ATi Mobility HD 6000 series Roadmap

With nVidia's GF104 showing great promise in the desktop GPUs, ATI better not slack. nVidia has caught up in terms of lowering power consumption, while delivering strong performance, unlike their first iteration (GF100).

 

I wonder which manufacturer will present some PC with the 6000 series towards the end of September.

My guess is Asus, Eurocom (though they are known for selling products which do not exist) and maybe Alienware.

 

Eurocom is not a manufacturer, and Clevo is in love with Nvidia (though a bit adulterous when the lusty 5870 came along).

Since we're seeing low-end and mid-range 6000 series first, I'd also say Dellienware's not gonna be it.

 

Just some information which confirms previous speculations:
Southern Islands reportedly coming in November - Graphics - Graphics - Fudzilla

Southern Islands to arrive in November. Only entry level and mainstream GPUs at first, Q1 2011 being the target for the high-end segment.

I was thinking... given that the Mobility 5870 is actually derived from the desktop 5770 which a mainstream card, maybe we will get to see the Mobility 6870 a bit earlier then expected. Because it will most likely be based on another mainstream card which are set to arrive in November.

 

If this is based off of locations in vancouver area, then capilano and seymour should be worse than robson.

Wonder if they will have any events in vancouver for ati, if so maybe i'll check them out.

 

Would that really matter? Since SB will only be available in 2011 for the biggest part of the world, I doubt one would rush and buy a laptop with a 6870 in Q4 of 2010.

 

Going off the roadmaps from the OP, mobile Blackcomb looks like it matches up with the desktop Barts part which is scheduled for a 2011 launch....and the mobile parts that are scheduled near a November 2010 launch are Granville, Capilano, and Robson, that maybe based off the desktop Turks and Caicos parts.

My guess is Granville and Capilano will be the Mob. HD6700 and HD6600 and Robson will be the Mob. HD6400 (or HD6500).

ATI will be trying to steal the thunder from Nvidia's Fermi GF106 & GF108 launch.

 

Wait, so will the Southern Islands comprise low and midrange, with the Northern Islands being upper mid-range and higher? I'm confused now. I was assuming Southern Islands would be one series (Radeon 6xxx), with the Northern Islands it's own series (Radeon 7xxx). I'd hate to see ATi split a series between a partial refresh and a brand new GPU architecture design.

 

That's the general consensus right now. When people talk about Northern Island they're referring to the 28nm GPUs expected out in late 2011...like Q3/Q4 2011.

The 40m SI cards out in Oct./Nov. and Q1 2011 will be the 6000 series with a new uncore....and the 28nm NI will be a new series with new architecture for the GPU core.

 

Numerically, ATI has no choice but to give SI the 6k moniker. Think of it as what Nvidia had to do, going from the G80 core to G92, 8000 to 9000.

 

Come to think of it ATI has already used model numbers in the 7000, 8000 and 9000 range for old Radeon cards back in 2001 thru 2004 so a naming scheme change is likely imminent.

 

Nvidia is going down the same path. GeForce - Wikipedia, the free encyclopedia

They're using 1 less significant digit, but they're still overlapping with previous number sequences. Both companies should be fine. I don't think we'll be confusing old AGP cards with PCI express 3.0 cards any time soon.

 

ATI should do away with the numerical system, and just use the codenames as their proper designations.

A name like Blackcomb sounds pretty cool.

 

Blackcoal would be more correct if they keep on bumping the bus while staying @ 40nm

 

they areadly have some card with 6xxx, 7xxx 8xxx 9xxxx naming

how ever they are AGP and quite hard to find these days and they quere just 9xxx not hd 9xxx to no mather what you do there will be a diferenc i still ope they won,t change thier naming scheme again

 

Radeon HD 7870, Radeon HD 8870, Radeon HD 9870

sounds fine to me : /

 

Hmm sounds to me the 6xxx will be improvement but if you have 5xxx now, better to wait until ATi/Nvidia moves onto 28nm architecture before upgrading.

 

Whooo Vancouver

I have to buy one just because of its codename.

 

They could do like Nvidia and drop one of the digits in the name. The last digit is never used anyway.

So you could have Radeon M680, M660 and so on. Put the M in front for mobility just like Nvidia puts it at the end. I like things to be simple and elegant.

I know "Mobility Radeon 6895" sound pretty cool for a redneck, but I doubt this has any affect on the sales besides confusing costumers.

 

Looks like some high-end 6000 series cards will come out in 2010 after all:
New dual chip card comes from ATI - Graphics - Fudzilla

 

Ken from GenTech posted some DX11 numbers for Blackcomb XT.

http://forum.notebookreview.com/asu...53-smaller-version-g73jh-114.html#post6599350

http://forum.notebookreview.com/asu...53-smaller-version-g73jh-116.html#post6603508


If ATI is releasing benchmarks it could indicate we may see the new cards sooner than expected. Maybe a notebook with a Mob. HD6870 available in January/February instead of March/April.

