Why are AMD and Nvidia releasing more 28 nm video cards?

Quizzical

A few weeks ago, AMD released the Radeon Pro Duo, a dual Fiji card.  Last week, they released the FirePro S9300 X2, another dual Fiji card, this one for HPC use.  And now just today, Nvidia released the Quadro M5500, basically a desktop GeForce GTX 980 in a laptop except with Quadro branding--and with a scorching 150 W TDP.

Now, none of those are really meant for consumer use.  The Radeon Pro Duo is the heir to cards like the Radeon R9 295 X2 and the GeForce GTX Titan Z, which could at least kind of be called consumer cards.  But FirePro and Quadro are not for consumers at all.

But if the next generation 14/16 nm cards to replace them were about to launch next week, don't you think these cards wouldn't exist?  AMD's latest public guidance is Polaris around the middle of this year and Vega early next year.  If Polaris can beat out Fiji in performance, why release more Fiji cards so near to them being obsolete?  Ditto for Pascal as compared to a GTX 980--and energy efficiency is a huge deal in a laptop.

On the other hand, these releases make a ton of sense if the early Polaris and Pascal cards are either very delayed or slated to be lower end cards that leave the Fury X and Titan X as the top of the line into next year.  The new process node will bring new high end cards eventually.  But it might take a while to get there.

grndzro

Trying to get rid of old stock before the new chips come out. Is normal.

Quizzical

Launching a brand new $6000 card is a rather strange way to try to clear inventory, don't you think?

Thomas2006

I think both AMD and Nvida are playing there cards close to there chests on this one. Both want to be the first to market with there new series and both want to claim that top dog / fastest card award.

But I also have a feeling that both cards are going to be fairly expensive compared to what is out right now. I am curious to know how much TSMC had to play in the cards being delayed for both companies so long. With both companies using TSMC for the production of there chips I wander how much of a supply there is going to be right off the bat. I bet there is going to be some serious shortage going on for awhile as they both vie for production space.

Also worth noteing
http://wccftech.com/nvidia-pascal-gtx-1080-8gb-1070-launching-summer-debut-april-gtc-2016/

" Swerclockers makes no mention of when we should expect desktop Pascal graphics cards but the site goes on to claim that Nvidia is facing challenges bringing Pascal up to speed on TSMC’s 16nm FinFET which they say will in all liklihood throw a wrench in the company’s plans."

Quizzical

Thomas2006 said:

I think both AMD and Nvida are playing there cards close to there chests on this one. Both want to be the first to market with there new series and both want to claim that top dog / fastest card award.

But I also have a feeling that both cards are going to be fairly expensive compared to what is out right now. I am curious to know how much TSMC had to play in the cards being delayed for both companies so long. With both companies using TSMC for the production of there chips I wander how much of a supply there is going to be right off the bat. I bet there is going to be some serious shortage going on for awhile as they both vie for production space.

Also worth noteing
http://wccftech.com/nvidia-pascal-gtx-1080-8gb-1070-launching-summer-debut-april-gtc-2016/

" Swerclockers makes no mention of when we should expect desktop Pascal graphics cards but the site goes on to claim that Nvidia is facing challenges bringing Pascal up to speed on TSMC’s 16nm FinFET which they say will in all liklihood throw a wrench in the company’s plans."

Polaris is 14 nm, and TSMC has no 14 nm process node:

http://www.amd.com/en-us/press-releases/Pages/amd-demonstrates-2016jan04.aspx
http://www.tsmc.com/english/dedicatedFoundry/technology/16nm.htm

The fabs that have 14 nm are Intel, Samsung, and Global Foundries.  And AMD chips aren't going to be made at Intel.

ceratop001

Torval said:

I think there are delays and I think they're setting a marketing stage for comparison. I also think it would be more worrying if they launched an entirely new series of consumer cards as opposed to low volume high dollar niche cards.

In a couple of months (or so) when they do release they will be able contrast the power and performance improvements to what has been recently released. Or, like you say, it's possible they're not seeing the dramatic improvements expected. I doubt this though because both GDDR5X and HBM2 are huge improvements in themselves, let alone the improvements in the process node shift.

I'm going with delays being the most likely culprit and they're trying to work those delays to their favor.

It is hard to put yourself in their shoes and strategize an specific outcome in the future. Could be delays or an uncertain market.

Cleffy

Workstation cards are usually months to a year after consumer cards. Its just that time.

olepi

I think the 20nm node fizzled, and everybody is squeezing 28nm while they get the 14/16nm chips ready.

After that, it's 10nm, 7nm, 5? I'm getting dizzy.

Hrimnir

These are all business cards.  In some cases a business might need the fastest thing available right now and may not be able to hold off 2-3 months for a newer faster version.  When it comes to business, time is money.

So, nothing really that crazy about whats happening here.

Darksworm

FirePro and Quaddro are workstation graphics cards.  They are not gaming cards.

