ATI Prepping GPU Refresh for 2H 2010

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
According to this news item, AMD’s CEO said that ATI is on track to refresh its line-up of graphics cards in 2H 2010. Thanks to H-street for the heads up.

“We are ramping the ATI Radeon HD 5000 series now and look forward to refreshing the entire lineup in the second half of next year*,” said Dirk Meyer, chief executive officer of AMD, during quarterly conference call with financial analysts.
Click to expand...
 
Oh crap, TSMC can't even produce enough of the current series to satisfy demand, and now they're talking refresh. :eek:
 
ATI is on a roll. Southern Islands recently taping out and refreshes coming for the HD5xxx series. With execution like that, they'll have a fully refreshed second generation DirectX 11 line up before Nvidia finishes getting their first out the door.
 
qdemn7 said:
Oh crap, TSMC can't even produce enough of the current series to satisfy demand, and now they're talking refresh. :eek:
Click to expand...
That hasn't stopped ATI from selling over 6 million 40nm DirectX 11 GPUs. :rolleyes:
 
Yeah, I wonder how overpriced the refresh is going to be? My guess is around $400 for the 5850 refresh and $500 for the 5870 refresh.
 
Darksword said:
Yeah, I wonder how overpriced the refresh is going to be? My guess is around $400 for the 5850 refresh and $500 for the 5870 refresh.
Click to expand...

Who cares, as long as they exceed performance-wise the 470 and 480, respectively. nVidia needs to get its sh!t together and become competitive again
 
NVIDIA's 480 scales pretty well when put together in SLI. Hmm, let's see..

NVIDIA:
Lower GPU and memory clocks
Higher temperatures
Pushing the envelope for electricity limits
Scales really well when put together with multiple cards
Single card at stock doesn't compete well; overclocks (with adequate cooling [=easy]) really well

ATI:
Higher GPU and memory clocks
Lower temperatures
Still has room for electricity limits
Scales alright when put together with multiple cards in Crossfire (could be better scaling!)
Single card (at stock) out performs a stock 480; overclocks alright??? (I don't know enough on this atmosphere)

Does that sound about right? Which has a smaller die for the most part?

At first, seeing as how the 480 has much lower clocks than the 5970 or whatever, you might think it's more efficient off the bat...but it generates more heat and electricity, so in this sense it doesn't.

But when you start putting the 480's into SLI, for those lower clock speeds, you get a competitive configuration in comparison to crossfire. Interesting!



I wonder if it's about improving scaling and crossfire/sli now, cause that's one way to go if their architectures can't be optimized and made more efficient in a timely manner..



Is what I say right for the most part? :?
 
They say "second half of next year." Does that mean fiscal year, and if so is that like 2011 or...?
 
mdk30 said:
They say "second half of next year." Does that mean fiscal year, and if so is that like 2011 or...?
Click to expand...
Donno, but ATI has quite an advantage with a lot of free time right now to experiment around and do their best to optimize their architecture as best as they possibly could.. :eek:
 
Darksword said:
Yeah, I wonder how overpriced the refresh is going to be? My guess is around $400 for the 5850 refresh and $500 for the 5870 refresh.
Click to expand...

I would presume them to be the same MSRP as the 5800s were at launch
 
Global Foundries has a 28nm process--and TSMC was reported as ignoring that and jumping right to 20nm, was it not? And the 5000 series is 40nm IIRC. I would hazard a guess that this refresh will be in the form of 28nm and it may be targeted at the low power market.
 
Does anyone know if ATI is doing anything to compete with nVidia's 3D Vision technology? I would think now would be the time to catch up on that front with all the 3D TV/Games/Movie stuff taking off right now.
 
mdk30 said:
Does anyone know if ATI is doing anything to compete with nVidia's 3D Vision technology? I would think now would be the time to catch up on that front with all the 3D TV/Games/Movie stuff taking off right now.
Click to expand...

A couple driver releases ago ATI included hooks for 3rd prty 3D technologies. ATI isn't making their own, instead they are going to rely on other people to support their cards.
 
Derangel said:
A couple driver releases ago ATI included hooks for 3rd prty 3D technologies. ATI isn't making their own, instead they are going to rely on other people to support their cards.
Click to expand...

Interesting. That would be awesome with EyeFinity I'd have to think. :)
 
Cerulean said:
NVIDIA's 480 scales pretty well when put together in SLI. Hmm, let's see..