 

I don't get it, are those benches for desktop or notebook ?

the GTX 460 is a middle-class card and thus the compared ATI card must also be a middle-class card.

If they are talking about mobile cards... the 460M is not even released yet and I thought that the 5870 already out perform it or at least is on par with it.

 

Keep in mind, those are DX11 only benches. However, I'm still confused as to how ATI has benchmark info on an unreleased, un-benchmarked nVidia part.

 

I really wanted to replace my 5870m's with one of the late 6xx0's but more and more I'm thinking I'll be better off waiting for 28nm... Which will also probably mean I'm going to end up needing to replace my system at that point... It's really too bad the 6xx0's didn't turn out more promising. Who knows though, they may end up having some really nice features and be well worth the swap or else may just be worth the slight improvement when the 7xx0's are released and the 6xx0 prices are wayyyyy down (if I don't want to replace my notebook that is).

 

Successive generations typically aren't monumental upgrades from top end GPUs from previous generations...

 

Did the 5970's ever make it to mobile cards? Will they ever?

If there's a chance of 6970m's then maybe that will be the large scale jump over the 5870m's that I'm (as well as many others) looking for.

I really don't think I've ever heard of the 5970's ever making it to a mobile platform so I'm not actually asking this with too much seriousness.

 

Especially when they're on the same node.

 

They won't be introducing dual GPUs on one card for notebooks anytime in the near future, unless it's extremely limited and customized.

 

Yeah, doesn't make much sense with limited space and TDPs which aren't as much of a concern in desktop rigs. It always begs the question why they didn't just release the same card with more ROPs/etc. if the TDP and space could handle it.

 

Ken sources the numbers he uses for the GTX 460M's DX11 benchmarks as coming from Nvidia.

http://forum.notebookreview.com/asu...g53-smaller-version-g73jh-75.html#post6500717

http://forum.notebookreview.com/asu...g53-smaller-version-g73jh-75.html#post6500967

http://forum.notebookreview.com/asu...g53-smaller-version-g73jh-77.html#post6503573

http://forum.notebookreview.com/asu...g53-smaller-version-g73jh-78.html#post6505322


The GTX 460M should beat the Mob. HD5870 in DX11 benchmarks but in DX10.1/9 games the Fermi card should be at or just under the Mob. HD5870. So a Blackcomb XT performing at the level of a GTX 460M in DX11 isn't entirely discouraging since it does indicate a performance increase. In fact, according to Nvidia's original marketing material, GTX 460M should perform twice as well as Mob. HD5870 in Stone Giant....that would mean Blackcomb XT is actually seeing a near 100% performance gain in that benchmark over ATI's current top end mobile card. :wink:

Seriously though there's no reason to get the coffin nails out just yet since there's still plenty to speculate over....for example the Blackcomb XT tests were run using Cat 10.4 and ATI isn't supposed to start using stream processors to assist tessellation until Cat 11.0.

 

All of your links go to the same thread, not a particular post. I must have missed where Ken quoted the source of the 460M benchmarks. Still, I'm as hopeful as can be for an arch built on the same node as the previous one.

 

They don't link to posts #745, #748, #762, & #776?

When I click them they do.


Also ATI's 3000 and 4000 series were on the same 55nm fabrication process and there was a nice performance boost between those two generations.

 

No, they all just take me to page 42 on laptop and iphone. My understanding is that 3xxx to 4xxx was major arch change. 6xxx is minor revision AFAIK. NI should be big jump with arch and node change, no?

 

They take me to the specific posts fine..

 

I understand now, thanx Phinagle.
In this case then both the 460M and the Blackcomb XT look like very nice cards. I can't wait to see some real benches not just "leaked" ones.

 

GTX 460M Ìú¶¨³ö£¡COMPUTEX 2010 ̨±±µçÄÔչй¶΢ÐÇGT760¹æ¸ñ£¡£¡  ×¼ÏµÍ³»ùµØ-±Ê¼Ç±¾DIY DIY±Ê¼Ç±¾ ±Ê¼Ç±¾×¼ÏµÍ³ ÉñÖÛ LG ·½Õý À¶Ìì ¹ã´ï Èʱ¦ MSI ΢ÐÇ Éñ´ï Ö¾ºÏ - Powered by PHPWind

use google translate! still makes little sense!
any one knows chinese?

 

The 4000 series wasn't so much a major arch change over the 3000 series (some tech remained the same, and some changed) but instead a major increase to the amount of hardware on the GPU. The original rumors for the 4000 series, however, also only expected minor changes over the 3000 series.

Mind you, none of that means that the 6000 series will be a repeat of the 4000 series, it's only to show that the potential exists to get a nice performance bump without a die shrink. GT200 would be another example since it originally rolled out on the same 65nm fab that was being used for the GeForce 8 and 9 series.


The 28nm GPUs are said to be a major architectural change....or maybe even two architectural changes since it's rumored ATI will be using both GF and TSMC to make them....but there's no guarantee we'll get an awesome performance increase out of those changes. I'll cite Fermi as evidence of that.

