ATi Mobility HD 6000 series Roadmap

you guys missed something they match the 480 with a 150w - 200w tdp on 40 nm ...... they could have gonne bigger they just don't need to

 

Yeah, the desktop 69xx launch was pretty much fail. AMD had the chance to really stick it to nVidia in the desktop enthusiast end and failed. Hopefully, the mobile 69xx series will be more successful.

 

I am just dumbfounded with a lot of the responses in this thread. What the hell where you expecting with the HD6000 series? Have you not learned anything since HD4000? They have followed the same path, with their high end cards being dual GPU, and fighting for the market with performance/price.

AMD ATi released products which offer great performance per price, and per watt. They took the HD5000 series an refined it, delivering better products. They didn't really need to release a stupidly high power card just to brag about having the fastest single gpu card. That's stupid. Want the fastest single card? They will offer it with their dual GPU HD6990. The important thing is the price. There is little point of releasing a 2000+ SP GPU only to defeat the GTX580. And then what? Nvidia would release a useless update in the form of GTX590 just to win a stupid title.

But as things are, AMD just refined their architecture, making faster products with less stream processors and ended up offering better producs for less money. They had all this time to improve upon, and further prepare a full upgrade later on.

Hell, HD6850s in CF seem like a great choice for a low cost high end gaming rig, and the new HD6900 have also great performance scalability.

Just as the GTX570 aims for, being a superb product for a great price, these new AMD GPUs offer bang for buck. Want bragging rights? wait for the HD6990.

Seriously, with the release of Hd6850/70, the performance of HD6950/70 was pretty much predictable. They went and did what they were meant to do: supersede current offerings. HD6990 will simply be a better HD5970, and thanks to the improvements in architecture, they can trully make a great dual GPU single card offering.

 

When is the mobile series coming out? I saw benchmarks of 6900m series in the alienware thread but seems like there is no mention anywhere else of the card...

 

Yea well give us a break about AMD being about performance per price, precisely because for about one year now they have been awfully bad at delivering that. 5770 ? Underperforming. 5850 ? Too expensive. 6870 ? Sweet when they were available for 220€. Now I can't seem to be able to find them for significantly less 250€, which is a joke since the GTX 460 1GB is available at 180€.

I can only imagine the number of people who wish they bought a 5850 about a year ago right now. I almost wish I did. Well now the GTX 460 is nice for the price, but we had to wait about 10 months for that.

 

No idea where you live. But these prices are insane for the performance you get.
Newegg.com - Computer Hardware,Video Cards & Video Devices,Desktop Graphics / Video Cards,Radeon HD 6000 series,Radeon HD 6870

And you are right, the GTX 460 is a bang for the dollar. And it is also well known, Nvidia is selling this undercost, but that doesn't mean the HD6970 isn't the BEST Price/Performance on the market....

But I agree with ryzeki. Those who are talking trash about the HD6970 couldn't use their brains? Why would AMD sell the HD6970 @ $360 if the GTX 580 is selling at $600? Isn't that a massive clue? Did you all forget the HD5970 is still the most powerful single card, for less than GTX 580 you still get more power? Granted dual GPU have their issues. But isn't that a hint of what AMD has in plans for their $500+ card? And I guess also forgot that CF scaling is now at 100% with the new HD6xxx.
- And the irony is. The HD6970 will again like the HD5870, will outsell the GTX 580, GTX 570. At least AMD has brains. Getting more desktops with AMD is what AMD needs. The more share AMD has, the less Nvidia can bully their way. Gaming industry can no longer just take Nvidia's money and give the finger to the majority of desktop gamers. Also AMD did surpass Nvidia this year for more desktop shares than Nvidia.
- And it looks like AMD will also surpass Nvidia this year for mobile. Dell and Apple and HP all favoring AMD over Nvidia is a huge loss for Nvidia. AMD is doing something right
- Even then, I wonder where the incentive is to spend $550 for a GTX 580 with console port games? Hmmm? Where is the incentive when even the supposed most advanced game, "Crysis 2" is just another console port? When Metro 2033 the DX11 supposed monster, IMO monster failure, is a console port? Which means whether you play Metro 2033 in DX9 or DX11, it barely matters visually.

 

It's because NVidia overprices so much that even Apple has issues with it!

