ATi Mobility HD 6000 series Roadmap

It's too soon for 28nm yet.

There's still plenty of model numbers for Turks and Caicos to take in the 6500s, 6600s and 6700s. A 640 SPU Turks could potentially take over for Juniper as the HD6670 or HD6770 while a 720 SPU GPU could even take up an HD6830 moniker.

 

Look what I found:
AMD Set to Introduce Radeon HD 7000 Southern Islands in Q2 2011, Rumor Has it - Softpedia

In all new 28nm flavour.

 

I won't trust Rumorpedia.net as an insider source until some of their info has been proven accurate....and since the site only just launched on Thursday, Nov. 4th that hasn't happened yet.

Honestly the site looks like it was set up as a spoof by a group of the fanboy-type commentators you find blasting articles on sites like SemiAccurate or Fudzilla (the articles are especially Fud-like), to drum up hits and then sell ad-space. Which, don't get me wrong, I think is a brilliant idea I wish I thought of myself. :wink:

 

this would mean that amd woud preatty much release it's card on the 28 nm of tsmc the day it will be released

 

How will the new mobile GPU compete with the 5870m?

 

a guess, the new midrange will be about the same speed and the new high end should be faster.

 

imo, i think the midrange would be slightly weaker than the 587m

 

sure could be, but unfortunatly until we see benchmarks its all a good guess.

 

New series wont be that much quicker, just more efficient in power usage. At most maybe 10-15% improvement over 5870m

 

Should preface that with "My guess is".

 

Hehehe, count me in on that idea.

 

Hardly and definitely a no. TSMC is in no condition to begin manufacturing 28nm. The only confirmation is Nvidia has a test unit @ 28nm and they don't plan on producing that until 2012.

And IMO I think there is more likely a chance AMD will release a 28nm first with Global Foundry. IMO after 40nm, TSMC won't be AMD's main producer anymore, there is a reason why AMD is in debt with funding the Global Foundry project.

 

A few examples of the amount of time that passes from first samples to launch:

AMD got Evergreen's first samples in March '09 and launched Cypress in the last week of Sept. '09. Supposedly AMD had Juniper ready before Cypress but held off on HD5700 cards until Oct. '09.

Nvidia saw the first samples of it's smallest Fermi GPU, GF108, before March '10 and started sending them out to OEMs by Sept. '10.

Back when GF100 was still called G300 the first rumors that Nvidia had gotten A1 samples started in May '09....GTX 480 didn't go on sale until near the end of March '10.

 

Some Cayman Radeon HD6970 specs - 1536 stream processors

4-Way VLIW. Notebooks should start seeing that arch at 28nm with Southern Island.

 

maybe nvidia don't plan on using it untill 2012 but TSMC will have it by 2011 and rumor say that amd will be using the process


also i don't know where you read that amd funded GFO but GFO where the foundry of amd befor they separated from them

that 6970 look sweet can't wait to see the 28 nm version of the 4way arch

 

From previous reports, 28nm production should start in the second half of 2011. That's what both TSMC and GFO aim for and I am pretty confident that in the second half of 2011 we will see 28nm cards enter mass production.

 

so 28 nm will be ready with 7xxx serie of card and the mobility varient will be up for ivy bridge

wich will be time to upgrade unless haswell promises enough new thing and dx12 is coming so i have to wait for the next gen of ati card to

 

It does look like that, yes.

 

Well, ATi has also used UMC (another Taiwanese company) for GPU & Chipset fabrication, too.

 

Well that's the whole the point of HD6xxx. It's a stop gap until 28nm fabrication is ready. Also reason why I don't see any point in upgrading from HD5870 to HD6970 when likely following year 28nm will be available.

 

AMD Bulldozer.

 

if it,s better then ivy bridge i could actualy switch to it but my curent setup is quite good (i love cfx) si i can still likely wait and see for 1 or 2 more years

but i do hope that amd bulldozer will be good enough to put intel back in it's place

 

AMD to Start Production of Desktop "Bulldozer" Microprocessors in April, i really hope AMD can compete with Sandybridge LGA2011/X68.

 

Yeah. Desktop Bulldozer in April hopefully means mobile Bulldozer in 2012.

AMD's APU don't have to be faster than Intel's top CPU, they just have to support Hybrid Crossfire with a top AMD discrete GPU (or two ) to be the better choice for gaming.


On a side note Anandtech's preview of Brazos has revealed the model numbers for the IGPs as HD6250 and HD6310 which should hopefully mean that Caicos/Seymour at HD6400 will have more than 80 SPU.

 

18w with the gpu on die that looks sweet a tablet with a such setup could be lees anemic on the gpu side then curent offering

if the clock for clock adventage is good enoug it could even be faster then my GF laptops

 

Here we go.



SemiAccurate :: AMD talks new mobile parts

Granville, Capilano, and Robson are back in the mix?

 

^ Sweet just what i've been wanting to see

 

And a general idea of the planned release date.

Fix'd

 

I kinda hoped to see BlackComb before Christmas.

