AMD: Sea Islands R1100 (8*** series) Speculation/ Rumour Thread

[CarstenS]

Fermi was not such a blast because of 40nm node as it was for being new design from the scrap after NV40-G71-G80-GT200 evolutionary steps.

I am still inclined to see a more than evolutionary step between G7x and G8x, more so than between GT200 and Fermi actually - ok, maybe not when talking on the basis of some "G7x with bolted on geo-shaders" or "fermi does tesselation in software" line of arguments. But no one is doing that, won't they?

 

[keritto]

Fermi was not such a blast because of 40nm node as it was for being new design from the scrap after NV40-G71-G80-GT200 evolutionary steps.

I am still inclined to see a more than evolutionary step between G7x and G8x, more so than between GT200 and Fermi actually - ok, maybe not when talking on the basis of some "G7x with bolted on geo-shaders" or "fermi does tesselation in software" line of arguments. But no one is doing that, won't they?

I was clear enough with thought that NV40 thru GT200 was evolutionary when it comes to front-end evolution and unified shaders were redesigned to fit that front end, as it was necessary for dx10 support. While Fermi was build from scrap with both front end being totally redesigned no more HS/DS and Tess engine could be "bolted", as you said, to aging 6years old design-. Along with that they fully redesigned their SMX/Core concept.

If i'm so wrong why they did need so huuge amount of time to offer dx10.1 hardware based on GT200 but still redesigned from scrap when it comes to Cores. They obviousl saw that they cannot just use old approach and attach many new features into front end without loss of too much performance.and we all saw how cumbersome this rush to Fermi design reflected to its power envelope.

But it's easier to attack with some pre-Fermi launch arguments like "fermi does tesselation in software". Which might be even true but for "original Fermi design" that was never produced for market, so that their front end was still so much screwed by "bolting" HS/DS to their "evolutionary dx9.0c pipeline". But they realized soon enough that concept would made Fermi, along with being power hog, but also falling huge margin behind competitions performance so envydia rather decided to delay it until they fix that problem.
Released Fermi architecture evidently didnt have no bolts with their predecessor GT200 and that was what i stated in larger than a line that you quoted.

 

[CarstenS]

If you mean this sentence, then yes, I did not read it as "front-end evolution"-only for NV4x to GT21x. I am sorry.
"All it had common with previous dx10 chips were unified shaders but these need to support new CUDA implementation."

 

[Love_In_Rio]

I am still inclined to see a more than evolutionary step between G7x and G8x, more so than between GT200 and Fermi actually - ok, maybe not when talking on the basis of some "G7x with bolted on geo-shaders" or "fermi does tesselation in software" line of arguments. But no one is doing that, won't they?

IMHO G80 was the more revolutionary step in NVIDIA´s design history, at least after TNT and Geforce 1.

 

[MDolenc]

Released Fermi architecture evidently didnt have no bolts with their predecessor GT200 and that was what i stated in larger than a line that you quoted.

I'll hate myself for this a bit but here it goes anyway...
What bolts did then the G80 have with G7x that you so eloquently proclaimed as evolutionary step from G71?

 

[fellix]

G80 was the first implementation based on the CUDA framework, that started as a parallel compute project way back in 2002. It is a completely separate architecture and the only remote common link with the older non-CUDA GPUs is the support for separate clock domains (NV40 had it too, but was never actually used).

 

[CarstenS]

FWIW, G70 and G71 had a slightly higher higher clocking vertex shader circuitry. Or so common monitoring programs would lead you to believe (need to be careful with Q&As wrt Nvidia as of late...).

 

[caveman-jim]

Im beginning to doubt there will be another set of cards launched.

 

[Love_In_Rio]

Im beginning to doubt there will be another set of cards launched.

mmm...we are approaching the point in which an architecture will last for many many years...Good that both Nviidia and Ati has reached it with good and efficient architectures...or it would have happened like in the CPU front in which Intel will make at its will...( cpus solded to the motherboard anyone? ).
Sea islands will be out because its a refinement of Southern islands in the same process but after that: 22nm and 14nm... I don´t see it in at least three years and i doubt we see a tock here. Consoles if powerful enough will restrain PC market again.

 

[Alexko]

Im beginning to doubt there will be another set of cards launched.

Why?

 

[swaaye]

There is already a rebrand doing duty as a Radeon 8550M. I think it's a safe bet that there will be more to come.
http://www.techpowerup.com/175785/A...ok-Pops-Up-with-Radeon-HD-8550M-Graphics.html

 

[Dave Baumann]

What makes you think its a rebrand?

 

[boxleitnerb]

It is more likely than not being one. Both AMD and Nvidia love to rebrand especially in the mobile segment.

