New rumor about ATI Southern Islands

[mindfury]

http://news.ati-forum.de/index.php/n...fikkartenserie

The new graphics card generation, the direct successor of the current HD5000er series, is due out later this year, in this regard we have received new information from an industry-related source. So shall have already come in the middle of next October, the first offshoot of the new HD6000er series and the code name "bear beard."

The small branch is located in view of the performance between a current HD5770 and HD5870. In addition comes an eight-layer PCBs, 1GB GDDR5 memory with 5Gbps and 2x6pin power connectors are used. The TDP of the new pixels accelerator should be slightly higher than 150W and when connections are 2xDVI, HDMI, 2xMini-DP, with the support of Eyefinity 4, available. In addition, the chip of the "beard" Pinkomaptibel with the PCB is the HD5800, which will drive down production costs significantly. The above specifications refer to "Bart XT.

More information is available to us, relate to the GPU "Cayment" which will appear in November and October. This information confirmed our of 16 July this year. This GPU will have a 10-layer PCBs and large 1GB GDDR5 memory with 6Gbps. The graphics card needs to power a 6pin and 8pin power connector, the TDP comes to less than 300W. The connections are 2xDVI, HDMI, and 2xMini-DP.

Also the code name for the upcoming dual-graphics card of the new generation is already known to us. This carries the code name "Antilles" and will be AMD's new spearhead. We keep you informed.

[Mats]

LOL lost in translation! I like the name Bear beard tho! For those who doesn't know, the german word for beard is Bart.

[Dan_c]

Makes my english look fantastic!

Finger's up tho, keep'em comin'

[Manicdan]

im so lost,
perf between 5770 and 5870 (might aswell be 5850)
power consumption close to a 5850 aswell

what the point? its going to be cheaper to produce is all that i see, so did watts per mm2 go up even higher somehow??!?!?!

[jam2k]

It looks like a GTX460 killer to me

[Frontl1ne]

The small branch is located in view of the performance between a current HD5770 and HD5870. In addition comes an eight-layer PCBs, 1GB GDDR5 memory with 5Gbps and 2x6pin power connectors are used. The TDP of the new pixels accelerator should be slightly higher than 150W and when connections are 2xDVI, HDMI, 2xMini-DP, with the support of Eyefinity 4, available. In addition, the chip of the "beard" Pinkomaptibel with the PCB is the HD5800, which will drive down production costs significantly. The above specifications refer to "Bart XT.

This sounds like either the HD6770 or the HD6850, if ATI keeps its current naming scheme. Most likely a 6850 since it uses two 6pins. The perf difference between a 5850 and a 5870 is roughly 10 - 15 %, so we'd only be seeing that much of an improvement for the HD6xxx series.

More information is available to us, relate to the GPU "Cayment" which will appear in November and October. This information confirmed our of 16 July this year. This GPU will have a 10-layer PCBs and large 1GB GDDR5 memory with 6Gbps. The graphics card needs to power a 6pin and 8pin power connector, the TDP comes to less than 300W. The connections are 2xDVI, HDMI, and 2xMini-DP.

That sounds like the HD6870 then. 1500MHz mem clock and 6+8 pins.

Doesn't really make sense, if the performance increase was only about 15% I don't think ATI would be bother releasing it as a next gen product.

[Florinmocanu]

Looks interesting. Basically they are going for 6770/6870/6970.

[madcho]

1200mhz mem clock ? wtf.

6/4 = 1.5 hey !!!

[Frontl1ne]

1200mhz mem clock ? wtf.

6/4 = 1.5 hey !!!

my bad

[Calmatory]

Doesn't really make sense, if the performance increase was only about 15% I don't think ATI would be bother releasing it as a next gen product.

What else would you expect? The new generation is still based on 40 nm process, so there isn't much they can do. They have the ~same transistor budget, they have to base their architecture on Evergreen and they only had approx 12 months to come this far. So really, what else can people expect?

They need to do whatever they can to minimize the costs, and if it is working under the limitations of pin compatibility with current gen, then let it be so. The chips will be bigger than current ones for sure.

What I believe is that this news is just to mislead the competition. Now it just gives an impression that it's nothing interesting, just a similar performing updated silicon with possibly lower price. But given the old rumours of focusing on efficiency of the chip, given that the 40 nm process should be pretty much mastered(I know.. ) by TSMC and those of AMD who work with them, they could achieve somewhat better perf/mm² for the uarch and somewhat better clockspeeds while keeping same yields.

I have hard time believing in "just" an updated silicon to minimize manufacturing costs. Then again, there isn't much what Nvidia could come up with in 40nm so AMD isn't really in position where it would desperately need more speed.. So the next real round will be fought at 28nm and thats where both companies focus.

[spursindonesia]

hmm, may i speculate ? Bart would be a 1280 SP (320*4) 16 ROP 128 bit chip 240-260 mm^2 performance ~HD 5850, Cayman would be a 1920 SP (480*4) 32 ROP 256 bit chip 390-420 mm^2 performance ~GF100 512 SP+, and dual Cayman power consumption would be very, very close to PCIE standard limit of 300 w, unlike HD 5970 TDP which is rather conservatively rated, or would simply be underclocked further compared to its older dual chips kamerad. This speculation is reasonably explaining the rumored TDP of the chips reported in the article.

[Panther_Seraphi]

Wasnt the HD4xxx series surrounded by the same rumours?

Oh it will only have 640 stream processors as the flag ship as its double what they have been doing the past two generations

HD2900 = 320
HD3870 = 320
Assumed HD4870 = 640
Actual HD4870 = 800



So i say lets wait a little bit longer as i can see there being either a memory boost to the chip e.g 384 or maybe a return to 512 ring bus?

