AMD RV870 rumored to launch in August

[onethreehill]

據台灣繪圖卡業者指出， ATI 終於在繪圖規格推出時程上再次領先 NVIDIA ，根據 AMD 最新繪圖晶片規劃，首款 Direct X11 繪圖晶片將會於第三季中上代，核心代號為 RV870 ，進度良好相信延期機會不大，而 NVIDIA 的 Direct X11 產品、核心代號為 GT300 的繪圖晶片，至少要到 10 月才能面市，失去搶佔市場的先機。

據繪圖卡業者指出，新一代 Direct X11 繪圖規格不僅再度提供繪圖技術的功能，亦同時能提升執行效率，現時 AMD 與 NVIDIA 已經透露了其下一代 Direct X11 產品的核心代號及推出時程，但最終規格則尚未明朗， RV870 及 GT300 最終誰能贏得首勝絕對是市場焦點，不過以推出時程而論， AMD 則佔上優勢， RV870 預計將於 2009 年 7 月底至 8 月初上市，而 NVIDIA GT300 則暫定 10 月中，兩者相差 2 個月，相信 AMD 能因佔上先機而取得銷售上的優勢。

新一代 Direct X11 不僅能提升編程上的彈性，令程式設計師更能有效應率地使用繪圖晶片，同時亦強化了 GPGPU 應用，預期 Direct X11 將會以長速度普及於市場上。

就 Direct X11 產品上市時程向 AMD 及 NVIDIA 查詢，兩者均表示不會對市場傳聞作出任何評論。

http://www.hkepc.com/database/images/thumbs/640x480.2009042011032909205380430.jpg

http://www.hkepc.com/2774
Translated (http://translate.google.com/translate?prev=hp&hl=en&js=n&u=http%3A%2F%2Fwww.hkepc.com%2F2774&sl=zh-CN&tl=en)

[jaredpace]

finally launched in the graphics process specification again leading NVIDIA, in accordance with the latest AMD graphics chip planning, the first Direct X11 graphics will be on behalf of the third quarter, core, codenamed RV870, is progressing well and believe that the extension of is very low, the Direct X11 and NVIDIA products, the core codenamed GT300 graphics chip, at least to 10 months to market, the loss of the opportunity to seize the market.
It has been pointed out that the graphics card industry, specifications for a new generation of Direct X11 graphics technology not only to re-provide graphics functions, but also can improve efficiency in the implementation of the existing AMD and NVIDIA has revealed its next-generation products, the core of Direct X11 code and the introduction of time, but in the end specifications are not yet clear, RV870 and GT300 Who won the final victory is the first market focus, but in terms of launch time, AMD accounted for on the edge, RV870 is expected to be by the end of July 2009 to market early in August, while the NVIDIA GT300 tentatively scheduled for 10 months, a difference of 2 months, I believe that AMD can be accounted for on a result of the opportunity to obtain the advantage of sales.
A new generation of Direct X11 will not only enhance the flexibility of programming, the programmers should be the rate of more effective use of graphics chips, but also strengthened the GPGPU applications, Direct X11 will be expected to speed universal long in the market.
Direct X11 on the time-to-market inquiries to AMD and NVIDIA, both on the market rumors that it would not make any comment.

[saaya]

dx11 on vista, yaaaay... so its not locked to windows 7 like they locked dx10 to vista alone..
oh wait, windows 7 IS vista... :D

what i wonder is if dx10.1 tesselation on current ati cards will work under dx11 or not... some of it should work at least, no?
or will dx11 treat dx10.1 cards with tesselation as if they didnt have any tesselation at all?

[hurleybird]

You'd think MS would make it similar to tessellation on Xenos, which would work in ATIs favor, but you never know.

[Shintai]

That would be a very shortlived 4890 and 4770. I doubt we see DX11 cards this year.


dx11 on vista, yaaaay... so its not locked to windows 7 like they locked dx10 to vista alone..
oh wait, windows 7 IS vista... :D

what i wonder is if dx10.1 tesselation on current ati cards will work under dx11 or not... some of it should work at least, no?
or will dx11 treat dx10.1 cards with tesselation as if they didnt have any tesselation at all?

Thats been sure since...ages. Less emo please.

