ATi Mobility HD 6000 series Roadmap

Woo... I like the signature, daamit.

I just hope the 6000m series can fit into 15.6" notebook. 17" is just too big and heavy.

 

Nothing will change, in that regard.

Mobility 6800 for heavy gaming notebooks.

Mobility 6700/6600 for everyone else.

Mobility 6500/6400/6300 for the peasants.

 

The 6000M series consists of several models that will fit from models probably as low as 13.3" (or maybe even less, like the 11" Alienware).

 

More information from here:
????? BARTS????5800?????GTX460??????? - ????? - CHIPHELL - ??????,????????DIY????. - Powered by Discuz!
also wikipedia has updated its page on 6000 series specs:
Comparison of AMD graphics processing units - Wikipedia, the free encyclopedia

I can already see some contradictions, Wikipedia states that the BART XT is 120W TDP while the leaked slides point towards +150W.

If the Wikipedia TDPs are correct, then it is likely that a mobile version of the BART core will be the next M6870.

 

According to Wikipedia:

Interesting if this is true, as those shader counts do worry me a bit. I guess we'll find out soon enough.

 

The new arch will need only 640 (800SP / 5 x 4) shaders instead of 800 to do the same work
(4 shader arch versus 5 shader arch without performance drop)

So if they release a 960 shader part (which I think will be true) then you still get a decent performance boost (from 640 to 960)
TDP will likely be higher too.
A 96 bit combined with GDDR5 and 640 shaders would be THE notebook card.
20W would be achievable with 96bit.
Decent performance with very low TDPs and temps!

 

Sadly though the leaked slide gives SIMD counts which suggest Barts will still use a 5D architecture. 12 SIMD for Barts XT and 960 shaders would be blocks of 80....Barts Pro also is an 800 shader part which isn't possible with the SIMD blocks of 64 that you get from a 4D arch.

This news actually worries me as I'm now afraid AMD will stick mobile uers with a Barts Pro GPU and we won't get the fullest potential bump from this generation.

 

Yeah, I'm getting the feeling that Barts PRO will be the Mobility 6870. The XT won't make it, unless ATI goes > 75W.

 

I also fear that AMD will go with Barts PRO for the Mobility Radeon HD 6870... they could try to fit an underclocked Barts XT in the 100W limit that Nvidia created but that is not their style. However, I fear that Barts PRO will not be able to outperform a 336 shader GTX 485M, but we'll get a clearer picture once barts pro is available.

 

Well, if AMD's claims that even the Barts PRO is > GTX 460 768MB are true, that could still lead to a sizable performance lead over the Mobility 5870.

Overclocked, we could be nearing ~5850 levels.

In short, it should be in a close fight, with a 336 shader Nvidia part.

 

Back from vacation and have had time to read up on the leaked Barts slide....

General feelings around the rumor mill is it's probably a fake.

 

Welcome back Phinagle.
That's good to hear.

 

I've realized that I totally neglected to consider that along with switching from a 5-way arch to a 4-way arch, AMD could have also bumped up the number of VLIW units per SIMD block from 16 units per to 20 units per in order to keep the number of stream processors per SIMD at 80....that would make an 800 shader part possible.

Those 800 NI stream processors could potentially be the equivalent of 1000 Evergreen stream processors....and a Barts XT with 960 SPU would have a relative shader potential just above the desktop HD5830.

 

For mobile, if you already have a Nvidia 4xx or AMD HD5xxx, skip the HD6xxx. I'd wait out a year for the 32nm or 28nm architecture. That I would think should boost the mobile GPU potential a lot more than reorganizing the architecture, for both power and thermal.

Anyways, that's how I see it. Unless games, applications all of a sudden exceed what the HD4870M and GTX 280M can provide, save your money for the next gen beast!

 

i'm still on my 48xx gen card i haven,t hit the limit of my dual card setup except when benching so i'll likely skip the 6800 gen to maybe i'll try it on desk but not on a new laptop unless they go back to 2.1 type III so i can get one in my m17

 

From the looks of the graph on the first post it looks to just be a refresh of the 5000 series.

 

For discreet graphic cards it's only 28nm. That's what I am also aiming for.

 

While I know the 6XXX series isnt a next gen GPU, shouldnt refreshes be 5X50 instead of a totally different series? Just my opinion. I know its all a marketing scheme to get people to buy hardware, but IMHO the 6XXX series should just be a 50 added and not a new seires.