NVIDIA:
Lower GPU and memory clocks
Higher temperatures
Pushing the envelope for electricity limits
Scales really well when put together with multiple cards
Single card at stock doesn't compete well; overclocks (with adequate cooling [=easy]) really well

ATI:
Higher GPU and memory clocks
Lower temperatures
Still has room for electricity limits
Scales alright when put together with multiple cards in Crossfire (could be better scaling!)
Single card (at stock) out performs a stock 480; overclocks alright??? (I don't know enough on this atmosphere)

Does that sound about right? Which has a smaller die for the most part?

At first, seeing as how the 480 has much lower clocks than the 5970 or whatever, you might think it's more efficient off the bat...but it generates more heat and electricity, so in this sense it doesn't.

But when you start putting the 480's into SLI, for those lower clock speeds, you get a competitive configuration in comparison to crossfire. Interesting!



I wonder if it's about improving scaling and crossfire/sli now, cause that's one way to go if their architectures can't be optimized and made more efficient in a timely manner..



Is what I say right for the most part? :?
Click to expand...

Everything now is pointing at excellent crossfire scaling compared to what it used to be.
The 58xx is a much smaller die than 4xx.
58xx OC capabilities are phenomenal, people (like me) are taking the 725mhz core on 5850's and pushing them to 900-1000mhz with a moderate voltage bump. 5870 can normally hit 1000mhz core with small voltage bumps.
 
If only amd/ati could be as competitive on the high end CPU side of things as they have been in the video card department.
 
badcaps said:
Global Foundries has a 28nm process--and TSMC was reported as ignoring that and jumping right to 20nm, was it not? And the 5000 series is 40nm IIRC. I would hazard a guess that this refresh will be in the form of 28nm and it may be targeted at the low power market.
Click to expand...

the roadmaps are a mess

but ok bear with me this is as much as i know based on what has been around for awhile


the ATI 5xxx refresh is "southern islands" its a mix (hybrid) of the "northern islands" gpu architecture (which was supposed to be 32nm but TSMC failed the process and later canned it) and the current 5xxx architecture so it basically looks like "southern islands" will be the 6xxx series
now because the 32nm failed and the 28nm is still unproven they'll have TSMC build "southern islands" in their proven and tested 40nm process


meanwhile nvidia is jumping to the 28nm process that TSMC will begin to test later this year (which doesnt really bode well for everyone involved since its still wonky and only seems to offer a theoretical 15% performance increase)

and ATI (i guess they had enough and realized they needed to have a different company work their silicon) won't have TSMC handle the 28nm production for "Northern islands" but rather they'll have Global foundries do it since they already have tested their 28nm process and found it doing well so far.

SO

to catch up TSMC was supposed to try doing the 22nm process but its not going well so they'll skip to the 20nm process (they've begun designing i guess) and they'll use some of the procedures that global foundries found useful in their 28nm process (high-K metal gate)


all that information is out there in the wild not just from charlie at SA
 
psychoticdream said:
the roadmaps are a mess

but ok bear with me this is as much as i know based on what has been around for awhile


the ATI 5xxx refresh is "southern islands" its a mix (hybrid) of the "northern islands" gpu architecture (which was supposed to be 32nm but TSMC failed the process and later canned it) and the current 5xxx architecture so it basically looks like "southern islands" will be the 6xxx series
now because the 32nm failed and the 28nm is still unproven they'll have TSMC build "southern islands" in their proven and tested 40nm process


meanwhile nvidia is jumping to the 28nm process that TSMC will begin to test later this year (which doesnt really bode well for everyone involved since its still wonky and only seems to offer a theoretical 15% performance increase)

and ATI (i guess they had enough and realized they needed to have a different company work their silicon) won't have TSMC handle the 28nm production for "Northern islands" but rather they'll have Global foundries do it since they already have tested their 28nm process and found it doing well so far.

SO

to catch up TSMC was supposed to try doing the 22nm process but its not going well so they'll skip to the 20nm process (they've begun designing i guess) and they'll use some of the procedures that global foundries found useful in their 28nm process (high-K metal gate)


all that information is out there in the wild not just from charlie at SA
Click to expand...

Really great info there; thanks. What a mess they're all in. I can only imagine how bad it's going to get when they reach the limits of the current materials and have to start experimenting with new ones--a situation we're hastening toward as I understand it.
 
I hope they get the prices of the 5xxx series to drop, that way I can upgrade.
 