[LordEC911]

dx11 on vista, yaaaay... so its not locked to windows 7 like they locked dx10 to vista alone..
oh wait, windows 7 IS vista... :D

what i wonder is if dx10.1 tesselation on current ati cards will work under dx11 or not... some of it should work at least, no?
or will dx11 treat dx10.1 cards with tesselation as if they didnt have any tesselation at all?

From my understanding, DX11 tessellation code will not work with current ATi Tesselators.
Though if devs code specifically for ATi's tessellator it will work with not just ATi cards, R600+, but also all DX11 hardware.

[eleeter]

I doubt we see DX11 cards this year.

I wonder how many sigs that will end up in. ;)

[Olivon]

I wonder how many sigs that will end up in. ;)

:ROTF: :clap:

[RejZoR]

As far as i understand from interview with Dave, there is a possibility for tesselator engine to be ported to DX11.
Just a possibility, probably through driver itself. Probably not as effective as native thing but should work i guess.

[HKPolice]

Uhhhh taped out in august maybe, but retail cards won't hit until oct/nov timeframe at the earliest.

[Smartidiot89]

That would be a very shortlived 4890 and 4770. I doubt we see DX11 cards this year.
4890 was just a fix in pricegap, nothing spectacular, HD4770 is just a card to put pressure on Nvidias aged renamed cards.

DX11 cards will come this year together with ATIs RV870 :cool:

[LordEC911]

Uhhhh taped out in august maybe, but retail cards won't hit until oct/nov timeframe at the earliest.

Uhhh... it takes longer than 3-4 months from tapeout to release...
RV870 should have tapedout in Feb.

[[XC] gomeler]

Yeah baby, yeah!

[RPGWiZaRD]

August? Hmm, I believe it when I see it, personally I expect an early-mid Q4 launch or so.

[Klarko]

August? Hmm, I believe it when I see it, personally I expect an early-mid Q4 launch or so.

Yup yup same here, i try not to get my hopes up anymore.

[Manicdan]

what would the be releasing between now and then? just the 4890x2, that should hopefully be stomped by the 5850 in CF. i think it will be too long of a period with no new advancements in the high end.

[RaZz!]

what would the be releasing between now and then? just the 4890x2, that should hopefully be stomped by the 5850 in CF. i think it will be too long of a period with no new advancements in the high end.

i hope the 5850 will perform like a 4870x2 and the 5870 even better. now that would be cool :D - even though it's unlikely i guess :p:

[Bo_Fox]

Yup yup same here, i try not to get my hopes up anymore.

LOL!!! "Why so serious?", as quoted by Joker in the recent Batman movie! It cracked me up!!!

Anyway, I expect ATI to "slack off" until they think that Nvidia is ready to launch their GT300 product.. it would most likely be an "announcement" kind of game where ATI makes announcements and then Nvidia counters with better-sounding announcements to the point where either one finally announces a launch date, and then it's finally set in stone. Personally, I would guess around October 2008... err, 2009! (Sorry, it's 2009 already, and still cannot get used to it..) when ATI does a hard launch and Nvidia does a paper launch when GT300 cards are not available for sale until maybe 2 weeks later.

[ownage]

That would be a very shortlived 4890 and 4770. I doubt we see DX11 cards this year.



4890 and 4770 are mid- and low-end (at this moment), and have a place in the market even when RV870 is in stores.
A few sources have already said RV870 will launch sooner than expected.

[turbox997]

4890 and 4770 are mid- and low-end (at this moment), and have a place in the market even when RV870 is in stores.
A few sources have already said RV870 will launch sooner than expected.

In addition the 4890 is just a overclocked and slightly tweaked 4870. I wouldn't see ATi losing too much if it was short lived anyways(more likely the 4870 will be eventually axed).

[spursindonesia]

Yeah, i have some personal infos from a few local ATi AIB distros friends here, they say RV 790 can very easily drop in price if the new ATi GPU generation comes in, the GPU yield is very2 good, GDDR5 price is dropping (especially for the lower binned types), and the OEM makers of the card charge less and less for the card BOM since the cards are truly matured in manufacturing process and the development cost would be depreciated quickly.

IMHO, the things that really bug RV870 development are the slow process of TSMC 40 nm fabrication ramping up and maturation. Also, the economic condition really hurts the enthusiast segment of VGA card market. IHVs will not launch the card unless it can be priced reasonably without taking a loss, and that would have to wait until yield and node performance improve.

[informal]

That would be a very shortlived 4890 and 4770. I doubt we see DX11 cards this year.


