060 would be GP106 die - which isn't the same as the GP104 currently used in the 1080/1070.

~If~ that is true - there is no cut down from a 1080/1070 to 1060. 1060 would be the top bin of that chip, and cut downs on it would roll to 1050/etc.

~If~ it's not true, and 1060 is another GP104 salvage part - nVidia absolutely will not cut down perfectly good 1080/1070 chips to make more 1060s. They will sell those as the highest end part they can. What they will do is just crank up the price on the 1060 to meet supply/demand. Yes, that will absolutely make no sense on Price/performance scale. But people who are Team Green won't care and will buy it anyway because the 1070/1080s aren't in their budget to do so. This has happened plenty before.

I expect the 1060 to be GP106, and if there is an additional salvage part from GP104, that it will be 1060Ti  type SKU. But that is just me guessing.

SEANMCAD

SEANMCAD said:

no in the sense that I can walk into a best buy in a small town on a random day and walk out with a 1070 in my hand that I pulled from the shelf kind of 'out'

Snapchat or lies !

 the idea behind what I said is not that its released because me and only me can do what is described but rather that its a repeatable pattern.

In other words, go right now get in your car and go to best buy

13lake

SEANMCAD said:

 the idea behind what I said is not that its released because me and only me can do what is described but rather that its a repeatable pattern.

In other words, go right now get in your car and go to best buy

Paper launch is paper launch, we're still waiting on that wide availability.

May i ask why have u switched to inferior 1070 all of the sudden ?

It wouldn't have anything to do with you checking nowinstock, seeing if a card is in stock somewhere for more than a few hours, and then pretending like you haven't done that, and coming up with the elaborate drive to best buy story.

I mean it must be pure coincidence that you are using Best Buy as an example, and that there's been 2 cases of a 1080 surviving as in stock on nowinstock for more 24 hours.

And now a 1070 staying green on nowinstock for more than 24 hours ?

Did you actually call any Best Buy, or ever had any intent to drive to one to check for availability ?

Quizzical

13lake said:

Malabooga said:

It has much in connection, seem you have very low understanding of how things work, NVidia inteded to lauch 980ti. And binning has nothing to do with this, half of the chip is just not done unless you have big problems with yields

You seen to not understand how economics works and how chip production works, sorry, but your post is just nonsense.

Nope.

Nvidia intended to launch 980 Ti, exactly after Fury X launch date got announced, they were waiting to counter. they followed tic for tac with the product announcement and then price announcement to ensure that AMD doesn't recover any market share.

Nvidia hoped to price 980 Ti much higher, AMD also hoped to price Fury X higher, however due to the game of cat and mouse, they both went for less starting msrp, the only difference being that Nvidia could afford to sell at loss or lower than wanted because of immense lead in "everything" over AMD, while AMD couldn't afford as in it hurt AMD more than it did Nvidia.

Binning has everything to do with this, half of the chip is being done, exactly because A yields are a huge problem, B the quantum effects + leakage is getting to ridiculous levels.

It is you who seems to not understand what Nvidia can and cannot afford to do, and is willing or not willing to do economics wise, and what they calculate is acceptable sacrifice with regards to chip production to hold onto the market share.

Let's consider their recent history.  With Kepler, Nvidia had four chips, excluding the oddball, double registers one in the Tesla K80.  The desktop cards based on those chips that they tried to sell to the general public are, by chip, give below.