 

I dont get why the you guys are comparing the GTX580 to the 6970, just because its their top end for each company doesnt mean you should. The GTX570 is on the 6970 toes for very competitive pricing atm, i would say both are very comparable, here where the value starts, the GTX580 its on league of it own for people that want to dish $$$$ for the extra performance, just like there are people that drop big $$$ for Xtreme procesors.

anadtech review

 

Which makes the GTX 570 a better value, and after all of the delays and such it's a huge disappointed. There was too much hype, and it backfired. AMD is expected to at least lead in price/performance, and they even failed in that. This was a terrible launch, plain and simple, and NVidia won this round big time.

 

Yeah, cheaper and good performing doesnt mean better, this release seems mediocre.

Wait for 28nm next year?
Dudes... One of my microelectronics teacher works for a semiconductors company in Mexico, He showed us a 28nm chip last week (a crappy audio chip ES)... The companies make things like 1 year before the release... You can ALWAYS make a better performing 40nm chip, but Im not gonna explain why because it takes too long...

This is why Im also dissapointed they just want to refresh their line(rebranding). Now: Who has more brain?

Surely Nvidia learned a lesson in the last generation of gpus. And now is the king of the hill again with nice performing and smart priced products.

I love ATI more than Nvidia, but this time Nvidia clearly won.

 

Why squeeze every last ounce out of 40 nm when 28 nm is on the horizon? Law of diminishing returns, my friend.

And nvidia "winning" is all relative. They're only winning because they're willing to go to a higher TDP to do so. Meanwhile, ATI/AMD is comfortable churning out chips that are less power hogs and probably get better yield on the process. They're still competitively priced on performance per dollar. Not sure what else you want.

 

At higher game settings the HD6970 beats the GTX 580 in the best cases or matches GTX 570 in the worst....the HD6950 plays a similar back and forth with the GTX 570. It's at lower game settings that things become sloppy and AMD's cards look like garbage.

Compare the disparity in results between a review like [H]ard|OCP's which showcases what a card's highest playable settings are and a review like TechPowerUp's which bothers to benchmark enthusiast cards at 1024x768 and include those results in their averages....there's a different picture there where, for those looking for a card to run at the highest playable game settings, both HD6900 cards are a much better value than either GF110 card.

 

Because my friend, it is very "expensive" to buy the new bulbs and plastic covers (dont forget the beers of the engineers who do that) to mass produce @ 28nm laws of the economy... Lol what drives me crazy is that most of the people talk and talk about 40nm 65nm 32nm 28nm and they dont even know what it means, they dont even check wikipedia which is soooo wrong hahahaha.

Also it is very stupid to delay a product... Come on they already had the cards like 3 or more months ago. How long does it take to print a box? lmao.

 

What most ATi fans wanted is another spanking to Nvidia in all depeartments like happen with 5870, trully a great card that caught nvidia with its pants down. But now we are back how it was before, and most fans expected to sustain what ati did in their last gen against nvidia.

 

they could have had if they did scaled up the 6970 it would have spanked nvidia hard the curent card has a 208 tdp where the nvidia has a 355 tdp (both are test tdp not anunced TDP) and yet at higher resolution it gives fermi a run for it's money

while the driver are ZERO optimised for a 4 way arch vs nvidia that did all it could to perfect driver on fermi
he diference between fermi and the 4way arch in a realistic situation (aka not 1024x768) is on average 1-5 damned frame

 

It refers to the minimum feature size of the process, usually the minimum channel length the lithography can handle. The smaller the channel width, the more transistors fit in a given area, and the faster the transistor (generally, switching speed is partially related to channel length but is also related to applied voltage and threshold voltage of the transistor. Threshold voltages have been fast approaching their lower limits, meaning gate voltages can no longer reduce to save power. This is greatly responsible for the stalemate we see on clock speeds even as chips get smaller in addition to saturated drift velocities.)

But I could be wrong. I didn't check wikipedia

 

Hey +Rep thats the basis of the MOSFET (Metal Oxide Semiconductor Field Effet Transistor) Basically speaking in terms of CMOS tech it is the minimum channel length (distance) between P+ charges "floating" on an N- charge well. OFTEN the smaller the better as it allows to store more transistors on a smaller area, also the thermal voltage is smaller due the reduced space that explains why it is cooler.


I guess Im not being fair with ATI, the mobile cards are not released yet. We dont even have benchmarks atm. I would like to see how optimized the HD6970 is. Perhaps these cards are very nice OCers like the HD 5850

 

I wouldn't count on them being good OCers, considering that Barts uses more power and runs hotter than Juniper. I've had doubts from the start that Barts is really suited for mobile.

BTW, AMD puts an "M" at the end of their mobile Radeon model numbers now.

 

But you have to deal with higher leakage due to punchthrough and tunneling, bringing heat back up. Thus, clocks stagnate.