So I'm not sure why the OP even cares at all that they released those.  You won't be buying them for gaming.  You wouldn't even think of buying them for gaming because in many cases they cost substantially more than even top of the line gaming graphics cards despite giving worse gaming performance.

They're optimized for completely different types of applications (CAD, Modeling, Photo and Video Editing, etc.).

Malabooga

Cleffy said:

Workstation cards are usually months to a year after consumer cards. Its just that time.

Geforce GTX980 (GM204) releae date: September 18,2014
980 rebrand
Quadro M5500 (GM204) release date: May-June 2016

Yah, its "that" time

Ridelynn

Darksworm said:

FirePro and Quaddro are workstation graphics cards.  They are not gaming cards.

So I'm not sure why the OP even cares at all that they released those.  You won't be buying them for gaming.  You wouldn't even think of buying them for gaming because in many cases they cost substantially more than even top of the line gaming graphics cards despite giving worse gaming performance.

They're optimized for completely different types of applications (CAD, Modeling, Photo and Video Editing, etc.).

The reason it's pertinent is that everyone is jumping up and down about the impending Pascal release.

This is further evidence that Pascal, while it may get announced soon, probably won't be shipping in significant volume or with significant SKUs any time soon.

That goes along with reports that HBM2 isn't ready yet in any capacity, and that GDDR5X won't be in full production volumes until the summer, and AMD's Polaris (which uses the same memory types) won't be available until possibly this fall. And let's not forget that JHH showed another woodscrew card this January when he talked about Pascal, which strongly suggests they didn't even have engineering samples at the time.

So there a lot of doubts about when Pascal will actually be available for consumers which are actual upgrades to existing cards on the market. And the fact that the professional market is moving forward with 2 year old technology implies that current day technology isn't ready or mature enough for that market.

SEANMCAD

Ridelynn said:

Darksworm said:

FirePro and Quaddro are workstation graphics cards.  They are not gaming cards.

So I'm not sure why the OP even cares at all that they released those.  You won't be buying them for gaming.  You wouldn't even think of buying them for gaming because in many cases they cost substantially more than even top of the line gaming graphics cards despite giving worse gaming performance.

They're optimized for completely different types of applications (CAD, Modeling, Photo and Video Editing, etc.).

The reason it's pertinent is that everyone is jumping up and down about the impending Pascal release.

This is further evidence that Pascal, while it may get announced soon, probably won't be shipping in significant volume or with significant SKUs any time soon.

lets just do this: lets see how long it was between the first release of the last generation and its mass adoption and assume the time line for this generation will be the same sound fair?

SEANMCAD

Torval said:

Ridelynn said:

The reason it's pertinent is that everyone is jumping up and down about the impending Pascal release.

This is further evidence that Pascal, while it may get announced soon, probably won't be shipping in significant volume or with significant SKUs any time soon.

That goes along with reports that HBM2 isn't ready yet in any capacity, and that GDDR5X won't be in full production volumes until the summer, and AMD's Polaris (which uses the same memory types) won't be available until possibly this fall. And let's not forget that JHH showed another woodscrew card this January when he talked about Pascal, which strongly suggests they didn't even have engineering samples at the time.

So there a lot of doubts about when Pascal will actually be available for consumers which are actual upgrades to existing cards on the market. And the fact that the professional market is moving forward with 2 year old technology implies that current day technology isn't ready or mature enough for that market.

To add a little bit, from the article I read on Anand GDDR5X won't even be ready to start production until August. I'm not sure how long it takes to ramp up to full production volumes from there but I imagine they won't have a lot of inventory immediately.

So will the first Pascal cards still use GDDR5? That's not necessarily a deal breaker, but according to the Anand article there is a significant difference between the two. There is a lot up in the air and a lot of variables with very little information. I would love to get a few more details that would provide better conjecture.

to get an answer on the timeline do  you think its reasonable to look at the past generation timeline to get a baseline expectation?

Ridelynn

SEANMCAD said:

lets just do this: lets see how long it was between the first release of the last generation and its mass adoption and assume the time line for this generation will be the same sound fair?

Because the first Maxwell cards were not the Titan X and 980Ti,  "Maxwell" was introduced with the 750Ti, and it took more than a year for the 980Ti and Titan models to come out.

And let's just look at the fact that the 680 was "announced" in March of 2012, but it wasn't until late that summer that it was possible to really get your hands on one because the supply was so tight. That was effectively a paper launch, and done because Fermi was a complete trainwreck and they wanted to run as far and as fast away from that as they could.

Yeah, I think it's fair we assume this generation will be about the same.

SEANMCAD

Ridelynn said:

SEANMCAD said:

lets just do this: lets see how long it was between the first release of the last generation and its mass adoption and assume the time line for this generation will be the same sound fair?

Because the first Maxwell cards were not the Titan X and 980Ti,  "Maxwell" was introduced with the 750Ti, and it took more than a year for the 980Ti and Titan models to come out.

