New rumor about ATI Southern Islands

[Manicdan]

as time goes on the 5800s will get weaker looking with the lackluster dx11 perf
but currently i dont think the 460s are really scaring them, otherwise we would have seen a real price drop (talking 10% at once, not 2% per month for what, 1 month so far, lol, 2010 sucked for gpus)

[safan80]

as time goes on the 5800s will get weaker looking with the lackluster dx11 perf
but currently i dont think the 460s are really scaring them, otherwise we would have seen a real price drop (talking 10% at once, not 2% per month for what, 1 month so far, lol, 2010 sucked for gpus)

Yeah I'm dreaming that the new single gpu ATI card or new Nvidia card that could actually beat the 5970 at 2560x1600.

[Etihtsarom]

Am I mistaken or are we really gonna care about more power when DX11 games are still just starting to show up AND the current gen is actually plenty for what's out there? What's the popular standard out there? 1920x1080, not 2560?

[Mech0z]

If this is true then we (I) should be able to use 2 x dvi + 1 x hdmi no?

[LordEC911]

For those of you that missed it...

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.

Sounds spot on to me other than the fact that Barts should be 256bit since Turks is the 128bit part.

The above was provided by neliz.

[Xoulz]

This.


Its been stated since day one Southern Islands was a stop gap between Evergreen and Northern Islands. It's been said that it would take the core from evergreen and match it with the uncore of northern islands. So based on that alone, I think most of the performance improvements will be in other areas other than just raw power. Logically it makes sense.

Maybe this analogy - kinda like the same CPU core getting a better chipset with more features later on its life.

correct.

not only that, but AMD is not aiming at ultra performance (80~120fps).

I suspect these new (half-breed?) cards, might not be any faster than their brethren, but their impedance, or minimum frames per second is constantly higher. Trading off pure performance for intelligent/efficient performance.
(?)

[Calmatory]

So according to that, only HD5470-HD5670 replacements in Q4/10?

[madcho]

So according to that, only HD5470-HD5670 replacements in Q4/10?

This roadmap is sh*t from a guy that set his name bottom right.

The true can't be near this. Why ? Just because Dirk Meyer, the CEO said the full new line up will be delivered to market before the end of the year.

So it's very bad and old speculation.

Current best is the ATI forum.de that offer the best idea on what's going next. The roadmap from Honma is just a joke

[Sn0wm@n]

correct.

not only that, but AMD is not aiming at ultra performance (80~120fps).

I suspect these new (half-breed?) cards, might not be any faster than their brethren, but their impedance, or minimum frames per second is constantly higher. Trading off pure performance for intelligent/efficient performance.
(?)

LOL fanboy comment ...


seriously these threads make me lol

[spursindonesia]

For those of you that missed it...



Sounds spot on to me other than the fact that Barts should be 256bit since Turks is the 128bit part.

The above was provided by neliz.

Yeah the buswidth and the ROP seems quite puzzling to me up to this moment. I think if ATi push for 384 bit for Cayman & 256 bit for Bart, they won't be pad limited, but the die size might grow excessively since 384 bit means 48 ROP & 256 bit 32 ROP consecutively. But with the slowing of GDDR speed increasing pace & the added die size from core improvement using the same process node, it's quite reasonable for them using the larger buswidth & the associated added ROP, if the chip performance justifies that. OTOH, that would mean the end of sweetspot strategy, or not ? aarrggh, i should stop speculating by now, a layman shouldn't think too much out of his league.

[xBanzai89]

Yeah the buswidth and the ROP seems quite puzzling to me up to this moment. I think if ATi push for 384 bit for Cayman & 256 bit for Bart, they won't be pad limited, but the die size might grow excessively since 384 bit means 48 ROP & 256 bit 32 ROP consecutively. But with the slowing of GDDR speed increasing pace & the added die size from core improvement using the same process node, it's quite reasonable for them using the larger buswidth & the associated added ROP, if the chip performance justifies that. OTOH, that would mean the end of sweetspot strategy, or not ? aarrggh, i should stop speculating by now, a layman shouldn't think too much out of his league.

I assume the "sweet spot strategy" may only apply to a new process. If you want to increase performance on the same process they will have to go a bit bigger. If they want a noticeable effect. I am sure you will see that sweet spot next for 28nm.