From here (http://seekingalpha.com/article/132172-advanced-micro-devices-inc-q1-2009-earnings-call-transcript?source=yahoo&page=-1),we see that Dirk Meyer expects dx11 hw to be ready in the second half of this year :

...Just introduced a new product on the high end. We talked about our 40 nanometer products, still got feature advantages in the form of DX 10.1 and GDDR5 technology which makes the products distinctive. And we plan to lead the market with the X11 solutions coincident with Microsoft Windows 7 release in the back half of the year. So we feel very good about the GPU situation....

...We’ll launch another product here soon and follow that up with I’ll say a wave of DX11 compatible products in the back half of the year on 40 nanometer....

There is a possibility windows 7 will be a 2010 thing,but DX11 hardware is ,as can be seen above,planned for H2 2009 launch(in AMD kitchen at least).

[LiquidReactor]

Uhhhh taped out in august maybe, but retail cards won't hit until oct/nov timeframe at the earliest.

Thanks for the info, makes a lot more sense then this rumour.

[LordEC911]

Thanks for the info, makes a lot more sense then this rumour.
Not really...
RV870 was always targeted at mid '09, late Q2, with some small delays it was pushed back into Q3. As of now, until we hear different, it is still Q3.

[DoubleZero]

There is a possibility windows 7 will be a 2010 thing,but DX11 hardware is ,as can be seen above,planned for H2 2009 launch(in AMD kitchen at least).
The RC is due May 5, the only question is how soon in 2009 is windows 7 going to be released.

[GAR]

Realistically i dont see these cards coming out till October time. Thats just me tho.

[SNiiPE_DoGG]

change the thread title guys, GAR has spoken and he knows because he is a TSMC insider...... :rofl: the people making these rumors know way better than you do so I wouldnt make baseless opinions from what you want to happen so Nvidia doesnt get smoked (I dont care what happens personally, GPU's will be released in the future is good enough for me)

[GAR]

I hope AMD is paying you well for marketing help, anyway i am testing 3 4890's on X58 ATM, so dont assume you know me and know what i want. Just keep on marketing AMD for free, they love fanboys like you. Every ATI thread i see i see you marketing them and praising them.

[SNiiPE_DoGG]

well I wasn't really saying anything about AMD being better than nvidia next gen nobody knows who's gonna do what there, it was more smoking nvidia in terms of speed to market.... but hey nvidia screwed me over in the past so no I'm not going to buy their products unless rv870 is like r600 (fail)

[Bo_Fox]

ATI Declares Intention to Lead with DirectX 11.
ATI Plans to Release Radeon HD 5000 in Time for Windows 7

http://www.xbitlabs.com/news/video/display/20090422075255_ATI_Declares_Intention_to_Lead_with _DirectX_11.html

"In time for Win 7".. so, does that mean ATI will wait until Win7 is released or what? When should Win7 be expected? November?

[informal]

http://www.xbitlabs.com/news/video/display/20090422075255_ATI_Declares_Intention_to_Lead_with _DirectX_11.html

"In time for Win 7".. so, does that mean ATI will wait until Win7 is released or what? When should Win7 be expected? November?

Posted on previous page and directly from earnings call transcript :D.

[Bo_Fox]

Posted on previous page and directly from earnings call transcript :D.

What do you mean by previous page? :confused: ahh never mind.. Dirk Meyer said it already. Maybe ATI is just trying to get Nvidia to slack off until November/Dec.

[informal]

I mean it's already posted here (http://www.xtremesystems.org/forums/showpost.php?p=3746941&postcount=22) and before xbitlabs picked it up :D.

[Shintai]

http://www.xbitlabs.com/news/video/display/20090422075255_ATI_Declares_Intention_to_Lead_with _DirectX_11.html

"In time for Win 7".. so, does that mean ATI will wait until Win7 is released or what? When should Win7 be expected? November?

Windows 7 as in "Released retail" is january 2010. Volume/OEm etc is sooner tho. Hopefully early Q3.

[ILikeCosmosS]

dam gpus and istanbul are making me post pone my rig for october not summer :(
if these 5870 gpus roll out in the summer time they are mine!

[LiquidReactor]

So this stuff is aimed at 2009 Christmas season? I guess ATI/Nvidia are hoping that by that time economy will more or less rebound and enthusiast gpu sales will recover.

