New rumor about ATI Southern Islands

[flopper]

if accurate, seems good.
the 6970 will be double the shaders so...

[Sn0wm@n]

if accurate, seems good.
the 6970 will be double the shaders so...

exactly .. and if rumors of a single gpu top dog card then we can only imagine the power :O

[slaveondope]

3DV跑分 P14872 X6549 how is that compared to the 5750 ??? or even 5870

Thats on par with a 5850.

[Vardant]

[RPGWiZaRD]

So yea looks like 6870 will be slightly ahead of HD5850 (X7500 vs X7300) which is in-line with my expectations. Now I really hope for some generous launch pricing and HD6870 could be like ATI's 8800GT. Cayman will be interesting for people requiring more GPU power, I would expect HD6950 to be around same as HD5870 performance and HD6970 a bit above GTX 480 performance (maybe 10 ~ 20%). Then dual-gpu version will be alone up there wondering where its competitors are.

Shame that Fermi's been such a weak response from nvidia, cuz of this we will most likely not see any generous pricing especially for Cayman and Antilles. I can already imagine that HD6970 being priced 450~$499 and dual-gpu at like 699~$749...

I think optimal prices for AMD taking into account manufacturer costs and compared to today's products would be sth like:

6850: $199 (GTX 460 1GB or slightly above performance for 20~$30 less)
6870: $239 (slightly above HD5850 performance for 20~$40 less)
6950: $329 (HD5870 performance for 30~$70 less)
6970: $449 (above GTX480 perf for ~$50 less)
6990: $699

I know this is expensier compared to Cypress ($299 and $399 launch pricing) but they would still look rather aggressively priced compared to the other products in performance comparision and no1 could blame AMD for being too greedy with such prices either as the products are simply that good to warrant the price.

[Ztx]

The test system of the charts you show are i7 965 @3.87 ghz, that is why you have much higher results.
The vantage preview of the HD6800 was using an i7 @ 2.93ghz resulted in X6549 for the HD6850 and X7538 for the HD6870
Here are results from a i7 975 @ 3.33 (stock)

[kaktus1907]

HD 6870 and HD 6850 performance figures
http://itbbs.pconline.com.cn/diy/11962049.html

WTH ATi logo ?

[Pantsu]

Engineering samples most likely don't warrant instant change to new logo stickers, the card might not even come with that fan design.

[tajoh111]

The test system of the charts you show are i7 965 @3.87 ghz, that is why you have much higher results.
The vantage preview of the HD6800 was using an i7 @ 2.93ghz resulted in X6549 for the HD6850 and X7538 for the HD6870
Here are results from a i7 975 @ 3.33 (stock)

Thats using ancient drivers and CPU has little effect in Vantage extreme.

With more recent drivers, the card 5870 scores around 9 thousand.

http://www.fudzilla.com/reviews/item...dition?start=7

3dmark 06 though is crazy effected.

[spursindonesia]

GTX 460 1GB gets X6500, GTX 470 gets X7700, HD 5850 gets X7300 and HD 5870 gets X9000. That looks bad.

Already makin judgement without knowing the price first ? Power consumption and noise ? LOL

The resegmenting will be confusing & rather misleading, if there is any renaming then it will suck no doubt, but if HD 6870 & HD 6850 (?) are priced at US$ 229 & US$ 179 at launch and decreasing as time goes with supply abundance, i think MANY peep will praise these little mighty value cards.

Courtesy of B3D poster :

http://forum.beyond3d.com/showpost.php?p=1480677&postcount=2866

OCClub   TechGage1  "2     Xbit    PConline
       i7-920   i7-975     "      "       i7-940
        @3.00+   @3.33      "      "       @2.93
        9.9      9.10       10.3   10.7    ???
5870:  8246     8213       9041   8870    
5850:  6590     6566
5770:           4311
480:                       9237   9828
470:                              7844
6870:                                     7538
6850:                                     6549

Not bad, not bad at all IF Bart really is a 960 SP chip.