[jam2k]

lol fanboy comment ...


Seriously these threads make me lol

+500

[saaya]

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)

very good post

[saaya]

Yeah the buswidth and the ROP seems quite puzzling to me up to this moment. I think if ATi push for 384 bit for Cayman & 256 bit for Bart, they won't be pad limited, but the die size might grow excessively since 384 bit means 48 ROP & 256 bit 32 ROP consecutively. But with the slowing of GDDR speed increasing pace & the added die size from core improvement using the same process node, it's quite reasonable for them using the larger buswidth & the associated added ROP, if the chip performance justifies that. OTOH, that would mean the end of sweetspot strategy, or not ? aarrggh, i should stop speculating by now, a layman shouldn't think too much out of his league.

how could it be 384bit if its pin to pin compatible with the 5000 series? maybe not pin to pin but i thought pcbs could be reused, cutting down ttm big time and reducing costs?

[spursindonesia]

how could it be 384bit if its pin to pin compatible with the 5000 series? maybe not pin to pin but i thought pcbs could be reused, cutting down ttm big time and reducing costs?

Perhaps it's not Cayman which is pin compatible with Cypress, but Bart instead. A 256 bit 32 ROP 1280 SP (320*4) Cayman might end up around 270-290 mm^2, with perhaps ~HD 5850-5870 performance, it should be a pretty potent competitor for GF 104 derivatives chip in US$ 180-220 price segment.

I see Bart as a HD 4770 vs HD 4850, but instead using new process node to gain added efficiency (perf/watt or die area), it uses new mArch development such as rumored shader/SIMD rearrangement of the current ATi's GPU mArch. Pretty similar to nVidia's approach in GF 104 development that succesfully beats its bigger brother in perf/watt & die area while still using the same proces node.

[saaya]

makes sense, and ati had more time to tweak their design than nvidia...
and they have more experience with tweaking 40nm and getting the most out of every transistor than nvidia as well it seems...

but looking at the specs...
1280sps will be enough to reach 5850s 1440sp performance?
that would be a performance boost of around 12%?
thats not too much... you could get that with clockspeeds alone
765->850mhz...
so then whats the point of rolling out new silicon?
cause its a 10% smaller chip?
idk... this sounds weird... that wouldnt really make sense...

if those die sizes are correct, then performance HAS to be higher... otherwise it wouldnt make sense to roll out new silicon...
it has to be at least as fast as a 5870... at least...

think about it, a 512sp 480 is how much faster than a 5870?
bart should have roughly the same perf boost as cayman, so it wouldnt be going from 5830 to 5850, thats a rather small boost... it should go from 5830 to 5870+

[tbone8ty]

also consider the lengths of ATI cards...i have a feeling we will be seeing short cards like the 460's

[Olivon]

Perhaps it's not Cayman which is pin compatible with Cypress, but Bart instead. A 256 bit 32 ROP 1280 SP (320*4) Cayman might end up around 270-290 mm^2, with perhaps ~HD 5850-5870 performance, it should be a pretty potent competitor for GF 104 derivatives chip in US$ 180-220 price segment.

I see Bart as a HD 4770 vs HD 4850, but instead using new process node to gain added efficiency (perf/watt or die area), it uses new mArch development such as rumored shader/SIMD rearrangement of the current ATi's GPU mArch. Pretty similar to nVidia's approach in GF 104 development that succesfully beats its bigger brother in perf/watt & die area while still using the same proces node.

Yeah, it sounds plausible spursindonesia

[madcho]

Bart will use 5870 PCB and is pin compatible if you read corectly ATI forums.de

[LordEC911]

but looking at the specs...
1280sps will be enough to reach 5850s 1440sp performance?
that would be a performance boost of around 12%?
thats not too much... you could get that with clockspeeds alone
765->850mhz...
so then whats the point of rolling out new silicon?
cause its a 10% smaller chip?
idk... this sounds weird... that wouldnt really make sense...

From what it sounds like and if the changes are true, a 1280SP Barts sound be performing around a 5850-5870. A 1280SP 4d design should perform very similar to a 1600SP 5d design, since they supposedly aren't often utilizing the 5th SP in gaming situations and that they are touting "efficiency" as their main goal.