 

Maybe partially, or maybe not at all.

In the 10.8 drivers several code-names were leaked including Blackcomb, Whistler and Seymour but no mention is made of the Robson, Capilano and Granville parts.

One possible explanation for this is that Robson, Capilano, and Granville are just refreshed 5000 parts but another possibility is that the three original parts slated for a 2H 2010 launch were scrapped and Blackcomb, Whistler and Seymour were bumped up. The second explanation is somewhat supported by the apparent earlier launch of the Turks, Caicos and Barts desktop parts which, according to the roadmap, were due for a late 2010/early 2011 launch but are now being launched in October.

Barts, Turks and Caicos are the most likely desktop counterparts to Blackcomb, Whistler and Seymour.

 

Here we go: Radeon HD 6000 launch in late October - Graphics - Graphics - Fudzilla

Due to be released in October 2010, just one month from now... all shall be revealed.

 

Catch up with the news, it's widely thought to be more than just a simple refresh (even if the opposite was the consensus a couple months ago) and therefore its 6xxx name shouldn't be discussed. Nothing in common with nvidia going from Geforce 8 to 9.

 

thanks for the link!

 

[Rumour] AMD Barts XT (HD 6770) faster than HD 5850 by VR-Zone.com

The rumors go: (Desktop) HD5870 > Barts XT > HD5850 > Barts Pro > GTX 460

If true, even a mobile card based on Barts Pro should be able to handle the return fire from a fabled GF104 based GTX 485M.

 

Everything depends on how these are priced compared to current generation chips. The 5770 is nice but a bit too weak for heavy 1080p gaming. The 5850 is expensive. There are some good deals on GTX 460 1gb, but getting a card priced between the 5770 & 5850 when they were released and outperforming these would be awesome.

 

You'll just have to hope AMD continues to price their GPU generously. Would be nice to have something that stomps all over the GTX 460 for either same price or less huh

 

5850 was expensive? It actually had a bit more buy value than the 5870; the 5850 can easily be overclocked to be on par with the 5870.

 

Yeah and 260€ or more is called expensive in my price scale, sorry to disappoint.

 

If the Mob. 6870 can be overclocked to = ~5850, Nvidia just can't match that with anything currently available.

 

I'm comparing the price of the 5850 to that of the 5870. I wasn't referring to your connotative meaning of expensive graphics cards.

 

OK, looks like we have the real specs for BARTS PRO and BARTS XT now:

source

I think BARTS XT in a notebook would be overkill, but even BARTS PRO looks like it can beat a GTX 460. If you look at the TDP BARTS PRO looks like the way to go and the Mobility Radeon HD 6870 would still have nearly the same TDP as the Mobility Radeon HD 5870. Although Barts XT could fit a 100W TDP with some downclocking, I don't think AMD will go this road simply because they don't need to in order to beat Nvidia.

 

According to the chart above, doesn't that mean the upcoming lineup doesn't have anything yet to unseat ATI's current 5870m?

If so, then I suppose my purchase of a proper gaming laptop right around now won't be bad timing.

 

The desktop 5870 is what you're looking at there. The 5770 is the mobility 5870.

The 6750 by that chart is much stronger than the 5770.

 

Ah, I see. Then again one can't wait forever when deciding to upgrade. Just as well 5870m or the crossfire solution will still last a good 3-5 years even with the 6 series coming out.

 

We haven't seen what a fully enabled 384 core GF104 can do...it might come close to a Barts PRO.

If a notebook can be built to cool a Barts XT then it's not overkill.

Barts Pro does look like the more likely candidate though...with a 4-way setup 1120 cores would be close to the count of a desktop HD5850's 1440 shaders in 5-way. Texture Units are lower than on Cypress though so performance may suffer a bit.

AMD could price Barts XT directly against GTX 460 but it's not necessarily in their best intrest to do so.

GF104 has a die size that's just larger than Cypress (desktop HD5870) making Nvidia's GPU more expensive to produce....which means if Nvidia is able to make a profit off GTX 460 at it's current price then AMD should have had no problem lowering it's price on Cypress GPUs to compete. The Barts parts, while larger than Juniper (desktop HD5700), will be even smaller than Cypress and GF104, and therefore even cheaper still to produce.