[ILikeCosmosS]

this is pissing me off that if i build my computer in the summer i will get i7 but then by christmas the new proccesors hit + alot better gpus

[LiquidReactor]

this is pissing me off that if i build my computer in the summer i will get i7 but then by christmas the new proccesors hit + alot better gpus

Just build now... I already got a good HD 4890 but waiting out to see if thoes HD 4890x2 1Ghz will show up.

[Helmore]

Windows 7 as in "Released retail" is january 2010. Volume/OEm etc is sooner tho. Hopefully early Q3.

Where did you get that from? Microsoft itself has not really made any official announcements about when they will release Windows 7. HP is hoping to be able to sell PCs with Window 7 on it to consumers in October and I think retail sales will be around the same time if that's the case.

EDIT: linky: http://www.enterprisenetworkingplanet.com/news/article.php/3807741/Windows-7-May-Be-Closer-Than-Thought.htm

[xVeinx]

Where did you get that from? Microsoft itself has not really made any official announcements about when they will release Windows 7. HP is hoping to be able to sell PCs with Window 7 on it to consumers in October and I think retail sales will be around the same time if that's the case.

EDIT: linky: http://www.enterprisenetworkingplanet.com/news/article.php/3807741/Windows-7-May-Be-Closer-Than-Thought.htm

Shintai is correct. HP = OEM. Vista was released earlier to businesses and OEMs, while being later to retail. The goal is to try and boost computer sales in the short term, and then release the OS to the broader community. This also increases the buffer time that developers have to make needed driver fixes, etc. There is a slim chance that it will arrive in time for Christmas (AKA, december) but I wouldn't count on it.

To be a bit more on topic, DX10, much less 10.1, is still not tremendously widespread. 21 games have support for it, and even then the implementation is suspect for a few of them. If RV870 does release soon, it's features will not be fully utilized nor its power (via multithreaded code) realized for quite a while.

[v_rr]

Windows 7 as in "Released retail" is january 2010. Volume/OEm etc is sooner tho. Hopefully early Q3.

I´m typing on a notebook runing Windows 7 7022 32bit. So it´s very relative tell people "on time for Win 7".

[Helmore]

Shintai is correct. HP = OEM. Vista was released earlier to businesses and OEMs, while being later to retail. The goal is to try and boost computer sales in the short term, and then release the OS to the broader community. This also increases the buffer time that developers have to make needed driver fixes, etc. There is a slim chance that it will arrive in time for Christmas (AKA, december) but I wouldn't count on it.

With Vista the volume license customers could get it a couple of months before retail and OEM availability. Volume licence is not OEM. Windows Vista Enterprise for example can only be had through a Volume License, not an OEM license. The article I linked to quotes the president of Compal Electronics (manufactures for HP and Acer for example), I'll just quote it here:


It's all in the timing
Part of the reason that Sinofsky and his team are so busy may be that they're working to ensure Windows 7 makes its debut ahead of Microsoft's public timeline.

That theory got another shot in the arm this week when Ray Chen, president of Taipei-based Compal Electronics, revealed to Bloomberg News that the actual general availability date for Windows 7 would be far sooner than the software giant had been claiming.

"According to current planning, it should be late September or early October," Ray Chen, whose company is a Microsoft hardware partner and makes laptop computers for HP and Acer, told Bloomberg.
http://www.enterprisenetworkingplanet.com/news/article.php/3807741/Windows-7-May-Be-Closer-Than-Thought.htm

BTW, I myself don't really care whether Windows 7 will be released in either October or December, as long as I can buy it around the start of next year or maybe a little earlier.

[xVeinx]

With Vista the volume license customers could get it a couple of months before retail and OEM availability. Volume licence is not OEM. Windows Vista Enterprise for example can only be had through a Volume License, not an OEM license. The article I linked to quotes the president of Compal Electronics (manufactures for HP and Acer for example), I'll just quote it here:


http://www.enterprisenetworkingplanet.com/news/article.php/3807741/Windows-7-May-Be-Closer-Than-Thought.htm

BTW, I myself don't really care whether Windows 7 will be released in either October or December, as long as I can buy it around the start of next year or maybe a little earlier.

Right. The emphasis here, for both Shintai and I, is that the general public will not get their retail copies until 2010. Volume licenses will be available sooner than retail, sure. They always are. However, you don't see people benching the latest cards on Vista Enterprise either ;).