 

Going back to naming scheme changes, rumors are now spreading 'cross the internets that the ATI logo may be retired this year.

ATI logo finally gone? | KitGuru

[Update] AMD bald ohne ATi-Logo?


I guess it would make sense to change the GPU brand names to have a more recognizable association with Vision and Fusion....and it would have the added bonus of getting the AMD logo on many an Intel powered PC. :wink:

 

Seems to me the smarter idea would be to replace the weaker name with the stronger one. I.e., AMD should be replaced by ATI. Their CPU department doesn't seem that strong to me - I always see them as the "cheap Chinese knock-offs" of CPUs (well, these days - back in the day they were alright, but they've been so far behind for so long). Whereas in the graphics department, ATI is doing surprisingly well.

Why get rid of a label that's really making a name for itself? That would be like Elvis Pressley getting his name legally changed to Harold halfway through the most successful tour of his career!

 

The HD series all share a similar design method.

HD 2900 used a ringbus memory controller that was overly complex. They cut out much of the rubbish in the HD3870 but the numbers remained the same leaving a very small chip. The HD4870 then beefed the numbers up (and used GDDR5 to make up for the bandwidth).

The GTX460 is similar to the HD2 -> HD4 just by missing a step. They cut some of the compute stuff out, added in some extra units and really aimed it at the gamer market.

 

if 11.0 will nail the ferni to the coffin, it would be excellent.. 11.0 is about 2 months from now... 10.8 is comming out next week.. i'm quite sure the GTX460M comming out in 2 months so basically bye bye NVIDIA.

 

If you think there is going to be some wonder set of drivers for your card to make it suddenly better then you are living in la la land.

The drivers may help in pure tesselation benchmarks, however just like fermi suffers when its actually having to work hard on the rest of the scene then the gains in real games will be very small.

 

You have never played with a HEX core from AMD have you?

As for the mobile sector, AMD never had the cash to invest in propper mobile designs. With propper fusion products however that may change. You can't just replace one set of engineers with another.

The design process of making a CPU and GPU are VASTLY different, not just in how they work but even how the physical properties of the silicon are different.

 

I've read that Nvidia's attempt to become a strong player in the super-computer market has been rather successful. They've already won some important super-computer projects. But I guess this comes at a price in the gaming sector.

People don't see this, but most of their current troubles come from their attempt to turn a GPU into a CPU. This is actually why I would still like to seem them retain a strong position in the GPU market, I really appreciate their craziness in what they have tried to achieve.

On the other hand, when I bought this laptop Nvidia was charging 700-800 USD for a 9800M GTX. I remember I've wished so hard then for ATI to do a come back and whip them for their greediness. I might have wished too hard.

 

Yeah but they are funding their compute chips by shoveling stuff onto their standard GPU buyers that they don't need which make them run hot and cost too much.

The GTX 460 is the style of chip they should have released from the get go.

 

What exactly are you referring to about the silicon? If northern islands is made on GF's 28nm node, it will be the same silicon as any 28nm AMD CPU built on that same node.

 

I agree with this. As important as ATi and Nvidia love to consider getting out the foundation GPU, you can't expect to create crippled versions of that foundation product without suffering certain adequacies (power leakage resulting in increased power needs and heat output). The GTX 465 was this inferior product. The GTX 480 and 470 serve their purpose well, and justifiably so being the foundation. It always takes that mainstream performance product to really start to push a new series of GPUs, especially so with Nvidia products. Examples: Geforce 6600, 7600, 8800 GTS 320 MB & 8800 GT, 9600 GT, GTS 250, and of course the GTX 460. They all were/are hot sellers that resonated a very fine price/performance ratio. As much as I like ATi stuff right now, I am nostalgic for my first ever desktop custom build with an 8800 GTS 320 MB + Athlon x2 5600 + WinXP. Brings back wonderful memories with the gratuitous blue LED lighting and that toxicating "new" smell lol.

 

CPUs and GPUs aren't built using the same fabrication process and CPUs generally aren't manufactured on half-nodes (e.g. Bobcat/Ontario will be an exception) so you won't see a 28nm AMD CPU coming out of Glofo.

Furthermore, even at the same node size there are tons of different methods of manufacturing. For example, TSMC will have three separate 28nm processes all of which will be different from Glofo's 28nm.


Globalfoundries Starts on 22nm Fab & Announces First non-AMD Customer - AnandTech :: Your Source for Hardware Analysis and News

TSMC splits 28-nm high-k metal gate process into two versions

Blog - AMD could move all ATI GPUs to SOI at 32nm - here's why | bit-tech.net

Why not all ATI GPUs will go to SOI at 32nm :: TweakTown USA Edition

 

Looks like I incorrectly assumed all GPUs used SOI these days. Thanks for the correction.

 

You can use different dopings of wafers, different qualities of silicon, different types of transistor designs. All on the same node.

This is not a simple science.