Sometimes being poor has its advantages, by the time i get around to upgrading i'll have a matured 5xxx in my system ^_^
 
psychoticdream said:
the roadmaps are a mess

but ok bear with me this is as much as i know based on what has been around for awhile


the ATI 5xxx refresh is "southern islands" its a mix (hybrid) of the "northern islands" gpu architecture (which was supposed to be 32nm but TSMC failed the process and later canned it) and the current 5xxx architecture so it basically looks like "southern islands" will be the 6xxx series
now because the 32nm failed and the 28nm is still unproven they'll have TSMC build "southern islands" in their proven and tested 40nm process


meanwhile nvidia is jumping to the 28nm process that TSMC will begin to test later this year (which doesnt really bode well for everyone involved since its still wonky and only seems to offer a theoretical 15% performance increase)

and ATI (i guess they had enough and realized they needed to have a different company work their silicon) won't have TSMC handle the 28nm production for "Northern islands" but rather they'll have Global foundries do it since they already have tested their 28nm process and found it doing well so far.

SO

to catch up TSMC was supposed to try doing the 22nm process but its not going well so they'll skip to the 20nm process (they've begun designing i guess) and they'll use some of the procedures that global foundries found useful in their 28nm process (high-K metal gate)


all that information is out there in the wild not just from charlie at SA
Click to expand...

Damn, 20nm? Im so old I remember when TSMC was making 22CM parts...LOL Just kidding.
 
Ah jeez maybe I should wait and do a complete system overhaul in the fall. I don't really need a big performance boost at the moment.
 
Anyone know if they plan on fixing the band-with problem for Eyefinity, and Crossfire with this re-hash?

If I remember correctly in an article(or was it review) I read, that stated there isn't enough band-with over the crossfire bridge connector for Crossfire to run properly in Eyefinity in resolution higher than 2560x1600. Or at least not how they want it to.
 
Good news :) I was planning on buying two 5870's later this year but instead I will be going for the new Refresh cards.
 
badcaps said:
Really great info there; thanks. What a mess they're all in. I can only imagine how bad it's going to get when they reach the limits of the current materials and have to start experimenting with new ones--a situation we're hastening toward as I understand it.
Click to expand...

20nm is pretty close. They won't be able to half it with out some major changes. It is likely that this will be terminal size for many if not most applications, as 16nm chips are not expected in 2018, and even then they will probably not achieve good yields.
 
psychoticdream said:
the roadmaps are a mess

but ok bear with me this is as much as i know based on what has been around for awhile


the ATI 5xxx refresh is "southern islands" its a mix (hybrid) of the "northern islands" gpu architecture (which was supposed to be 32nm but TSMC failed the process and later canned it) and the current 5xxx architecture so it basically looks like "southern islands" will be the 6xxx series
now because the 32nm failed and the 28nm is still unproven they'll have TSMC build "southern islands" in their proven and tested 40nm process


meanwhile nvidia is jumping to the 28nm process that TSMC will begin to test later this year (which doesnt really bode well for everyone involved since its still wonky and only seems to offer a theoretical 15% performance increase)

and ATI (i guess they had enough and realized they needed to have a different company work their silicon) won't have TSMC handle the 28nm production for "Northern islands" but rather they'll have Global foundries do it since they already have tested their 28nm process and found it doing well so far.

SO

to catch up TSMC was supposed to try doing the 22nm process but its not going well so they'll skip to the 20nm process (they've begun designing i guess) and they'll use some of the procedures that global foundries found useful in their 28nm process (high-K metal gate)


all that information is out there in the wild not just from charlie at SA
Click to expand...

That's MOSTLY right from what we've heard.

I would just add that TSMC is supposed to be 30% greater clocking at the same power/die size on 28nm general process vs. 40nmG, not 15%. If you parse GF and TSMC's low-power 28nm claims, you get (an unmade claim of) 24% greater clocking for GF's 28G process versus 40nmG at TSMC. So basically, 5% faster in the same power envelope at TSMC...When/If it becomes mature and/or they don't fuck it up.

But if you look at SRAM sizes for 28nm though, .12 micron for GF, .127 for TSMC versus TSMC 40nmG which is .242, you see GF has a transistor density of 2x, while TSMC has 1.9x. Basically, GF should make the same amount of transistors with a little greater than 5% smaller die than TSMC on the same node.