[LedHed]

Sounds like ATI is more worried about a 2 month window of having the first DX11 chips on the market than they are about overhauling the GPU architecture as NVIDIA has with the new GT300 specs. ATI can't simply keep doubling SPs/ROPs/TUs, NVIDIA isn't that easy to compete against..they are going to need some fundamental improvements or they can say goodbye after that 2 month window closes and the GT300 shows it's face.

[SNiiPE_DoGG]

Sounds like ATI is more worried about a 2 month window of having the first DX11 chips on the market than they are about overhauling the GPU architecture as NVIDIA has with the new GT300 specs. ATI can't simply keep doubling SPs/ROPs/TUs, NVIDIA isn't that easy to compete against..they are going to need some fundamental improvements or they can say goodbye after that 2 month window closes and the GT300 shows it's face.

lol those GT300 "specs" are PURE speculation you cannot possibly be telling me you consider them a fact....

[informal]

Sounds like ATI is more worried about a 2 month window of having the first DX11 chips on the market than they are about overhauling the GPU architecture as NVIDIA has with the new GT300 specs. ATI can't simply keep doubling SPs/ROPs/TUs, NVIDIA isn't that easy to compete against..they are going to need some fundamental improvements or they can say goodbye after that 2 month window closes and the GT300 shows it's face.

Many said the same thing before GT200 launched and we know how that one turned out :p:

[LedHed]

lol those GT300 "specs" are PURE speculation you cannot possibly be telling me you consider them a fact....

I'm not saying they are 100% accurate, but you can't tell me that all of it is false. Most pre-release data can always be called speculation, but as we know these "rumors" tend to turn out somewhat true.

[SNiiPE_DoGG]

it 6 months till release and the chip isnt even close to taped out yet, no one who would post it on the internet knows the real specs of the GT300, yet - the only thing we can assume is that it is faster than GT200 :rolleyes:

[informal]

A safe bet is to look in the past and see what previous gen. brought us.GT200 was 1.7x-1.8x G80 in pure performance in high quality modes,RV770 was a similar increase over RV670.Today RV790+GDDR5 is only some ~15% slower than GTX285 which is a card based on quite larger and more complex GPU(and yet less efficient).
So for GT300 a doubling in resources is not out of the question as well as dx11 support.Similar goes for RV870,a 2000 SP,2x/2.5x count in rops and tmus when compared to RV770.

[clayton]

It's great to have rumors for both ATI/nVIDIA's next-generation hardwares today. That'll contain fanboys for sure!

[LordEC911]

it 6 months till release and the chip isnt even close to taped out yet, no one who would post it on the internet knows the real specs of the GT300, yet - the only thing we can assume is that it is faster than GT200 :rolleyes:
If nvidia wants to actually hit an early Q4 release, they should have just finished tape-out or wrap it up in May.
6-8 months from tape-out to release is the norm.

But you are right. We don't hear any concrete rumors until first samples comeback from TSMC and some of them fall into the wrong hands or some confidential docs get glanced over by a curious intern, meaning we are still at least a month or two away from hearing any solid info.

[Xoulz]

Windows 7 as in "Released retail" is january 2010. Volume/OEm etc is sooner tho. Hopefully early Q3.

Hmmm... last news I heard was that Windows7 will be on store shelves before X-Mas. Microsoft has said numerous times that W7 is moving along nicely! I think MS will push W7 publicly for the holiday season, with most sales opportunity on holiday laptop and PC purchases.

[Shintai]

Hmmm... last news I heard was that Windows7 will be on store shelves before X-Mas. Microsoft has said numerous times that W7 is moving along nicely! I think MS will push W7 publicly for the holiday season, with most sales opportunity on holiday laptop and PC purchases.

Windows 7 is actually moved back about 2 months from the most optimistic schedules. With a public RC coming next month. With a luck it will be RTM in august or september. July if a miracle happens. Then about 2 months later OEM should start ship. And 3 months after the OEMs the retail in the stores.

[Bo_Fox]

Windows 7 is actually moved back about 2 months from the most optimistic schedules. With a public RC coming next month. With a luck it will be RTM in august or september. July if a miracle happens. Then about 2 months later OEM should start ship. And 3 months after the OEMs the retail in the stores.

I thought it should take around 6 months after the public release candidate, for all the testing, bug-ironing, etc..??