GK110 (15):  GeForce GTX 780 Ti (15), GeForce GTX Titan Black (15), GeForce GTX Titan (14), GeForce GTX 780 (12)

GK104 (8):  GeForce GTX 770 (8), GeForce GTX 680 (8), GeForce GTX 670 (7), GeForce GTX 660 Ti (7), GeForce GTX 760 (6)

GK106 (5):  GeForce GTX 660 (5), GeForce GTX 650 Ti (4), GeForce GTX 650 Ti Boost (4)

GK107 (2):  GeForce GT 640 (2), a big mess of cards that sometimes had 1 compute unit enabled and sometimes 2

So let's look at Maxwell.  They had four chips based on the architecture:

GM200 (24):  GeForce GTX Titan X (24), GeForce GTX 980 Ti (22)

GM204 (16):  GeForce GTX 980 (16), GeForce GTX 970 (13)

GM206 (8):  GeForce GTX 960 (8), GeForce GTX 950 (6)

GM207 (5):  GeForce GTX 750 Ti (5), GeForce GTX 750 (4)

Yes, there are a lot of salvage parts there.  But nothing near half of the chip disabled, except for GK107, where if only one compute unit is defective, disabling it means you've disabled half of them.

Now, there are some other, OEM-only parts that I don't list, in some cases including further cut down salvage parts.  But those are such low volume parts that Nvidia didn't bother trying to sell them to the general public.  If you have few enough chips in such a low bin to sell that Dell will take all of them off your hands at an agreeable price, Nvidia did stuff like that.

If the GeForce GTX 1060 roughly matches a Radeon RX 480 in both price and performance (as seems plausible), there will be vastly more demand for it than for the GTX 1070 or GTX 1080, simply because there is far more demand for $200 cards than for $400 or $600 cards.  Do you really think Nvidia wants 80% of the GP104 chips they sell to be half disabled?  For the volume Nvidia has, it's much more efficient to just make another, smaller chip unless yields on GP104 are so astoundingly awful that 80% of the dies have to be cut down that far.

13lake

I completely agree that a 3rd salvage part is idiotic, extremely risky, and generally just a bad and dumb idea.

What i'm trying to get across is that there is a pretty good chance Nvidia will do it. And not only could they do it, they would benefit from it, no matter what the shitstorm it creates, and how bad in a myriad of ways it turns out to be.

we're talking about a company which completely fucked up everything from their 8000 series > 900 > 100/200 > "300" > 400. And only managed to get back on track with their 500 series.

They had no intermediate or longterm repercussions whatsover.

Their drivers have been going to shit for ~2 years and you still have tens of thousand people answering with : "Nvdia driver perfect, AMD driver shit, AMD driver always was shit like 10 years ago".

Have they lost customer numbers because of drivers ?, nope, negligible loss, it's chugged down to over-enthusiastic consumer centric approach, and just a unfortunately negligible consequence of being too fast, too good, with so many new features released , etc, ...

Nvidia is changing the landscape so fast with their drivers and gameworks, they don't have time to dabble with small insignificant little exponentially increasing bugs

It's the same active denial method, that was used the whole time AMD cards were more efficient (before maxwell) with the convenient excuse : "Peasent AMD is obsessed with power efficiency which is irrelevant to pro PC enthusiast gaming, Nvidia knows what PC gamers need, and it's pure power at any power consumption cost"

As soon as Nvidia cards became more efficient the same regular suspects turned their story around to "power efficiency over all for pro PC enthusiast gamers".

13lake

Man who the hell enters any random shop anyway in the US to buy any pc part ?? The only shop you go without checking prices online and calling the shop itself is Microcenter to get cheap cpu and/or mobo combo

As for a statistical example of availability, i've already given my opinion on that

SEANMCAD

13lake said:

Man who the hell enters any random shop anyway in the US to buy any pc part ?? The only shop you go without checking prices online and calling the shop itself is Microcenter to get cheap cpu and/or mobo combo

As for a statistical example of availability, i've already given my opinion on that

basically its rare and rather silly for someone to walk into a store to pick up a PC part which is EXACTLY why I use it as a point of reference.

Products that are on the shelf at Best Buy are the extreemly common ones. If you can get something on the Shelf at Best Buy you should have ZERO trouble getting it online and shipped to you within a few days pretty much 100% of the time.

Ridelynn

You would think - yet it seems there is trouble getting it online and shipped.