 

Seems chiphell believes the delay may have been caused by a PCB design flaw that resulted in a power connector physically interfering with the cooler or fan...requiring some of the plastic be trimmed off by hand on all of the initial run of cards.

The true reason behind the delay of AMD Cayman 6970/6950? - you'd never wanna know - XtremeSystems Forums


Now Nvidia fanboys have jokes to counter the wood screw jokes.

 

Seems like the new Cayman isn't that great as the HD 6970 performs around the GTX 570, so the GTX 580 is still top of the desktop cards.

AMD's Radeon HD 6950 and 6970 graphics processors - The Tech Report - Page 17

 

But it's also 15% better on performance per watt (6970 vs. 580) and 18% better on performance per dollar. ATI is still winning where they traditionally have. (it also eeks out the 570, which it is a direct competitor of).

 

Pretty interesting indeed, f? thats a serious design flaw! Well a delay is better than dealing with angry people like Nvidia did with the 8600 and 8400 chips.

 

At this point we don't really know what kind of architecture will the 6000 mobile series have. I seriously doubt it will be a 4D architecture so the whole discussion around the 6970/6950 performance is pointless. Why not? Because this will mean that AMD has to design a whole different chip specially for the mobile segment, and that just doesn't make economic sense. It's much simpler for them to derive the mobile cores from the desktop ones, and that means the 6870/6850 cores, because those fit the size and thermals for mobile cards.

I am inclined to believe that the 6970M will be derived from the desktop 6850. Which means about 20-25% more performance than the current 5870M while keeping the TDP at around 70W. And for those that can't wait, I am almost certain that the 6970M will be announced at CES 2011.

 

A mechanical issue like that is of far less impact than an electrical design flaw like the solder bump debacle. (actually, I'd say it was a materials design flaw because they chose a crappy solder makeup)

 

The VLIW4 thing isn't too important since it doesn't appear to me that it brings much benefits to the table in the end. I'm perfectly fine with the new high-end chips being based off Barts, which is a solid architecture and yet already much more potent than Juniper.

 

And if people actually bothered to read it all would see it only affects a few that use that particular fan. Not all and certainly not all AIB will be using it. And of course it's a cosmetic underneath the casing so you wouldn't even see it and has no affect on actual performance. Not like a fake FERMI with fake FERMI results...

As others have noted, the HD6970 is a huge success if you base it on what AMD has been producing for the last few years now. It's their best performance/watt/dollar yet. The dual vbios switch should be interesting and their new TDP system should be also. The HD6xxx also introducing MLAA, while still needs to be improved is something great to have considering so many games are now console ports with no AA support. So even if mobile can't use, for those who buy a HD6970 for AA performance, they should be very happy, no more trying to bypass or trying to find driver flaws that will allow them to enforce AA like you have to with Nvidia (Nvidia admitted the AA fix is actually a driver flaw).

But really, why is NBR so hung up over HD6970 when that will not have any impact for NBR? This is a laptop, mobile forum. It's the HD6870/HD6850 that should be scrutinized over. And please don't tell me Nvidia fanboys actually think a 580M would be a 570 or a 580...

The GTX 580 is the clear fastest GPU, but at what cost? I don't see a win here since anyone who is financial conscience at all would see the HD6970 is definitely the better buy. It is rumored AMD drivers will soon have Bullet Physics supported and now supports Stereoscopic 3D and all drivers now support OpenCL along with AMD Stream. So I don't see what there is to complain about here.

- And considering every AAA title people are excited about in early 2011 so far are console ports, EyeFinity is still > Nvidia's Tessellation/PhysX/3D.

How is this a GTX 580 win? A $525 card is only about 2-5 FPS than a $360?

 

VLIW4 has the same validity to HD6000M parts as 28nm does. Might even be VLIW4 has more. I'd personally wager all my money on them still being VLIW5, but if this were Vegas, and I were setting the odds, I'd set Whistler or Seymour as less of a long shot for being VLIW4 than 28nm.

I will agree that we could lose the GF110 is better than Cayman ipso facto GTX 580M is better than HD6970M until after the mobile GPU launch. I, myself, will try to keep it about the arch.

In the most positive light, at least it means there wasn't some more serious technical issue with the GPU arch itself. It does go towards dispelling the rumors that AMD was making last minute tweaks to the drivers because they were caught with their pants down by GF110.

VLIW4 was supposed to bring near 100% graphics performance for 90% of the shader space, but some are estimating that the new arch is a good bit below that mark still. That would be one of the areas where AMD tweaks things before VLIW4 comes to mobile.