[Bo_Fox]

A safe bet is to look in the past and see what previous gen. brought us.GT200 was 1.7x-1.8x G80 in pure performance in high quality modes,RV770 was a similar increase over RV670.Today RV790+GDDR5 is only some ~15% slower than GTX285 which is a card based on quite larger and more complex GPU(and yet less efficient).
So for GT300 a doubling in resources is not out of the question as well as dx11 support.Similar goes for RV870,a 2000 SP,2x/2.5x count in rops and tmus when compared to RV770.

RV770 was more like 2x faster than RV670. The GTX285 is only like ~10% faster than the RV790, since RV790 is roughly equal to the GTX275, and the 275 has only like 5% slower core, 15% less memory bandwidth..

However, I do not believe that 2000sp's is possible. First, ATI will definitely have to increase the number of ROP's, which will increase the chip die size a lot more than adding just a few shaders. 40nm is so new and unrefined, plus the power consumption and heat dissipation is going to be even more of an issue. Even a 1.5x increase in SP/TMU's is a bit optimistic. I just do not expect ATi to make a mammoth chip like the GT200, from all the hinting and the direction that they have been heading towards (smaller chips, multiplied in Xfire, etc..). It's not like as if they are going from 65nm all the way to 45nm. 55nm down to 40nm isnt that dramatic. Unless the R870 has a new kind of architecture, the power consumption efficiency is not as good as Nvidia's architecture when measured by "watts per die size area", so there's no way that ATI is going to ever make a 500mm^2 chip (unless ATI wants to forget about an X2 version).

[SNiiPE_DoGG]

have you seen the rv770? its smaller than a gddr5 chip....

[Helmore]

RV770 was more like 2x faster than RV670. The GTX285 is only like ~10% faster than the RV790, since RV790 is roughly equal to the GTX275, and the 275 has only like 5% slower core, 15% less memory bandwidth..

However, I do not believe that 2000sp's is possible. First, ATI will definitely have to increase the number of ROP's, which will increase the chip die size a lot more than adding just a few shaders. 40nm is so new and unrefined, plus the power consumption and heat dissipation is going to be even more of an issue. Even a 1.5x increase in SP/TMU's is a bit optimistic. I just do not expect ATi to make a mammoth chip like the GT200, from all the hinting and the direction that they have been heading towards (smaller chips, multiplied in Xfire, etc..). It's not like as if they are going from 65nm all the way to 45nm. 55nm down to 40nm isnt that dramatic. Unless the R870 has a new kind of architecture, the power consumption efficiency is not as good as Nvidia's architecture when measured by "watts per die size area", so there's no way that ATI is going to ever make a 500mm^2 chip (unless ATI wants to forget about an X2 version).

Going from a 65 nm. process to a 45 nm. process is the same step as going from a 55 nm. process to a 40 nm. process. They are both are a full process node apart. Although I do agree with you, power consumption is probably going to be a major issue for ATI and probably their main performance limitation. Unless they don't mind a big die, where you throw a lot of execution units at your problem but clock them a lot lower to decrease voltage and thus power consumption. But ATI has so far said to not go that route again.

[Bo_Fox]

Going from a 65 nm. process to a 45 nm. process is the same step as going from a 55 nm. process to a 40 nm. process. They are both are a full process node apart. Although I do agree with you, power consumption is probably going to be a major issue for ATI and probably their main performance limitation. Unless they don't mind a big die, where you throw a lot of execution units at your problem but clock them a lot lower to decrease voltage and thus power consumption. But ATI has so far said to not go that route again.

Hmm, I guess you are right, but it's not AS dramatic as from 65nm to 45nm. 40nm is hardly considered a half-shrink between 45nm and 32nm, since 40nm is so much closer to 45nm that it's almost the same thing.

Also, ATI would have to reduce the voltage and clock speed so much that it looses way too much performance compared to Nvidia's cards. ATI's chip architecture uses 4+1 shaders for each unit, and consumes more power per transistor than Nvidia's for similar performance. The benefit that ATI has is reduced die size area (less trannies needed), which is better from a financial manufacturing standpoint. I'm guessing that for Larrabee, it would be even worse than Nvidia, needing a huge chip real estate, but can perform exceptionally well at lower voltage without giving off too much heat.

[LordEC911]

RV770 was more like 2x faster than RV670. The GTX285 is only like ~10% faster than the RV790, since RV790 is roughly equal to the GTX275, and the 275 has only like 5% slower core, 15% less memory bandwidth..