[Vardant]

Already makin judgement without knowing the price first ? Power consumption and noise ? LOL

I was talking about the naming scheme, if that wasn't obvious.

[Olivon]

[tajoh111]

I was talking about the naming scheme, if that wasn't obvious.

If the new naming scheme is true, AMD obviously renamed the lineup to justify higher pricing. A 6770 for 249 would be hard to justify to the public a 6870 at that price sounds like a good deal. Pricing is going to be jacked up substantially if AMD is going to name its 6870 to 6970.

This is deceptive marketing. Considering the price of the 5770 and the 4770, the 6870(6770) is going to be one of the worst generations ever, as their price increase was the most expensive ever, but at the same time, the least amount of improvement(the 5770 was a lot closre to the 4870, the 4770 beat the 3870).

We might not know pricing, but we all know that AMD is going to charge more than 200 dollar pricing for these cards, even the 6850. I lost confidence in AMD pricing after they charged 239 dollars for the 5830 and will lose even more faith if the naming scheme is true which seems certain at this point.

[~CS~]

I think this generation is going to be pretty cheap IMHO,market is saturated with DX11 cards at this point and there is plenty of competition.


Cypress is smaller then Fermi ,but still quite large compared to rv770 ,I think barts will be about the size of rv770 or a bit smaller. AMD knows they will need to compete on perf/$ basis ,it will be hard to do with cypress which is too large and juniper which is too underpowered. I am still not convinced about the naming,but honestly it doesn't matter to me.

[spursindonesia]

LOL, Bart will be a great x7xx chip if 32 nm process wasn't canned, but happens, and there's just so much you can do on the same process. Increasing absolute performance while increasing performance/$, performance/die area, and performance/watt aka overall efficiency at the same time, increasing product value for costumer somewhat (a US$ 229 HD 6870 Bart that's faster than 299 US$ HD 5850 of today is indeed giving more value for costumers, right ?) while making good progress for the company's bottomline altogether, all achieved using the same TSMC's troubled 40 nm process, and you guys already shooting so much flak to AMD ??

[Chickenfeed]

Considering with current drivers Cypress XT can do just shy of 10k GPU score @ 1ghz in Vantage extreme ( with a highly clocked i7 ) x7000 total isn't too great of a showing if that is infact the 6870. That does place it in 5850/470 territory. Now the new gpus likely do better with DX11 content ( still not relavent yet in my opinion but it will be eventually ) but current DX9/10 performance is what most will care about. I could care less about naming schemes but 6870 as the mid range part would be a bit of a marketing fail as far as im concerned.

The DX11 varriant of 3Dmark should make for a more interesting comparsion once it arrives.

[Zloyd]

ooh nice so now AMD have some video playback features .... SWEET that was one thing lots of nvidia users like the better video playback this should negate that now.

[Chumbucket843]

LOL, Bart will be a great x7xx chip if 32 nm process wasn't canned, but happens, and there's just so much you can do on the same process. Increasing absolute performance while increasing performance/$, performance/die area, and performance/watt aka overall efficiency at the same time, increasing product value for costumer somewhat (a US$ 229 HD 6870 Bart that's faster than 299 US$ HD 5850 of today is indeed giving more value for costumers, right ?) while making good progress for the company's bottomline altogether, all achieved using the same TSMC's troubled 40 nm process, and you guys already shooting so much flak to AMD ??

i'll tell you why AMD is getting flak.

their shader architecture has not changed much in 3 years. they barely had any innovations in evergreen, and it was only 50% faster than last gen. this is not an exciting upgrade for most even with dx11.

TSMC does not have a troubled 40nm process any longer. the blame game is very fun but failures are not always caused by uncontrollable events aka excuses. TSMC does a good job for the most part. most of the people who criticize them dont know what they are talking about. furthermore, there is a lot you can do on an existing process. AMD can improve their architecture and physical design on 40nm by a nice margin.

faster gpu's are demanded by consumers. someone must supply this good. in this case most are expecting AMD to deliver due to nvidia's hot mess on a pcie slot.

