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

[MooMoo444]

Ahh well, I guess me and my 980s will remain quite intimate for some time yet.

 

[ICDP]

You hope that AMD release their next top GPU at a price higher than Nvidia's current top card like they did with the 7970?

Are you really going to try and claim this is a fact? Nividia's current top single GPU was GTX 580 3GB version priced from £440 in Jan 2012 and HD 7970 was priced from £420 on the same date. So 7970 actually came in cheaper and faster than Nvidia's top current GPU. Please don't bring up the 1.5GB GTX 580 for comparison because that was NOT Nvidia's current top GPU at the time 7970 was released.

 

[andybird123]

I didn't pay anywhere near that price for my 580 3GB

http://hexus.net/tech/news/graphics/33453-amd-radeon-hd-7970-launch-day-pricing-availability/

pre-order no name was £420, actually in stock brand names were £450-530 - £450 was the official rrp on launch
a 580 is a 580, having a different amount of VRAM doesn't make it a different GPU core

I suppose going forwards we always ignore the 80ti cards and only use the Titan as the price comparison point

so basically going back to what was just said, "pulling a 5870/7970" will just be releasing a card slightly faster than a TitanX at £700 will be fine?
the 5870 is a good comparison and something to hope for... lumping it in with the 7970 is doing it a diservice

 

[TheF34RChannel]

NVIDIA “Pascal” GeForce 1000 series arriving at Computex

http://videocardz.com/58953/nvidia-pascal-geforce-1000-series-arriving-at-computex

The graphics card players will begin mass shipping their Pascal graphics cards in July and they expect the new-generation graphics card to increase their shipments and profits in the third quarter, the sources noted
– DigiTimes

 

[Rroff]

Well last ridiculous mock up was fermi. we all remember how that went
Do we have any more of such examples with ridiculous mock ups?

Either way, it is entertaining when we go from High End Pascal launch in April to first test boards pictures in April and from consumer reveal at GTC to 'they never planned to show off consumer Pascal at GTC', and from Computex launch to probable Computex reveal of consumer cards. So no let's start the clock for Computex

Actually can't remember entirely but I remember people posting similar not long before the original Titan dropped, etc. not just that they were showing ridiculous mockups but the lack of hardware on show at various events.

April was technically the earliest they could have come to market - though it was kind of unlikely - I said for awhile that there wouldn't be much information in relation to consumer Pascal at GTC

 

[melmac]

Unfortunately there's only amd to blame.

That's rubbish. Nvidia charge what they charge because the customers pays for it.

AMD have released their own really expensive cards, that customers have bought.

Competition only affects the more main stream cards, when AMD released the 290 cards, Nvidia dropped the price on the 780 cards. When Nvidia released the 9xx cards, AMD dropped the price of the 290 cards.

Did the release of the 290 affect Titan prices? Nope. Did the release of the 980Ti affect Titan prices? Nope.

 

[LoadsaMoney]

NVIDIA GeForce GTX 1080 and GeForce GTX 1070 With Pascal GPUs Under Full Production – Rumors Point to Computex Reveal

Reports from APAC based sources suggest that NVIDIA already has production of their Pascal based gaming GPUs in full swing.

http://wccftech.com/nvidia-geforce-gtx-1080-1070-computex-2016/

 

[Ian Evey]

There is always something not quite right with an amd launch, why would this time be any different.

 

[Harlequin]

digitimes is saying Maxwell based pascal cards.....

 

[LoadsaMoney]

^ Yeah they do.

 

[Mauller]

digitimes is saying Maxwell based pascal cards.....

Lol, if they know more than they let on, it more than likely means desktop parts that are pure SP shaders in hardware, the same as Maxwell.

And it is cheaper for nvidia to do this since a pure 3800 SP Pascal part would be less than 2/3 the size of GP100.