Neliz was saying something quite awhile ago about the 5850 replacement performing around a 5870. Cayman will be taking on the 512SP GF100 with the cutdown/salvage part being slightly ahead of the current 480SP GF100. Think about it, you need around a 20-25% performance increase over 5870 to match/beat GTX480 in most situations. Cayman needs a 10-20% increase over that to match/beat a 512SP GF100, if it stays around 700-750mhz.

So all that math above comes out to a fully specced Cayman needing to be roughly 35-40% faster to fall within those guidelines, which is what I was originally speculating awhile ago.

On a side note, the info from the roadmap is a good 4 years old but the codenames are correct. Also, the dates are there for mass availability, do not base launch/release dates off the roadmap. Most early roadmaps like these usually give a quarter or so for leeway.

[spursindonesia]

makes sense, and ati had more time to tweak their design than nvidia...
and they have more experience with tweaking 40nm and getting the most out of every transistor than nvidia as well it seems...

but looking at the specs...
1280sps will be enough to reach 5850s 1440sp performance?
that would be a performance boost of around 12%?
thats not too much... you could get that with clockspeeds alone
765->850mhz...
so then whats the point of rolling out new silicon?
cause its a 10% smaller chip?
idk... this sounds weird... that wouldnt really make sense...

if those die sizes are correct, then performance HAS to be higher... otherwise it wouldnt make sense to roll out new silicon...
it has to be at least as fast as a 5870... at least...

think about it, a 512sp 480 is how much faster than a 5870?
bart should have roughly the same perf boost as cayman, so it wouldnt be going from 5830 to 5850, thats a rather small boost... it should go from 5830 to 5870+

I dunno Saaya, i think every mm^2 counts especially if the chip is gonna be used in sub-200 US$ segment, and considering supply problem that has been plaguing ATi for since the launch of Evergreen family, this will help alleviating that problem somehow.

Regarding performance, i think ~HD 5850 can be counted as being conservative, with 320*4 config, if this rearrangement is successfull, we can expect ~HD 5870 like performance in many occasions, and good performance is not only determined by max & avg FPS, i expect some decent improvements in min FPS & a competitive tesselation performance in these upcoming chips from ATi.

Well, i guess these speculations are rather leaning on an optimist side, but considering what ATi has achieved since HD 2900 debacle, the optimism is quite warranted. In this scenario, nVidia's only competitive in the mainstream segment (GF 104 is pretty good chip, especially if the full spec one can be produced in good quantity), but would have much less profit & incentive to play the price war game in maintaining market share.

[mindfury]

http://forum.beyond3d.com/showpost.p...postcount=1318

According to the latest Catalyst 10.8 Antilles is Cayman based.

By the way, the driver reveals also the relations between marketing and engineering Codenames and all of the new "Southern Islands" GPUs have the prefix "NI" to it, e.g. "NI Cayman", "NI Barts", or "NI Caicos". So it appears SI is really NI in 40nm.
And some additional thing is that the mobile versions will have Pro, XT, and LP versions, whereas one has XT, Pro and sometimes LE versions for the desktop. And also the mobile chips sometimes appear as Gemini (X2?) variants.

223,CAYMAN GL XT (6700),NI CAYMAN
224,CAYMAN GL XT (6701),NI CAYMAN
225,CAYMAN GL XT (6702),NI CAYMAN
226,CAYMAN GL XT (6703),NI CAYMAN
227,CAYMAN GL PRO (6704),NI CAYMAN
228,CAYMAN GL PRO (6705),NI CAYMAN
229,CAYMAN GL (6706),NI CAYMAN
230,CAYMAN GL LE (6707),NI CAYMAN
231,CAYMAN GL (6708),NI CAYMAN
232,CAYMAN GL (6709),NI CAYMAN
233,CAYMAN XT (6718),NI CAYMAN
234,CAYMAN PRO (6719),NI CAYMAN
235,ANTILLES PRO (671C),NI CAYMAN
236,ANTILLES XT (671D),NI CAYMAN
237,BLACKCOMB XT/PRO (6720),NI BLACKCOMB
238,BLACKCOMB LP (6721),NI BLACKCOMB
239,BLACKCOMB XT/PRO Gemini (6724),NI BLACKCOMB
240,BLACKCOMB LP Gemini (6725),NI BLACKCOMB
241,BARTS GL XT (6728),NI BARTS
242,BARTS GL PRO (6729),NI BARTS
243,BARTS XT (6738),NI BARTS
244,BARTS PRO (6739),NI BARTS
245,WHISTLER XT (6740),NI WHISTLER
246,WHISTLER PRO/LP (6741),NI WHISTLER
247,WHISTLER XT/PRO Gemini (6744),NI WHISTLER
248,WHISTLER LP Gemini (6745),NI WHISTLER
249,ONEGA (6750),NI TURKS
250,TURKS XT (6758),NI TURKS
251,TURKS PRO (6759),NI TURKS
252,SEYMOUR XT/PRO (6760),NI SEYMOUR
253,SEYMOUR LP (6761),NI SEYMOUR
254,SEYMOUR XT/PRO Gemini (6764),NI SEYMOUR
255,SEYMOUR LP Gemini (6765),NI SEYMOUR
256,CAICOS GL PRO (6768),NI CAICOS
257,CASPIAN PRO (6770),NI CAICOS
258,CAICOS PRO (6779),NI CAICOS