However, I do not believe that 2000sp's is possible. First, ATI will definitely have to increase the number of ROP's, which will increase the chip die size a lot more than adding just a few shaders. 40nm is so new and unrefined, plus the power consumption and heat dissipation is going to be even more of an issue. Even a 1.5x increase in SP/TMU's is a bit optimistic. I just do not expect ATi to make a mammoth chip like the GT200, from all the hinting and the direction that they have been heading towards (smaller chips, multiplied in Xfire, etc..). It's not like as if they are going from 65nm all the way to 45nm. 55nm down to 40nm isnt that dramatic. Unless the R870 has a new kind of architecture, the power consumption efficiency is not as good as Nvidia's architecture when measured by "watts per die size area", so there's no way that ATI is going to ever make a 500mm^2 chip (unless ATI wants to forget about an X2 version).

Wish I had my numbers that I worked out yesterday in front of me...
Anyways, RV870 will, most likely stay under 300mm2.
Linear shrink of RV770 to 40nm puts it around ~155mm2.
A 2.5x increase to shaders and texture units from RV670 to RV770 only increased transistor count by 1.44x.

I am not expecting another 2.5x increase to shaders/texture units but at least a 2x increase. I really like the idea of a 32sp per cluster with 10 clusters. I would think they would increase ROPs, either by number or by buffing them up. Use 6.3ghz GDDR5 to hit 200Gbps, clock the core at the very least 650mhz to hit 2TFlop and even with all those increases you are still only talking about a chip around ~250mm2 w/ ~1.3b trannies.


Hmm, I guess you are right, but it's not AS dramatic as from 65nm to 45nm. 40nm is hardly considered a half-shrink between 45nm and 32nm, since 40nm is so much closer to 45nm that it's almost the same thing.
Just a 20% decrease in size... guess that is something to scoff at.
Plus with 40nm, it allows for a decent improvement in density as well.

[SNiiPE_DoGG]

w/ ~1.3b trannies.


ew.....:p:

[Bo_Fox]

Wish I had my numbers that I worked out yesterday in front of me...
Anyways, RV870 will, most likely stay under 300mm2.
Linear shrink of RV770 to 40nm puts it around ~155mm2.
A 2.5x increase to shaders and texture units from RV670 to RV770 only increased transistor count by 1.44x.

I am not expecting another 2.5x increase to shaders/texture units but at least a 2x increase. I really like the idea of a 32sp per cluster with 10 clusters. I would think they would increase ROPs, either by number or by buffing them up. Use 6.3ghz GDDR5 to hit 200Gbps, clock the core at the very least 650mhz to hit 2TFlop and even with all those increases you are still only talking about a chip around ~250mm2 w/ ~1.3b trannies.


Just a 20% decrease in size... guess that is something to scoff at.

Wow, let's just hope that you're right! I do hope so!!! :D

But something just tells me that it's too good to be true. *sighs* If increasing shaders only required a few more trannies, why the heck didnt the 4890 come with at least 960 shaders and 48 TMU's so that ATI could destroy GTX 285 for good? Instead, ATI went through all the trouble to modify the chip so that it could be clocked 10% more, and that's it. That was a lot of trouble...

I think that ATI will be adding more than just shaders. ROP's are known to bring a massive increase in die size, and ATI will definitely be needing more ROP's to avoid bottlenecking anything beyond RV790. Not just that, but some new features, like an improved DX11 tessellator unit, and a few surprises here and there...

[LordEC911]

Wow, let's just hope that you're right! I do hope so!!! :D

But something just tells me that it's too good to be true. *sighs* If increasing shaders only required a few more trannies, why the heck didnt the 4890 come with at least 960 shaders and 48 TMU's so that ATI could destroy GTX 285 for good? Instead, ATI went through all the trouble to modify the chip so that it could be clocked 10% more, and that's it. That was a lot of trouble...

I think that ATI will be adding more than just shaders. ROP's are known to bring a massive increase in die size, and ATI will definitely be needing more ROP's to avoid bottlenecking anything beyond RV790. Not just that, but some new features, like an improved DX11 tessellator unit, and a few surprises here and there...

The problem with that is ATi was able to make those changes in less than ~6 months, from the launch of RV770. To even slightly revise the architecture to add another cluster and another quad of texture units would take close to a year which would have put a 960SP RV790 within a quarter of RV870.