ATI Radeon HD 5890 Coming in Q1 2010

[003]

something that isnt real or is a paper dragon, cant get anyone nervous

This would imply that Fermi is a farce, does not exist in any form and never will, or that it is big on paper but will never reach retail. Both of which are false statements.

ATI is going to have some massive competition come Q1. A 5890 will help but does not appear that it will do any dethroning. Also I believe a 28nm process in mass production in Q1 10 to be a complete and total load of crap. They can't even get 40nm working.

It is extremely hard to take a large, high power and high complexity chip such as the RV870, shrink it a full node, while at the same time increasing power and complexity, apparently switching foundries, and putting it into mass production, all the while getting this done in a matter of months. That simply can't happen.

EDIT:
Anddddd.... the graph is already proven to be mere speculation:

this originally came from a Turkish site, the picture was not in the article.....it was posted by a forum poster.




http://www.donanimhaber.com/ATi_Rade...emde-16991.htm

[JohnZS]

5890 out before 5870 stocks really ever picked up.... Weird...and a shame

ATi have done that before.... with the X1800 series, by the time the cards were finally in stock in the UK the X1900 series was announced!

It's all a bit meh, especially since the default configuration of the 5890 does NOT have 2GB and the default configuration of the 5990 does NOT have 4GB.

Guess I am going to have to wait for Fermi..... (unless lots of ATi AIB's pack their cards with 2GB)

John

[RaZz!]

lol, this data is from wikipedia (and also states: "unconfirmed variant" on the right...) and the linked article doesn't offer any detailed information as well, so this is all speculation respectively made up

[Mungri]

Also I believe a 28nm process in mass production in Q1 10 to be a complete and total load of crap. They can't even get 40nm working.

ATI do have 40 nm working, it is TSMC that cant make the chips right. Nvidia will also have the same problem with their 40 nm chips.

Though I agree that we wont be seeing 28 nm any time soon, at least not for another year or two.

But really, please dont bother with taking such nonsense threads like these seriously, nor believing everything that you read in a graph or see posted on a forum, especially when the source is so rubbish - this is to everyone posting in this thread, not just the quoted person.

[SamHughe]

lol, this data is from wikipedia (and also states: "unconfirmed variant" on the right...) and the linked article doesn't offer any detailed information as well, so this is all speculation respectively made up

And that's the reason I have been thread-crapping in here (and in the Fermi pictured thread). Nothing but speculation based on random pictures/graphs found on the internet. There should be a "Xtreme Speculation" or Xtreme Gossip" forums. I'd recommend calling it Semi-accurate but I guess that one is taken.

[iTravis]

Remind me of this

[Mungri]

Lol @ 5870 x8 and its awesome power consumption.

Man, that thing would even totally cream Sam's balls in ballfire Y.

[Kej]

No ones answered my question about GloFlo skipping 32nm or not. Those articles listed contradict each other. I thought bulldozer was coming out on 32nm?

Edit: Answered my own question. Recent roadmaps show AMD with Bulldozer in 2011 at 32nm. TSMC isn't going to make these chips I'm assuming which leaves GloFlo. Why does everyone keep saying GloFlo is skipping 32nm?

Also what the heck, I just remembered..Llano cores are 32nm Phenom IIs. GloFlo must be skipping 32nm for GPU only, not CPU..is that what everyone means?

You have to differ between SOI and bulk-process here.
See page 6 on this PDF from AMDs Financial Day Earlier this month.
SHP are SOI, the others I presume are all bulk.
Sofar CPUs from AMD have been done on SOI while the GPUs from ATI have
been done on bulk. If any of this will change, that seems to be a matter of
a lot of discussion nowadays. :-)
GLOBALFOUNDRIES Presentation

[Kurz]

28NM! omgs... I am holding off still on 5870, but if 28nm comes out I am going to jump on that.

Though I doubt it'll happen.
I still have time before I need a new Video card.

[Jamesrt2004]

Somehow I find it hard to believe they're going to move to a 28nm process when they're having problems with the 40nm process. I'd surely like to see it, but I'm not holding my breathe.

tmsc wont be able to it up though 28nm = global foundries iirc

[madcho]

The first thing intel me that was a fake :

5890 : 2.5b transistors.

5990 : 4.7b transistors.

same amount of shaders ...

Same bus, all same.

Fakers are great in math they need the fields medal.

I'm sure 5890 is in foundry but past proven ATI with RV670 and RV770 can fake the informations about their new weapon, until 10-15j of the launch of the ship.

because there is a lot of fake information about a 5890, i think there is a large chance the chip is in foundry. but don't try to create your own 5890, your dreaming chip. ATI will do the best they can i'm sure. And i don't believe in 32nm tech. It's not the good frame time for 32nm.

And sam i love your photoshops, it's more fun than evrything i have seen before. Create your own chronic.

[SocketMan]

One thing is almost certain: ATI will bake their next gen. parts at G.F.,switching from TSMC.
I wonder what Nvidia's plan is for beyond the 40nm.
Will they stay with TSMC ?