However just because Barts XT could potentially be priced against GTX 460's manufacturing cost, AMD has every right to price an HD6770 at it's performance level. If for desktops, HD6770 performs between an HD5850 and HD5870 then AMD probably should put the price in that range and let one of it's lower model cards compete against Nvidia's GF104. With the smaller die size on Barts, AMD's profits will be even better than they were with Cypress at those price points....and for a company that's still heavily in debt, more money is the better than trying to force your competition out of business.

 

Te real issue, for us, is that 336 cores and 384 cores will likely require significant downclocking, to go mobile.

 

Exactly, and it's a bad sign that the even GTX 470M is downclocked. In order to fit a stock clocked GTX 460 / 256 bit into a 80W TDP you need to halve the original TDP. Nvidia can only manage this with heavy underclocking... even with the GF104. It would be interesting to know which exact TDP the GTX 460M has. The GTS 450 has a TDP of 106W. If GTX 460M has a 75W TDP that would mean a 30% reduction. If it has a 60W TDP it would be a 40% reduction. 40% reduction would mean the GTX 460 / 256 bit could go mobile at a TDP of 90W. A fully enabled GF104 has 14% more shaders, so that would mean at least 14% higher TDP (most likely more) so that would mean just above 100W. A stock or slightly downlocked mobile 384 shader GF104 with a 100W TDP would be great, but if it can be outperformed by a 75W TDP AMD card most manufacturers will not go with the 100W GPU since they would have to adapt their design, which in return means that only a few of those cards would be produced (or only a few would produce those cards) which means the price would be just as high as it is with the GTX 480M. Now people only buy the GTX 480M because they simply want the fastest card. If AMD comes up with something faster at a lower price, Nvidia has a problem.

But the GF106 architecture is more energy efficient than the GF104 architecture. The fact that Barts PRO has the same TDP as the GTS 450, which is a crappy card compared to the HD 6750, basically speaks for itself.

 

I'd imagine that they'll also have higher yields from before given that this is their second go at the 40nm process. If the rumors are to be believed, Nvidia has demonstrated the dangers of not knowing your process that well.

 

Has anyone looks one wikipedia lately ? At Nvidia GPU comparison chart ... it says that the 470M has more GFLOPs than the 480M.

Comparison of Nvidia graphics processing units - Wikipedia, the free encyclopedia

Does this mean that the 470M is better than the 480M ?

 

Definitely true.

It'll be AMD's third go at 40nm...though there weren't many RV740s produced because of yield problems.

Mob. HD5870 will definitely last you but if a Barts PRO does perform in the GTX460 to HD5850 range then a single mobile Blackcomb may get very close to Crossfire Mob. HD5870 levels. If the rumors turn out true, Mob. 6000 will give us a bump like we got from Mob. 3000 to Mob. 4000 but didn't get with Mob. 5000.

If a buyer can hold off until February/March that kind of bump would be worth the wait....but always buy now when you need to.

 

Those numbers for GTX 470M are for the maximum processing power in the best case scenario, but being superscalar means GF104 could have a harder time reaching those maximums than GF100 does.

 

So what's the consensus? Some believe the 6000 series will be a big step from 5000, others say not so much.

 

I'd say it will be big enough to beat the 400M series and also super cheap compared to those. Dunno if the tessellation part will be better than with Nvidia GPU's though.
The real jump (10-40x) in GPU performance is expected in 2 years from now

 

from what im reading the 6700s should be on par or faster then the 5800's so could be a substancial jump especially on a 6870.

 

Desktop performance doesn't always translate directly to laptop performance. Although I think it is likely that AMD will to base the top end 68xxM parts on the 67xx desktop parts... in which case it will be quite the improvement (but how are they going to cut down that 118W TDP? less memory bandwidth and lower clock).

 

They just lower the clocks.

 

like on the 4xxx serie the original 4870 has a quite high TDP but in my laptop they are not much higher then 45-65 without to much clock cut back

 

Makes me wonder if getting a laptop with the 5650 GPU right now would be a bad investment.

 

depends on if you need it now or not. SB and AMD 6000 series does have my curiosity up

 

I'd like to have a laptop now (been 7 months since my last one's mobo died). But having waited so long I wonder if it's worth it to wait a little longer.

The 4820TG with i5-460m and 5650 for $800 is a damn good deal.

 

that is is, and if it suits you for a year or 2 id be comfortable getting it and then chinging up later on