 

I think the pattern for AMD releasing their high performance mobile cards is to announce them at CES and then release them a month or two later

 

Yup, but patterns tend to change. I was expecting less linear progress.

 

that would be sweet if they did come out before Christmas since I'm waiting for those to come out instead of pulling the trigger on a G53, which I probably won't get until the end of the year anyway considering the delays, but that roadmap does have the Vancouver series listed as 2011 cards which seems they'll end up sticking to their habits

 

The problem is that in March Nvidia will greet BlackComb with their fully optimized 485M/580M and this time top tier RED card wouldn't cost 350$

 

Ultra High-End Enthusiast sounds a lot like a 1120 SPU part to me.

A Barts based Blackcomb is plenty capable of beating a GF104/114 in performance and price.

 

I'm not so sure. Blackcomb is supposed to be somewhere between HD5830 and HD5850, right? If so, the later isn't too scary even compared to 470M, the first optimized card of the fiasco batch. I guess, 30-40% performance increase is not gonna happen...

 

A 1000+ SP Barts based mobile GPU would be quite more powerful than current HD5800m series. Considering that on the desktop front, a 1120sp part beats the 1440sp HD5850, a 1120sp barts vs a 800sp broadway will be quite an upgrade.

I just want it to be 32 ROP and 256bit GDDR5

 

What do you mean between HD5830/HD5850? Desktops?

Barts PRO is 4% slower than HD5850 on the desktop and neck and neck with a 336 Core/256-bit GTX 460 while using less juice. 470M is based off a 288 core/192-bit GF104 derivative.

If PRO can beat a GTX 580M with 336 cores, XT could crush one.


Consider a hypothetical line-up like this:

Blackcomb: New Barts XT; 1120 SPU/256-bit; SKU: Mob. HD6950/70
Granville: Rebranded Broadway; 800 SPU/128-bit; SKU: Mob. HD6850/70 (or hopefully Mob. HD680v)
Whistler: New Turks; 560-640 SP/128-bit; SKU Mob. HD6750/HD6770
Capilano: Rebranded Madison; 400 SPU/128-bit; SKU Mob. HD6650/70 (or Mob. HD660v)
Seymour: New Caicos; 160-240 SPU/64-bit; SKU Mob. HD6550/6570
Robson: Rebranded Park; 80 SPU/64-bit; SKU Mob. HD6450 (or Mob. HD640v)

 

Based on drivers (think; I'll try to find the post but I know it was mentioned in the drivers thread), it seems that HP at least will be using the HD6550 to replace the HD5650 in their current notebooks, so the 6550 will most likely either outperform or be a rebranded version of the 5650. I just thought that I should mention that.

 

Sorry, but could you elaborate the "could crush one" part?
Bart PRO is 4% slower than a HD 5850. If, BlackComb is 70-80% of the Bart PRO performance wise it would only slightly surpass a GTS 450.
470M is equal to/ slightly stronger than a desktop GTS 450. It also is nicely OC'able by 25%. If 485M is 10-15% above 470M - I fail to see much crushing.

But my main frustration isn't the performance but time frames.
Giving Nvidia 3 more months would result in more ferocious competition with the possibility of 580M entering the scene in March.
Guess, I'm one of those flawless victory dreamers, lol. When Moor is happy - I'm bored

 

Desktop HD6850 is ~28% faster than GTS 450....at 70-80% of it's clocks HD6850 would still be 15-20+% faster than GTS 450. The 2 extra SIMD on Barts XT could add another 5-10% on top of PRO.

Barts is also about the same size as the GF106 used for the GTS 450 (GTX 460M) which means it's cheaper to make than the GF104 used for GTX 470M or a possible GTX 580M. If the rumors are true Cayman won't be that much larger than GF104.

 

Has anybody seen this? Very interesting news, this could mean immediate availability of Mobility Radeon HD6xxx GPU's at Sandy Bridge's launch!
Is this the mobile variant of Barts (Blackcomb) we've all been waiting for?

 

Updated roadmap:

 

I wonder if those highlighted in orange will be 5000 series rebrands.

 

looks like they are not even writen second gen dx11

 

SemiAccurate :: Acer leaks a pair of Mobile Radeon HD 6000 models

Probably a valid assumption. HD6370M and HD6550M could be Robson and Capilano parts.

 

Maybe rebrands. But granville can't be a rebrand becaues the 5700 series simply don't exist.

 

There are three Mob. HD5700 parts. Only the Mob. HD5730 made it into notebooks though and most manufacturers just went with the HD5650, which uses the same GPU.

ATI Mobility Radeon? HD 5700/HD 5600 Series Graphics

 

If its still using 40nm which it is then my prediction is probably correct.

 

If rumours are correct and the 6000 Mobility series is just a few months away, then there is no room for 40nm cards.

Still, we are looking at a 20% performance jump, but I've already said this multiple times.

Edit: I meant "28nm cards"

 

the drop in power consumsion when they will jump to 28 NM will be great to see if they fit the 6970m in the standar mxm format