[Manicdan]

AMD knows they will need to compete on perf/$ basis ,it will be hard to do with cypress which is too large and juniper which is too underpowered.

per $ is a choice made by a company, per mm2 is what generally drives profit margins. amd already has great perf per mm2, so why would they be making less money than nvidia per card todays pricing? comparing the size of 200$ cards, amd is much smaller and perf differences are close enough. its safe to assume they already are doing great with current $ per mm2 or perf/$.

[~CS~]

per $ is a choice made by a company, per mm2 is what generally drives profit margins. amd already has great perf per mm2, so why would they be making less money than nvidia per card todays pricing? comparing the size of 200$ cards, amd is much smaller and perf differences are close enough. its safe to assume they already are doing great with current $ per mm2 or perf/$.

I am not sure what you are trying to say.If Barts performs as good as GF104 ,do you expect it to cost more then GF104 ? I am not saying it is not possible,but not quite sure AMD brand can demand a premium over NV brand,they are getting closer,but not quite there. Perf/mm2 is irrelevant from consumer's point of view.

[spursindonesia]

i'll tell you why AMD is getting flak.

their shader architecture has not changed much in 3 years. they barely had any innovations in evergreen, and it was only 50% faster than last gen. this is not an exciting upgrade for most even with dx11.

Their shader arch is still yielding added performance with unit expansion, whether that's considered sufficient enough or not that's certainly a relative, debatable thing for sure. I think AMD is just trying to best nVidia's GT200 with their Evergreen gen, DX 11 is basically just a checker mark feature for this generation. Evergreen is relatively cheap to make compared to nVidia's current DX 11 generation, while absolutely faster and more advanced compared to nVidia's past generation.

For all its achievement, AMD Graphic division is certainly still lightweight compared to nVidia sizewise & resourcewise, for them being able to eventually bypassing nVidia after R600 debacle, that's quite an achievement in itself. Exciting or not, Evergreen did and does dominate DX 11 scenery for the past year (Fermi late entrance is certainly helpful), Northern Islands on 40 nm is basically a contingency product out of TSMC canceled 32 nm process, which IMHO affected AMD worse compared to its competitor (fact that AMD is always in front in using new process node and they've been very good in hitting their R&D timeline target lately).

With NI, they should and IMHO will still give added value to the costumers, how much and how far, that's a relative thing to judge, but one thing for sure, G92 and its nth incarnates are making tons & tons of money for nVidia, i don't think AMD's decision to resegmenting its new gen graphic cards, reevaluating & reworking the marketing strategy, can be deemed as a totally bad thing and harmful to the customers. Yes, it will cause some confusion, and some can consider this as a misleading marketing tactic, but i think the bottomline is giving more absolute value to the customer while being profitable in the process, that's all that matter. Anything else is just forum noise from geeks & nerds, green and red rose colored notwithstanding.

Though, if AMD dares to rebrand RV 840 from HD 5770 to HD 6770, they WILL get some flak from me personally, no doubt about that.

TSMC does not have a troubled 40nm process any longer. the blame game is very fun but failures are not always caused by uncontrollable events aka excuses. TSMC does a good job for the most part. most of the people who criticize them dont know what they are talking about. furthermore, there is a lot you can do on an existing process. AMD can improve their architecture and physical design on 40nm by a nice margin.

How much margin can be deemed nice enough so it won't be called a failure ? Perhaps TSMC 40 nm process is much better that what was available in the past, but troubled isn't just related to how well a process node is yielding. i certainly haven't seen the kinda competitive drive like in 55 nm old days, even after GF 104 is introduced that offers a tremendous value and give such a good fit against Cypress, AMD decided to rather enjoy maximal profit rather than getting a bit agressive pricewise to gain more marketshare.