Or a dramatic improvement in tessalation in a response to nvidias domination

[Manicdan]

was the assumption for the 4870 to go with 1+3, and they did 1+4?

theres really only 2 things to look at
perf per watt = efficiency
perf per mm2 = costs

it seems that only 1 of those is really changing by much. which isnt bad considering its on the same process and we cant really expect too much with perf/watt (unless its like nvidia where they chopped off the server stuff to shave power for the gf104)

[ajaidev]

Wasnt the HD4xxx series surrounded by the same rumours?

Oh it will only have 640 stream processors as the flag ship as its double what they have been doing the past two generations

HD2900 = 320
HD3870 = 320
Assumed HD4870 = 640
Actual HD4870 = 800



So i say lets wait a little bit longer as i can see there being either a memory boost to the chip e.g 384 or maybe a return to 512 ring bus?

Or a dramatic improvement in tessalation in a response to nvidias domination

Yes 3870 and 4870 were both in 55nm but then again 38xx was a copy of 29xx and there was not lot changed it was a simple die shrunk where as 48xx was a redesign.

As it is the 58xx are not copies of the 48xx and this means that the 58xx were also redesigns, the 68xx has to be radically redesigned from the 58xx to achieve any spectacular results but at the same nm it will be rather hard to do.

I agree that tessellation might be improved quite a bit.

[Chumbucket843]

What else would you expect? The new generation is still based on 40 nm process, so there isn't much they can do. They have the ~same transistor budget, they have to base their architecture on Evergreen and they only had approx 12 months to come this far. So really, what else can people expect?

They need to do whatever they can to minimize the costs, and if it is working under the limitations of pin compatibility with current gen, then let it be so. The chips will be bigger than current ones for sure.

What I believe is that this news is just to mislead the competition. Now it just gives an impression that it's nothing interesting, just a similar performing updated silicon with possibly lower price. But given the old rumours of focusing on efficiency of the chip, given that the 40 nm process should be pretty much mastered(I know.. ) by TSMC and those of AMD who work with them, they could achieve somewhat better perf/mm² for the uarch and somewhat better clockspeeds while keeping same yields.

I have hard time believing in "just" an updated silicon to minimize manufacturing costs. Then again, there isn't much what Nvidia could come up with in 40nm so AMD isn't really in position where it would desperately need more speed.. So the next real round will be fought at 28nm and thats where both companies focus.

not all transistors are created equal. there is a lot of room for improvements.

[Panther_Seraphi]

Yes 3870 and 4870 were both in 55nm but then again 38xx was a copy of 29xx and there was not lot changed it was a simple die shrunk where as 48xx was a redesign.

I thought that the HD2-4xxx were basic reiterations with improvements and only the HD5xxx was the next complete redesign?

[Manicdan]

2900 was a big change
3870 was a shrink
4870 was similar with more SPs packed together
5800s were just doubled

in theory it sounds like SI is about the same as 3870 to 4870 as far as architecture changes, not sure if the improvements will be as great, but just amount of things changing it seems like a similar comparison.

[Helloworld_98]

if the TDP of Cayment is correct and the performance scales fairly well per watt then it will be very powerful

but also very hot and noisy :P

[ElSel10]

Looks interesting. Basically they are going for 6770/6870/6970.

And not a very big performance boost compared to the cards they will replace. For a new generation, HD6000 really seems like it will be underwhelming. The good part is that hopefully they will drive down prices of HD5000 cards.

[Panther_Seraphi]

Well ill be looking forward to seeing what teh HD6xxx is like. I can see them being like the HD3xxx was to the HD2xxx a short but sweet update

FWIW
HD2900 released May 14 2007
HD 3870 released Nov 19 2007

[Manicdan]

that price drop was from process changes, not architecture.

AMD already can drop prices for 5800s by a massive amount, but have decided not too.
if this new arch is going to reduce the chip size but leave the same performance, I WILL NOT BE EXPECTING A PRICE DROP, unless nvidia lowers first, why would AMD care to get more market share when they can just make killer margins.

[PaganII]

Hope it has 3 timers like the Sapphire 5770 Flex.

[Manicdan]

Hope it has 3 timers like the Sapphire 5770 Flex.

with the desire for multi monitor support, but lack of adoption, i think that would be a great idea to include it globally

[ajaidev]

I thought that the HD2-4xxx were basic reiterations with improvements and only the HD5xxx was the next complete redesign?

2900 was a big change
3870 was a shrink
4870 was similar with more SPs packed together
5800s were just doubled

in theory it sounds like SI is about the same as 3870 to 4870 as far as architecture changes, not sure if the improvements will be as great, but just amount of things changing it seems like a similar comparison.

By redesign i mean space redesign not architecture. The 48xx and 58xx are reiterations of 38xx with added features but they were more extensive space redesigns than the 28xx to 38xx.

To achieve stealer performance compared to the 58xx the GPU has to be space redesigned again and in limited space "due to 55nm" a dynamic addition cant be made.

5970's has 3200 shaders and TDP is <300w so this new wonder single chip that is suppose to have 300w TDP seems to be well using different type of shaders because combining 3000-3200 shaders under on chip on 40nm will be very hard to do.

[Sentential]

Well this appears to explain alot but I am sure that the ATi fans wont be happy by what this means. It appears we are having a de-evolution of GFX cards from higher complexity cards to lower cost cards. Case in point is highlighted by this info.

Example:

GTX480
--->GTX460 (lower cost, equivalent peformance high end is GF104X2)

HD5870
--->HD6770 (lower cost, equivalent peformance high end is a dual card)

The real question now is this, will their card beat nVidia's 384 shader GF104?