 

[Alexko]

What makes you think its a rebrand?

Are you saying it isn't?

(PS: new BIOSes don't count.)

 

[turtle]

What makes you think its a rebrand?

Is it from Mars or something?

 

[swaaye]

Help me deciper this post

http://www.techpowerup.com/175785/A...ok-Pops-Up-with-Radeon-HD-8550M-Graphics.html

The 8550M uses Graphic Core Next but it is a new die. Radeon HD 7850M / 2 = Radeon HD 8550M 500 -> 900 series are all Graphic Core Next. 500 -> 5 CUs -> 320 ALUs (A6 Kaveri, 8560G) 600 -> 6 CUs -> 384 ALUs (A8 Kaveri, 8660G) 700 -> 6 CUs -> 384 ALUs (A10? Kaveri, 8760G) 800 -> 12 CUs, 16 CUs -> 768 ALUs, 1024 ALUs 900 -> 20 CUs, 24 CUs -> 1,280 ALUs, 1,536 ALUs All of the new GPUs are based on 28-nm HP with straining where the previous generation was 28-nm HPL. There is no 8400/8300/8200/8100 as those are taken by Kaveri A4/Kaveri E2/Jaguar E2/Jaguar E/Jaguar C/Jaguar Z. Graphic Core Next 2.0 is not Enhanced Graphic Core Next. --- AMD1305.1 = "KAVERI1 DESKTOP (1305)" AMD1304.1 = "KAVERI1 MOBILE (1304)" AMD1307.1 = "KAVERI2 DESKTOP (1307)" AMD1306.1 = "KAVERI2 MOBILE (1306)" AMD9833.1 = "KB 12W 2C (9833)" AMD9834.1 = "2C 5W KB (9834)" AMD9831.1 = "KB 4C 17W (9831)" AMD9832.1 = "KB 17W 4C (N-1) (9832)" AMD9830.1 = "KB 4C 25W (9830)" AMD6600.1 = "MARS (6600)" AMD6601.1 = "MARS (6601)" AMD6602.1 = "MARS (6602)" AMD6602.1 = "MARS (6603)" AMD6602.1 = "MARS (6606)" AMD6602.1 = "MARS (6607)" AMD6600.1 = "MARS (6620)" AMD6601.1 = "MARS (6621)" AMD6602.1 = "MARS (6623)" AMD6610.1 = "OLAND (6610)" AMD6611.1 = "OLAND (6611)" AMD6631.1 = "OLAND (6631)" AMD682B.1 = "VENUS LE" AMD6823.4 = "VENUS PRO" AMD6821.1 = "VENUS XT" AMD6820.2 = "VENUS XTX"

 

[iMacmatician]

Help me deciper this post

http://www.techpowerup.com/175785/A...ok-Pops-Up-with-Radeon-HD-8550M-Graphics.html

Here's the first half (I'm not sure about the second).

For mobile:

"All of the new" [CI?] GPUs will use 28 nm HP with straining instead of the 28 nm HPL in Southern Islands.

GCN 2.0 ≠ Enhanced GCN.

The 89x0M - 85x0M series are all GCN. The chip used by the 8550M is also a new chip.

89x0M: 1280 SPs, 1536 SPs.
88x0M: 768 SPs, 1024 SPs.
87x0M: 384 SPs, also possibly A10 Kaveri, 8760G.
86x0M: 384 SPs, also A8 Kaveri, 8660G.
85x0M: 320 SPs, also A6 Kaveri, 8560G. In particular, the 8550M is half of a 7850M.
84x0M - 81x0M: Absent, their positions filled with Kaveri A4/E2 and Jaguar E2/E/C/Z.

 

[RedVi]

No need, i'd bet on it all being made up. It doesn't make any sense to release a GCN low end GPU when the APU's they have to crossfire with won't be GCN until late next year at the earliest.

I guess I could be and would like to be wrong there, but additionally much of the other 'info' he spews doesn't make much sense. 8900 series has 20-24CU's? 8700 series has 6CU's? Clearly he's trying to talk about an APU with that last one, but his rule doesn't make any sense.

EDIT: Oh, well if he's talking about mobile (which makes sense...) then I guess it could line up. :\

 

[AlexV]

And once again AMD is truly on top of the game when it comes to getting its message out! I'll go cut up some chikkinz and inspect their entrails, perhaps the answer to the life and deeds of the 8550M is hidden right there! Is it a tired underpaid ASUS employee mistyping? Is it the seeekrit project that will take the world by storm? Tune in for the next exciting episode of "BeyondEntrails"!

 

[I.S.T.]

People here have really gotten angry with AMD as of late... Even as someone who mostly prefers Nvidia it's beginning to bug me.