These are options and trade-offs from the Gate-First and Gate-last approaches. While TSMC doesn't have anything proven from it's fab, it makes sense this is (generally) how it will go down.

Remember GF has been testing 32nm HKMG and it's 28nm shrink for what seems like forever, and has actual power/performance claims for it's 28nm ARM chips versus 40nm. I havn't heard jack from TSMC on actual results from it's first HKMG process; 28nm. You're right in the regard nVIDIA may be asking for disaster by jumping so quick.

So anyway, I havn't seen any claims ALL NI parts will be from GF, just AMD 'will make GPUs on GF 28nm', so it would make sense that either high volume parts (Redwood/Cedar replacements) for better yield or any parts designed with a higher transistor density but aimed at a lower clockspeed. nVIDIA OTOH, probably has a set design not aimed towards the strengths or weaknesses of either company, so it would make sense for a Fermi2 that they would use the process that creates the highest clock per watt, instead of greater density, hence TSMC. nVIDIA never has much cared about die size/yield anyway it seems.

As for TSMC nodes, they've basically given up on half-node steps by skipping the full nodes and making a half-node a full node. Bear with me:

Normal stepping would be 45nm full node, 40 half-node, 32F, 28H, 22F, ~18H, 16F.

TSMC is going 40F, 28F, 20F, ~14F.

They're just making one maximized node at each step to maximize profit and return on investment while hopefully leap-frogging the competition at each introduction. The thing is though, Moore's Law is supposed to kick everyone in the face at 18nm and slow down introduction of new nodes. IBM/GF will have a node at 18, TSMC will not. While TSMC may be out first with 20nm while GF is on 22nm (and perhaps testing 18nm), I would not have high hopes on them producing a well-working 14nm process any time largely before GF hits 16nm. There will likely be a time when GF has 18nm and TSMC is stuck on 20. It sounds good now, but we'll have to wait to see how it pans out.

Hope that helps a little...good synopsis of the situation though!
 
psychoticdream said:
the roadmaps are a mess

but ok bear with me this is as much as i know based on what has been around for awhile


the ATI 5xxx refresh is "southern islands" its a mix (hybrid) of the "northern islands" gpu architecture (which was supposed to be 32nm but TSMC failed the process and later canned it) and the current 5xxx architecture so it basically looks like "southern islands" will be the 6xxx series
now because the 32nm failed and the 28nm is still unproven they'll have TSMC build "southern islands" in their proven and tested 40nm process


meanwhile nvidia is jumping to the 28nm process that TSMC will begin to test later this year (which doesnt really bode well for everyone involved since its still wonky and only seems to offer a theoretical 15% performance increase)

and ATI (i guess they had enough and realized they needed to have a different company work their silicon) won't have TSMC handle the 28nm production for "Northern islands" but rather they'll have Global foundries do it since they already have tested their 28nm process and found it doing well so far.

SO

to catch up TSMC was supposed to try doing the 22nm process but its not going well so they'll skip to the 20nm process (they've begun designing i guess) and they'll use some of the procedures that global foundries found useful in their 28nm process (high-K metal gate)


all that information is out there in the wild not just from charlie at SA
Click to expand...

not quite sure where you get that

"ATI have tested 28nm with GF and doing good so far", this seems to be a huge FUD there...

I highly doubt GF can even make them yet..
 
Obi_Kwiet said:
20nm is pretty close. They won't be able to half it with out some major changes. It is likely that this will be terminal size for many if not most applications, as 16nm chips are not expected in 2018, and even then they will probably not achieve good yields.
Click to expand...

Yep! You posted while I was writing my big-ass thesis. Didn't think anyone else would add that,and it's very important to note. Being as close to 18nm as possible is important for longevity, and whatever commission fab creates the first well-yielding working sub 18nm is going to have a definite leg up. No one should assume TSMC will get there first because they will be on on < 22nm first.

On the same note, while 18nm for IBM/GF doesn't guarantee that either, I would bet they will be testing applications on that node pertinent to future development; like Intel using HKMG at 45nm. From watching TSMC struggle on past nodes, I doubt they'll deploy anything before they need to (or in the case of 40nm, it's too late) and would venture they will not take that extra step on 20nm. They could surprise us though!
 
One GPU generation lasts about 18months, and theres a refresh halfway through life, so to everyone who is complaining because they just bought a card, this was to be expected and you should have expected it too.
 
You must log in or register to reply here.
Back
Top