That signals supply constrained situation, and considering Cypress decent yield, basically AMD doesn't get much enough 40 nm wafers from TSMC to optimize their profit while maximizing the marketshare as well like what RV 770 has done for them in the past. That shows a trouble for a process that should be maturing by now and costing much less if its really trouble free.

Northern Islands seems to be a reengineering of AMD graphic mArch, if it really delivers the added efficiency as leaked, while costing less to the customers for the same or even more performance, i think it's a job well done and we should be grateful as customers. Designing graphic chip isn't an easy job, Fermi designers can attest to that statement, hehehe.

faster gpu's are demanded by consumers. someone must supply this good. in this case most are expecting AMD to deliver due to nvidia's hot mess on a pcie slot.

And i think they WILL deliver, whether the amount is good enough or in the way that satisfy people egos, that's really a matter of personal judgement.

per $ is a choice made by a company, per mm2 is what generally drives profit margins. amd already has great perf per mm2, so why would they be making less money than nvidia per card todays pricing? comparing the size of 200$ cards, amd is much smaller and perf differences are close enough. its safe to assume they already are doing great with current $ per mm2 or perf/$.

Cypress is great against GF 100, but aggressively priced GF 104 is really a thorn on AMD's side. Yes Cypress is still (debatable) more profitable even compared to GF 104, but the margin is much thinner if AMD decides to get aggressive against GF 104 marketshare offensive from nVidia, matching the price toe to toe and fighting for added marketshare.

Using a 220-240 mm^2 (if the 960 SP rumor is true) Bart will certainly matches GF 104 better from AMD's bottomline perspective compared to using a 334^2 mm Cypress, while still being supply constrained by TSMC. With its deep pocket, nVidia can be aggressive and accept thinner margin or slight loss for maintaining strategic marketshare position.

A small efficient yet good performing Bart and Northern Islands family overall will certainly give AMD more flexibility in their strategy for optimising both profit and marketshare against current nVidia's offerings.

LOL, what a long post to make.

[Manicdan]

A small efficient yet good performing Bart and Northern Islands family overall will certainly give AMD more flexibility in their strategy for optimising both profit and marketshare against current nVidia's offerings.

i fully agree that it will give them more flexibility. but i think people are a little too negative about the current price to perf of cypress. since the chip is only a portion of the cards cost, while the rest is all a cost (board, vrm, ram, box and shipping), i think that a 10% price drop could mean 30% less profit, but might not get anywhere near 30% more sales. the consumer side of things only see a few little factors and its hard to base why just from those.

just to point out, i HATE the current price to perf ratio. for the last year is was WORSE than 4000s/200s as long as you were not playing any dx11 games. 180$ for my 4850xfire, and even in the next few months it seems like 180$ still wont bring any noticeable difference (except for power consumption)

[Chumbucket843]

per $ is a choice made by a company, per mm2 is what generally drives profit margins. amd already has great perf per mm2, so why would they be making less money than nvidia per card todays pricing? comparing the size of 200$ cards, amd is much smaller and perf differences are close enough. its safe to assume they already are doing great with current $ per mm2 or perf/$.

there are two elements to a chip's cost. manufacturing is extensively talked about but dont forget that R&D expenses are ~$900 million per year for nvidia and i would assume ATi would be in that range. this is ~30% of nvidia's revenue and ~60% is manufacturing. the rest is business administration and marketing.

[Sly Fox]

I hope the renaming rumors aren't true. ATI's been relatively good these last few years about not arbitrarily misleading customers through names. One of the "ethical" advantages they've had over Nvidia. Seems kind of rash to just start going down that slippery slope.

[bamtan2]

lots of smoke being blown here based on RUMORS

RUMORS!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!

the only thing we KNOW IMO is that naming something 6870 when it is worse than 5870 is a STUPID IDEA. so why expect them to do it.