 

I'm almost convinced that Ruckus is ziddy's new account.

Seriously though, kill the desktop talk, please.

 

I thought I was the only one who thought that

 

The 570 and 6970 perform the same at the same price point and power consumption according to the review I linked. However, they have unique differences which make one better than the other in certain situations.

The highest end GPUs are never bought for their price/performance ratio. This applies to all parts too - CPU, memory, etc. While the 69xx series (and high end nVidia) might not be directly used in the next iteration of mobile GPUs, it's still useful to use as a reference for future possible GPUs and for comparing notebook and desktop GPU performance.

 

A HD6870m / HD6850m based off of the Barts architecture (a refined VLIW5 architecture from Cypress/Juniper) sounds very nice indeed.

1120sp with a 256bit memory interface and GDDR5 but at lower clocks would at least be quite a bit faster than the current HD5870m.

Ati proved they could take the rather power inefficient (idle specifically) HD48xx series and directly turn them into the mobile series with reduced clocks so hopefully they can do it again with the far more power efficient HD68xx series.

 

You mean HD6970M and HD6950M.

I wouldn't count on 1120SPs. Even the 960SP HD6850 is noticeably more power-hungry than the HD5770, so they might already have issues squeezing that in. With 1120SPs, chances are that clocks would have to be reduced to the point that performance is reduced drastically.

 

yes they can

if you look there http://www.tomshardware.com/reviews/radeon-hd-4870,1964-15.html

the 4870 consume at least 180 more then the 3870 at idle (we don't know how much the 3870 consumed at idle but it's safe to assume it was more then 20) so the 4870 was likely around 200wats and yet they managed to squeze it into into a laptop for factor

so they could eve squeze the 6970 too

 

Dream on. Keep on dreaming. Fantasy land is fun land! Desktop 4870 is still more powerful than HD5870M. And a 4870 is not a 6970... which consumes 20% more power than a 5870.

 

I guess we'll see.

And no, they can't get a 6970 in there. Heat is a factor, too.

 

i never sait they could make it have as much power just that they could squeze the card without killing all performance

definitly a desktop 4870 is more powerfull then a hd5870m but my dual 4850 are depending of the game less powerfull then a 4870 and on other more powerefull then a 4890

so in dual gpu laptop models there is room for more powerfull setup then single card


heat is a factor to yess but it's not like the 4870 ran cold in it's time

 

Just wondering as a stockholder,

how is the AMD 6xxx series holding up against nVidia's new series?

I have heard from some reviews that while nVidia wins in performance in the high, high end, AMD owns nVidia overall in bang-for-buck, heat, energy, and drivers. Is this correct?

 

Other than the drivers part, that's more-or-less correct. Still no idea how things will translate to mobiles, though.

 

edit: ignore

 

Though, if AMD were to make the top mobile card based on the 6950, is it even likely that NVidia could beat them by much, if at all?

 

Oops, I meant to say 6850, and I'm then looked at the wrong prices. The 6850 is going for as low as $180, about where the 5770 was, when it became the Mobility 5870.

Therefore, it's safe to expect the 6900M to cost about the same as the 5870M does right now.

Nvidia has zero chance at matching it, in price or performance.

 

If it's based on the 6850, I think they actually do stand a chance. After all, we haven't even seen a GTX 560 yet.

 

We know just enough to speculate.



That can't go mobile without a severe dose of downclocking.

 

True, but even downclocked it could come close to a downclocked 6850.

 

Nvidia must be so much better than AMD that they can just downclock a full GF114 GPU to reach mobile TDPs while AMD is stuck having to use a cut-down Barts.

 

So true I had to read it with like three finger snaps.

 

Which is why, after looking into it more and reading what has been said here, I think think that a downclocked 6870 is a possibility. That being the case, NVidia wouldn't stand a chance. Now, this is all based on the assumption that, in the desktop market, the GTX 560 clearly beats the HD 6870 and actually competes well against the HD 6950. If the difference is only like 10-15%, then there's no way the GTX 580M will beat the HD 6970M. Of course, we won't know for sure until both cards are out.

 

The easy way to put it, is nVidia pushed the TDP with the gtx480m all the way to 100W - about... 50W more than the HD5870m. Not that means a lot, since the MXM spec allowed for it, but it means AMD doesn't have to worry about being the first to breach the barrier.

 

But the question is whether or not they'd be willing to. Personally, I think that they should keep the 6970M at around the same TDP they'd normally use and make a 100W 6990M for hyper-enthusiasts.