 

[muziqaz]

NVIDIA “Pascal” GeForce 1000 series arriving at Computex

http://videocardz.com/58953/nvidia-pascal-geforce-1000-series-arriving-at-computex

The graphics card players will begin mass shipping their Pascal graphics cards in July and they expect the new-generation graphics card to increase their shipments and profits in the third quarter, the sources noted
– DigiTimes

I LOVE it how it is twisted:
Computex is June, isn't it? May 31st-June 4th.

yet the article says it will be mass shipping in July, which is not end of May, is it? And it 'starts' shipping, which means not in shops right away, unless 1000 series deep learns their way into the shops

 

[muziqaz]

Lol, if they know more than they let on, it more than likely means desktop parts that are pure SP shaders in hardware, the same as Maxwell.

And it is cheaper for nvidia to do this since a pure 3800 SP Pascal part would be less than 2/3 the size of GP100.

Do you think it is cheaper to book two separate product lines in the fab where you have very limited wafer capacity available for you, AND you are trying to fulfill your HP contract obligations for massive chip on brand new process and new memory tech?

 

[Mauller]

Do you think it is cheaper to book two separate product lines in the fab where you have very limited wafer capacity available for you, AND you are trying to fulfill your HP contract obligations for massive chip on brand new process and new memory tech?

Yes because you get better yields on smaller dies. You have a higher percentage of working dies.

Gp104 and gp106 wont be cut down gp100. Which is why they can make them with only SP shaders and cut out any dedicated DP units.

 

[ICDP]

Yes because you get better yields on smaller dies. You have a higher percentage of working dies.

Gp104 and gp106 wont be cut down gp100. Which is why they can make them with only SP shaders and cut out any dedicated DP units.

This ties in with the rumours that consumer Pascal is a die shrunk Maxwell. A rebrand if you will

Note: the above statement may or may not be based on actual facts.

 

[muziqaz]

Yes because you get better yields on smaller dies. You have a higher percentage of working dies.

Gp104 and gp106 wont be cut down gp100. Which is why they can make them with only SP shaders and cut out any dedicated DP units.

But if you read my question more carefully you realise that it is not. It is not about the yields. And it is not simple to cut out DP units.
if nvidia had 2 designs for this new process, they would have 2 separate engineer teams, it is not like you just take big pascal, erase DP units in their software and cook the chip
As I said as well, TSMC is very busy and has very limited capacity for nvidia, so if nvidia wanted to buy out extra slot at busy fab, it would cost.
And that is without me mentioning yields which I guess are very low with 600mm2 tesla dies.

 

[muziqaz]

This ties in with the rumours that consumer Pascal is a die shrunk Maxwell. A rebrand if you will

Note: the above statement may or may not be based on actual facts.

No, please don't joke with rebrands I don't think it will be rebrands, but again, I don't think we will get to buy pascal based consumer cards this summer

 

[Harlequin]

Really do think we'll only see fat pascal on gtx 1080 and ti with 1070 and lower as rebrand maxwell

 

[Mauller]

This ties in with the rumours that consumer Pascal is a die shrunk Maxwell. A rebrand if you will

Note: the above statement may or may not be based on actual facts.

I think they will do a Maxwell and make them mostly with SP units. But still keeping all the improvements pascall brings of course.

But if you read my question more carefully you realise that it is not. It is not about the yields. And it is not simple to cut out DP units.
if nvidia had 2 designs for this new process, they would have 2 separate engineer teams, it is not like you just take big pascal, erase DP units in their software and cook the chip
As I said as well, TSMC is very busy and has very limited capacity for nvidia, so if nvidia wanted to buy out extra slot at busy fab, it would cost.
And that is without me mentioning yields which I guess are very low with 600mm2 tesla dies.

They have already shown a small unannounced die on a drive px2. For its size and 3-4tflops sp, it points to it being gp106.

The die itself was small. You don't get a die that small by cutting down a gp100 into a smaller size. They will have been purposely made. And it is easy for them to swap out parts of a uARCH design since they make them scalable. Gp100 will more than likely be its own design of Pascal SM compared to consumer versions because it is cheaper for them to have the two separate designs. You end up wasting up to 1/3 of the die on all consumer part if they keep the dedicated DP units in the gp106-104 parts. And that adds up to 1/3 less usable dies or more on every wafer they make. That is a lot of money wasted in the end.

 

[Appleseed]

Looking to upgrade. 1070 looks nice.