NVIDIA GeForce 4XX Series Discussion

[bobzilla2009]

well, since GD are primarily AMD linked (i.e. amd own a fair chunk), the chips shown on the 28nm wafer will almost certainly be something to do with AMD. If it is proclaimed to a be non-cpu chip, then maybe the first lot could be a couple of hd5 series refreshes to test the waters. I don't think amd would go for a new architecture on a new fab unless they were very confident, unlike nvidia who just tried to board the 40nm train without hardly any experience bar the lamentable uber low end chips.

 

[cadaveca]

cadaveca: you've misinterpreted.

120Hz does not mean 120FPS. it means 120FPS maximum.

 

[GSquadron]

The rambus patents were lawly right and nvidia needs to pay big money now. We all hope that it wont ban or even delay the launch of fermi cards. http://www.fudzilla.com/content/view/17390/34/

 

[crazyeyesreaper]

who cares about the rambus patents no one nvidia has been fighting with rambus for YEARS ati and everyone else just said screw it and paid them off maybe if nvidia wasn't so damn greedy they could have done the same thing intel did to amd heres your money go piss off oh wait thats right nvidia cant even produce chipsets or high end gpus these days there moneys running low. I guess they should have put a little more thought into what they were doing only reason i want Nvidia to release fermi is so ati will finish there gpu refresh and i can sell my 5850s and update to a newer better gpus for little cost

 

[Benetanegia]

Who is lying.... well I guess both could be lying. Both could be tring to prempt each other. But if I had to go with recent history. I would say Nvidia is probably more likely to be lying.

With the new release of Femi very shortly. Also with how long it is taking them to release it.... I would say the odds of us seeing any refresh from them this year would be highly unlikely!

But at the same time ATI might also be just finally feeling the nervousnous from the release of femi shortly. And they could be just saying this to say "don't worry we have something coming too.... Look at us!"

But like I said I lean towards more of my first thought then the second. Not only that GF did show off a 28nm waffer quite recently. That was proclaimed not to be a CPU chip. They didn't say what it was but it is speculated to be for AMD/ATI.

http://www.semiaccurate.com/2010/01/09/global-foundries-28nm-wafer-spotted/

I know it is semi accurate but it's still a rumor out there.

Well the HD2900XT was released in May and HD3870 was released in November, that's just 6 months later and it included a node change, which I don't think it's going to happen this time, not in 2010 at least. If Nvidia is releasing a refresh in November too and Fermi is released in March, that's 2 months more than AMD had to release the RV670.

The thing is that no other Nvidia card has to be delayed necesarily, just as no other AMD card was delayed. It's different working groups taking care of the different chips always. Mainstream Fermi cards have not been affected by the delay at all. There was going to be around a 6 month difference between high-end and mainstream cards (as always) and now it's just going to be 3 months or so. Same for the refresh, it was going to come 1 year after the release of high-end cards, but now it's going to be just 8 months later. That's talking about the technical aspect, if they decide to delay them or not, that would be a different thing. Fermi has been delayed because of production problems and not because the architecture fails in any particular place.

About the waffer. Not a single serious publication has speculated to be a GPU from AMD/Ati. It's obvious it's not a GPU, it's too irregular. Most sites, including Semiaccurate agree that it's most probably an ARM based SoC or a complex test pattern.

 

[erocker]

If there is going to be a refresh of Fermi by November.. A refresh of a card that isn't released yet, that doesn't say much about what's about to get released.

 

[troyrae360]

If there is going to be a refresh of Fermi by November.. A refresh of a card that isn't released yet, that doesn't say much about what's about to get released.

Lol, We will probley still be waiting for the damn thing in November, by then we will be deciding if we want to buy ATI 67** series or Fermi

 

[Benetanegia]

If there is going to be a refresh of Fermi by November.. A refresh of a card that isn't released yet, that doesn't say much about what's about to get released.

Why? Like I said HD3xxx was planned before the release of HD2xxx, much earlier than the date that HD2xxx was planned to be released, before it got delayed. GPUs are not designed in a pair of months it takes a lot of time, more than a year. In any case the Fermi refresh was planned as early on as when we first heard of Fermi and there were many reports. A chip designed for Geforce/Quadro/Tesla and a (cheaper) refresh just for GPUs has always been the plan since the G80. The first one usually comes with much higher memory bandwidth which is highly required on Quadros and Teslas, always based on memory width, because slower memory is required for reliability and higher densities on the profesional markets.