[Mungri]

also consider the lengths of ATI cards...i have a feeling we will be seeing short cards like the 460's

Id really like that. GTX 460 / HD 5770 are about as large as I want my cards to be.

I dont know if its just me, but when it comes to graphics cards, I am not a size queen.

[Solus Corvus]

http://forum.beyond3d.com/showpost.p...postcount=1318

According to the latest Catalyst 10.8 Antilles is Cayman based.

By the way, the driver reveals also the relations between marketing and engineering Codenames and all of the new "Southern Islands" GPUs have the prefix "NI" to it, e.g. "NI Cayman", "NI Barts", or "NI Caicos". So it appears SI is really NI in 40nm.
And some additional thing is that the mobile versions will have Pro, XT, and LP versions, whereas one has XT, Pro and sometimes LE versions for the desktop. And also the mobile chips sometimes appear as Gemini (X2?) variants.

223,CAYMAN GL XT (6700),NI CAYMAN
224,CAYMAN GL XT (6701),NI CAYMAN
225,CAYMAN GL XT (6702),NI CAYMAN
226,CAYMAN GL XT (6703),NI CAYMAN
227,CAYMAN GL PRO (6704),NI CAYMAN
228,CAYMAN GL PRO (6705),NI CAYMAN
229,CAYMAN GL (6706),NI CAYMAN
230,CAYMAN GL LE (6707),NI CAYMAN
231,CAYMAN GL (6708),NI CAYMAN
232,CAYMAN GL (6709),NI CAYMAN
233,CAYMAN XT (6718),NI CAYMAN
234,CAYMAN PRO (6719),NI CAYMAN
235,ANTILLES PRO (671C),NI CAYMAN
236,ANTILLES XT (671D),NI CAYMAN
237,BLACKCOMB XT/PRO (6720),NI BLACKCOMB
238,BLACKCOMB LP (6721),NI BLACKCOMB
239,BLACKCOMB XT/PRO Gemini (6724),NI BLACKCOMB
240,BLACKCOMB LP Gemini (6725),NI BLACKCOMB
241,BARTS GL XT (6728),NI BARTS
242,BARTS GL PRO (6729),NI BARTS
243,BARTS XT (6738),NI BARTS
244,BARTS PRO (6739),NI BARTS
245,WHISTLER XT (6740),NI WHISTLER
246,WHISTLER PRO/LP (6741),NI WHISTLER
247,WHISTLER XT/PRO Gemini (6744),NI WHISTLER
248,WHISTLER LP Gemini (6745),NI WHISTLER
249,ONEGA (6750),NI TURKS
250,TURKS XT (6758),NI TURKS
251,TURKS PRO (6759),NI TURKS
252,SEYMOUR XT/PRO (6760),NI SEYMOUR
253,SEYMOUR LP (6761),NI SEYMOUR
254,SEYMOUR XT/PRO Gemini (6764),NI SEYMOUR
255,SEYMOUR LP Gemini (6765),NI SEYMOUR
256,CAICOS GL PRO (6768),NI CAICOS
257,CASPIAN PRO (6770),NI CAICOS
258,CAICOS PRO (6779),NI CAICOS

Well, those are all islands in the northern hemisphere...

[blindbox]

Which means that it's NI that's coming first, not SI. We were never sure which one is going to come first.