[Tao~]

28nm mass production ain't happening till Q4 10 at the earliest.
Unless GF has 'magically' acquired it AND has exclusively licensed it to AMD ; both seem equally improbable to me.

The chart is imaginative as well, it proposes more architectural changes in 5870->5890 than in 4870->5870

[skugpezz]

This would imply that Fermi is a farce, does not exist in any form and never will, or that it is big on paper but will never reach retail. Both of which are false statements.

ATI is going to have some massive competition come Q1. A 5890 will help but does not appear that it will do any dethroning. Also I believe a 28nm process in mass production in Q1 10 to be a complete and total load of crap. They can't even get 40nm working.

It is extremely hard to take a large, high power and high complexity chip such as the RV870, shrink it a full node, while at the same time increasing power and complexity, apparently switching foundries, and putting it into mass production, all the while getting this done in a matter of months. That simply can't happen.

EDIT:
Anddddd.... the graph is already proven to be mere speculation:

ati had 40nm working since the 4770.

I have not seen fermi up to now, maybe I will see it around april next year. By that time I should have a 5850 and a 5950.

[Calmatory]

People, just let people dream. Gives you good laughs afterwards.

Like "The refresh will be on GF hooray".

[Stukov]

Remind me of this

You do realize that the 5850/5870 stats are correct in that right?

[ColonelCain]

This is just a speculation sheet, with actual specs from currently released cards to make it look more legitimate.

28nm? No way that this is even possible. And even if ATI wanted to do a pilot chip on 28nm as they did with the 4770 on 40nm, it would not be with a chip that is over 2billion tranies. That is stupid.
384bit bus width? This would be a dramatic change in thinking for ATI.

Again, this is just a speculation.


In the likely event that we will see a 5890, it will still be on the 40nm process. But, I would expect it to come out once TSMC gets their acts together.

[Helloworld_98]

donanimhaber is now added to the FUD website list

28nm isn't coming for a long time, AMD has admitted it, along with the fact that they have higher clocks in 90's, not more shaders.

[BrowncoatGR]

No ones answered my question about GloFlo skipping 32nm or not. Those articles listed contradict each other. I thought bulldozer was coming out on 32nm?

Edit: Answered my own question. Recent roadmaps show AMD with Bulldozer in 2011 at 32nm. TSMC isn't going to make these chips I'm assuming which leaves GloFlo. Why does everyone keep saying GloFlo is skipping 32nm?

Also what the heck, I just remembered..Llano cores are 32nm Phenom IIs. GloFlo must be skipping 32nm for GPU only, not CPU..is that what everyone means?

CPUs are made on SOI processes and the one beeing skipped(supposedly) is the 32nm bulk process. Specifically only the high performance one iirc

[Chickenfeed]

Realistically I see any 5890 to be clocked at 1Ghz with 1300-1400GDDR5. I have my doubts at 2GB reference even. Everything else so far = total fail.

[Manicdan]

i hope we see 1ghz minimum on 5890s with 2GB standard with OCed versions up to 1200Mhz stock. and the ram, it saddens me, im glad we can OC the crap out of it, but GDDR5 has not improved much in the last 2 years. but they sure do love to up the prices.

[Stukov]

To quote myself from another forum...

My math may be sketchy here, but let me give it a try.


1) Going from 40nm to 28nm is a 30% reduction.
2) 30% reduction in space increase # of dies per wafer ~30%
3) Current 40nm yields are reported ~50%
4) Hypothetically 125 dies per wafer @ 40nm x 50% yield = 62 usable dies
5) 125 x 30% increase dies per wafer = 162 total dies
6) 162 dies per wafer @ 28nm x 50% yield = 81 usable dies
7) 62 wafers @ 28nm divided by 162 = 38% yield

Conclusion, should 28nm have yields 38% or higher when compared to 40nm at 50% yield it is more economical to produce when dealing with the BOM (bill of materials). Given the increase performance, decrease in cooling, from an updated chip, profit margins are likely higher. There is no reason AMD wouldn't attempt this with current situation and with Fermi coming out around that time frame as well.

All 28nm has to do is be within ~12% yield of 40nm to economically feasible and assuming no additional increase in surface area to the die with adjusted math to the die.

[JoeBar]

This is fake. My

[turbox997]

As many have already implied, the 28nm seems rather bogus.

I see it impossible to have even a tape out product using GF's 28nm process(which doesn't even exist yet) by 1st quarter 2010.
The chart doesn't have much credibility, to say the least.

[Manicdan]

is the 6000 series going to be a new arch? and if so then should we see 5000 cards on 28nm first? or if 28nm is going to be ready so soon (due to bad tsmc forcing them to jump sooner to regain profit margins) then is it really that bad for them to jump to 28nm with a new arch, in a new fab? from a business perspective if ATI built a 6770 as their first card on 28nm at GF, it would make more sense, and then later scale that up 2x and make a 6870 once they get the bugs out.