 

[erocker]

Why? Like I said HD3xxx was planned before the release of HD2xxx, much earlier than the date that HD2xxx was planned to be released, before it got delayed. GPUs are not designed in a pair of months it takes a lot of time, more than a year. In any case the Fermi refresh was planned as early on as when we first heard of Fermi and there were many reports. A chip designed for Geforce/Quadro/Tesla and a (cheaper) refresh just for GPUs has always been the plan since the G80. The first one usually comes with much higher memory bandwidth which is highly required on Quadros and Teslas, always based on memory width, because slower memory is required for reliability and higher densities on the profesional markets.

I realize this completely. It's just silly to hear news about something when it's "forefather" for the lack of a better term is delayed and late with still too little real information on it. I guess it's all just getting old and tiresome for me.

 

[phanbuey]

Why? Like I said HD3xxx was planned before the release of HD2xxx...

1. I think thats exactly why...

29xx = teh fail. Too hot, too big, and bottlenecked. (refresh planned too soon before release)

3xxx = less fail.

2. G80, g92 refresh was to counter 38xx series launch and target mainstream with cheaper cards..

GT200 refresh was to counter 4890 with more expensive cards than the originals, as 4890 was more expensive.

I agree with erocker, this is getting tedious, and there are no solid performance figures, and all the rumors are disappointing, which is something else it has in common with 29xx release.

 

[20mmrain]

Why? Like I said HD3xxx was planned before the release of HD2xxx, much earlier than the date that HD2xxx was planned to be released, before it got delayed. GPUs are not designed in a pair of months it takes a lot of time, more than a year. In any case the Fermi refresh was planned as early on as when we first heard of Fermi and there were many reports. A chip designed for Geforce/Quadro/Tesla and a (cheaper) refresh just for GPUs has always been the plan since the G80. The first one usually comes with much higher memory bandwidth which is highly required on Quadros and Teslas, always based on memory width, because slower memory is required for reliability and higher densities on the profesional markets.

I just don't see a refresh of possibly the most complicated card ever made 6 to 8 months after it is released. By then they probably won't even have most of the line out anyway.

ATI probably yes! By then it will have almost been a year if not a full year.

But Nvidia No Way! Well Maybe........ hey wait look there goes a pig out of my Ars!

I also have a feeling that if Nvidia was able to pull that off...... then they would piss off a lot of their customers who just spent 6 Months waiting for this card and forking out all of that dough. They would have then to buy the new cards all over again!

I'm really trying to look at away that a Nvidia refresh would be feasible .... but I just can't see it.

 

[trt740]

Anyone know when these chip/ cards are scheduled for release?

 

[cadaveca]

I'm really trying to look at away that a Nvidia refresh would be feasible .... but I just can't see it.

Maybe this can help you.

If we beleive this info to be correct, Jen Hsun visited TSMC in October in person:

http://www.fudzilla.com/content/view/15587/34/

Now, he could be going for the reasons stated @ the link, or he could have been going to drop off fresh designs.

Fermi has been in the works for some time now. More than likely, back in October, nV already had test silicon for GF100, and knew what the issues were, and what was involved to fix them. At the same time they got back first working silicon, new designs HAVE to be worked on. It's impossible to get perfect silicon, so a few revisions before release is generally the norm.

There are many ways to fix issues. It seems to me that the first GF100 chip we see will be very castrated in it's function, and there will be a new chip within 8 months of Fermi's actual release, that will bring in all the "extra" functions that will be disabled on the first chips.

Go back through the media, and you'll see Jen Hsun quite often over the summer. This was purposeful...and to me, rings of fund-raising. To me, that was for the refresh, that he dropped off for initial runs, back in October. There were probably several designs, too.

The reason for the lack of current info is that nV doesn't have enough chips yet for launch, and is unsure of how much exactly they will have to cut back current chips to satisfy demand until the refresh. They need all the chips before they can decide the final bins.

That, to me, is the only reason we are hearing about refreshes now.

 

[trt740]

Maybe this can help you.

If we beleive this info to be correct, Jen Hsun visited TSMC in October in person:

http://www.fudzilla.com/content/view/15587/34/

Now, he could be going for the reasons stated @ the link, or he could have been going to drop off fresh designs.

Fermi ahs been in the works for some time now. More than likely, back in October, nV already had test silicon for GF100, and knew what the issues were, and what was involved to fix them. At the same time they got back first working silicon, new designs HAVE to be worked on. It's impossible to get perfect silicon, so a few revisions before release is generally the norm.

There are many ways to fix issues. It seems to me that the first GF100 chip we see will be very castrated in it's function, and there will be a new chip within 8 months of Fermi's actual release, that will bring in all the "extra" functions that will disabled on the first chips.

Go back through the media, and you'll see Jen Hsun quite often over the summer. This was purposeful...and to me, rings of fund-raising. To me, that was for the refresh, that he dropped off for initial runs, back in October. There were probably several designs, too.

The reason for the lack of current info is that nV doesn't have enough chips yet for launch, and is unsure of how muc hexactly they will have to cut back current chips to satisfy demand until the refresh.