And let's just look at the fact that the 680 was "announced" in March of 2012, but it wasn't until late that summer that it was possible to really get your hands on one because the supply was so tight. That was effectively a paper launch, and done because Fermi was a complete trainwreck and they wanted to run as far and as fast away from that as they could.

Yeah, I think it's fair we assume this generation will be about the same.

from my research looking at wikipedia by looking at the Sandy Bridge lifecycle it appears to me that Haswell will become mainstream between now and no more than 12 (corrected) months from now.

Ridelynn

Who had that gif from princess bride - I think we need it again here

SEANMCAD

Ridelynn said:

Who had that gif from princess bride - I think we need it again here

so I am excited that I might be able to get my Haswell CPU before my Rift ships. I doubt it but it looks like it will be close. I wonder if the generation before Sandy Bridge has the same pattern. I will check

Ridelynn

So nVidia did announce a Tesla Pascal model, based on HBM2. Server availability Q1'17.

Ridelynn

And what do Intel CPUs have to do with an nVidia GPU timeline.

SEANMCAD

Ridelynn said:

And what do Intel CPUs have to do with an nVidia GPU timeline.

I wonder if GPU chips follow the same pattern. I have some time I will check and let you know

SEANMCAD

SEANMCAD said:

Ridelynn said:

And what do Intel CPUs have to do with an nVidia GPU timeline.

I wonder if GPU chips follow the same pattern. I have some time I will check and let you know

yeah it appears at least going all the way back to the 600 series that all the cards for that generation are on the market within 12 months of the family having been 'created' which in this case I think means today.

Ridelynn

So you'll be perfectly ok when "Pascal" is announced as the 950Ti with GDDR5 available this summer, and that the "1080Ti" / GP200 die isn't announced until March 2017, and released for retail in July 2017.

You called it, I guess.

Quizzical

Ridelynn said:

Darksworm said:

FirePro and Quaddro are workstation graphics cards.  They are not gaming cards.

So I'm not sure why the OP even cares at all that they released those.  You won't be buying them for gaming.  You wouldn't even think of buying them for gaming because in many cases they cost substantially more than even top of the line gaming graphics cards despite giving worse gaming performance.

They're optimized for completely different types of applications (CAD, Modeling, Photo and Video Editing, etc.).

The reason it's pertinent is that everyone is jumping up and down about the impending Pascal release.

This is further evidence that Pascal, while it may get announced soon, probably won't be shipping in significant volume or with significant SKUs any time soon.

That goes along with reports that HBM2 isn't ready yet in any capacity, and that GDDR5X won't be in full production volumes until the summer, and AMD's Polaris (which uses the same memory types) won't be available until possibly this fall. And let's not forget that JHH showed another woodscrew card this January when he talked about Pascal, which strongly suggests they didn't even have engineering samples at the time.

So there a lot of doubts about when Pascal will actually be available for consumers which are actual upgrades to existing cards on the market. And the fact that the professional market is moving forward with 2 year old technology implies that current day technology isn't ready or mature enough for that market.

There's a decent chance that the card Nvidia showed off in January actually worked--but it was Maxwell, not Pascal.

The Polaris cards that AMD has shown publicly were using GDDR5.  If you're willing to use GDDR5, not GDDR5X or HBM2, then you don't have to wait on memory vendors.

Malabooga

Ridelynn said:

Darksworm said:

FirePro and Quaddro are workstation graphics cards.  They are not gaming cards.

So I'm not sure why the OP even cares at all that they released those.  You won't be buying them for gaming.  You wouldn't even think of buying them for gaming because in many cases they cost substantially more than even top of the line gaming graphics cards despite giving worse gaming performance.

They're optimized for completely different types of applications (CAD, Modeling, Photo and Video Editing, etc.).

The reason it's pertinent is that everyone is jumping up and down about the impending Pascal release.

This is further evidence that Pascal, while it may get announced soon, probably won't be shipping in significant volume or with significant SKUs any time soon.

That goes along with reports that HBM2 isn't ready yet in any capacity, and that GDDR5X won't be in full production volumes until the summer, and AMD's Polaris (which uses the same memory types) won't be available until possibly this fall. And let's not forget that JHH showed another woodscrew card this January when he talked about Pascal, which strongly suggests they didn't even have engineering samples at the time.

So there a lot of doubts about when Pascal will actually be available for consumers which are actual upgrades to existing cards on the market. And the fact that the professional market is moving forward with 2 year old technology implies that current day technology isn't ready or mature enough for that market.

They showed same woodscrew board today

Its getting sooooooo embarassing for NVidia. But they are taking "preorders" 129 000$ for 8...

SEANMCAD

Ridelynn said:

So you'll be perfectly ok when "Pascal" is announced as the 950Ti with GDDR5 available this summer, and that the "1080Ti" / GP200 die isn't announced until March 2017, and released for retail in July 2017.

You called it, I guess.

I am predicting that in the future the consumer release for this generation of GPUs will be like the last two generations. 'calling it' would require something to have already happened. like an announcement of upcoming GPUs which is what I 'called'. I hope that clarified anything unless you were talking to someone else.