That, to me, is the only reason we are hearing about refreshes now.

The launch date has to be close look at the low inventory of 200 GTX cards.

 

[cadaveca]

I think the launch will be in march, with wide availability in April, as I've already said a few posts back. Maybe partners will get chips next month, these need to go on boards, which are probably already ready, then then sent worldwide..still a couple months out yet.

 

[Binge]

Massive disappointment only a few months away? Gee willagers I'm excited!

 

[troyrae360]

I think the launch will be in march, with wide availability in April, as I've already said a few posts back.

I would be supprised if they launch in march

 

[cadaveca]

I'm trying to remain optimisitic. I truly beleive that if TSMC did not have issues, we would have had cards back in November.

Yields are just so poor now that they have to wait it out until they have large quanities of chips, before they decide what to do, IMHO. Not thier fault, so I could care less.

 

[Benetanegia]

I just don't see a refresh of possibly the most complicated card ever made 6 to 8 months after it is released. By then they probably won't even have most of the line out anyway.

ATI probably yes! By then it will have almost been a year if not a full year.

But Nvidia No Way! Well Maybe........ hey wait look there goes a pig out of my Ars!

I also have a feeling that if Nvidia was able to pull that off...... then they would piss off a lot of their customers who just spent 6 Months waiting for this card and forking out all of that dough. They would have then to buy the new cards all over again!

I'm really trying to look at away that a Nvidia refresh would be feasible .... but I just can't see it.

I think what you guys don't understand is that it's not:

- Ok, we have to released X chip, let's start work on chip Y.

NO.

The reality is:

- Ok, it's Q4 of 2007, let's start working on chip X, it's going to be released around Q4 2009.
- Ok, it's Q1 2009 chip X is shaping up, we can maybe go for a Q4 2009 release, let's start work on chip Y (refresh) to release it in Q4 2010.
- Internal release dates: Team 1 working on Chip X - Q4 2009; Team 2 working on Chip Y Q4 2010.
- Chip X gets delayed: Team1 - Chip X Q1 2010; Team 2 - Chip Y Q4 2010.

Nothing changes!! There is no sense changing the schedule for the refresh chip! Whatever is wrong with chip X is inmediately fixed in chip Y, so whatever the problems are those are never inherited to chip Y. There's no need to mention that another team (Team 3) has been working on the mainstream variants from the beginning too...

Annoying people has never been a problem. Was people annoyed with the X850? With the X1900? With the GF 7900? X1950? HD38xx? All those cards were released in less than 8 months from the previous oneand this supposed Fermi refresh is going to supposedly be released in at least 8 months after GF100.

 

[20mmrain]

I do understand what you are explaining. I just don't see them doing it because of profit's reasons.

 

[r9]

Like I have said million times before. NVIDIA was late with DDR4 DDR5 DX10.1 DX11 and 40nm. And design big chips which are difficult to manufacture. I would be surprised only if NVIDIA released those cards earlier.

 

[Binge]

Like I have said million times before. NVIDIA was late with DDR4 DDR5 DX10.1 DX11 and 40nm. And design big chips which are difficult to manufacture. I would be surprised only if NVIDIA released those cards earlier.

ZzzzZZzzzZZzzz if you said it before what makes you say it again? NVIDIA really didn't go the large scale DDR4 route. They practically skipped it. Their 10.1 compatible cards were laughable. I think they just got fat from their previous generations and then got lazy.

 

[mastrdrver]

well, since GD are primarily AMD linked (i.e. amd own a fair chunk), the chips shown on the 28nm wafer will almost certainly be something to do with AMD. If it is proclaimed to a be non-cpu chip, then maybe the first lot could be a couple of hd5 series refreshes to test the waters. I don't think amd would go for a new architecture on a new fab unless they were very confident, unlike nvidia who just tried to board the 40nm train without hardly any experience bar the lamentable uber low end chips.

From what I understand, since no two fab companies (GF, TSMC, Intel, etc) have the same process for even the same node, ATI can't take their R800 from TSMC and just move it over to GF. The R800 is designed with the TSMC process in mind.

Also, the S/A article did not mention if that wafer had cpus or gpus on it. It just stated that the wafer had irregular looking chips on it which suggested that they were not SRam structures.

As for G100, I could see a quick refresh to 28nm at TSMC when it become available. It has been suggested that nVidia wouldn't release the full 512 core (GTX 380 if you will) G100 until it goes to 28nm because it is too hot and can't get enough chips out of 40nm.

 

[trt740]

for a company in trouble there stock is up.

 

[[I.R.A]_FBi]

for a company in trouble there stock is up.

Dont you know the formula, have press releases announcing mindless achievements