AMD: R9xx Speculation

[Lightman]

I have a positive feeling that the changes from HD5k to HD6k are more fundamental than the ones going from RV770 to Evergreen. Dunno why, though.

Regarding that stupid table with horrible name mixing just ignore it!
It says in 3DCentre article that table was compiled from FORUM LEAKS posted at http://we.pcinlife.com/thread-1516392-1-1.html

It's not anything from AMD or even leaked press deck so just forget it and stop spreading FUD guys :smile:

 

[Forrest]

I have a positive feeling that the changes from HD5k to HD6k are more fundamental than the ones going from RV770 to Evergreen. Dunno why, though.

You're both right and wrong!

 

[MarkoIt]

You're both right and wrong!

Ah, maybe i got it.
Bart goes for HD67x0
Cayman goes for HD68x0
Juniper (!) goes for HD66x0/HD65x0

In this way he is both right and wrong.

 

[neliz]

That's not fair Marko

 

[Kaotik]

Ah, maybe i got it.
Bart goes for HD67x0
Cayman goes for HD68x0
Juniper (!) goes for HD66x0/HD65x0

In this way he is both right and wrong.

AMD is as bad as nVidia with naming if Juniper ever gets a HD6k name

 

[caveman-jim]

You're both right and wrong!

What makes you say that?

 

[MarkoIt]

AMD is as bad as nVidia with naming if Juniper ever gets a HD6k name

Actually, since there is no new feature set introduced by N.I, the only thing that should concern us is the performance/$$. AMD might have decided that instead of replacing Redwood with an another 40nm design, they could have just used Juniper.
Right now a cut-down version of Juniper is already sold as 5670. I don't see why a possible HD6670 couldn't use a full Juniper.. priced as a x6xx series it would be a great deal!
Turks and Caicos may be 28nm parts that will replace Juniper later on, let's say in 6-8 month ?
It would fit the schedule for a pipe-cleaner @28nm for next generation.

 

[OgrEGT]

(...)
Either way, I have a hard time giving that chart any validity. Unless as no-X was suggesting they just got their wording wrong and they actually 90% MORE performance rather than 90% of the performance.

But then that still brings into question its validity as I have a hard time believing 6870 is going to be 190% of the speed of 5870.

Regards,
SB

If this refers back to the news on 3dcenter.org, they are more like talking about 90% of the 5870 performance. The site is in German, so am I

In the next sentence they are thinking about, why should AMD do such a marketing "harakiri" without any good case and getting the unavidable disastrous feedback from the press on top.

 

[gamervivek]

What makes you say that?

Mama always told me "Life is like a box of chocolates: You never know what you' re gonna get."

 

[LordEC911]

Turks and Caicos may be 28nm parts that will replace Juniper later on, let's say in 6-8 month ?
It would fit the schedule for a pipe-cleaner @28nm for next generation.

Juniper is ~180mm2 at 40nm, 40nm pipe cleaner RV740 was ~140mm2.
I think a 28nm Juniper is a little too small for a pipe cleaner.
IMO a 28nm Barts would be a better subject.

I also can't rule out 28nm Turks and Caicos 1H '11, though that does seem a bit too optimistic.

 

[Kaotik]

Actually, since there is no new feature set introduced by N.I, the only thing that should concern us is the performance/$$. AMD might have decided that instead of replacing Redwood with an another 40nm design, they could have just used Juniper.
Right now a cut-down version of Juniper is already sold as 5670. I don't see why a possible HD6670 couldn't use a full Juniper.. priced as a x6xx series it would be a great deal!
Turks and Caicos may be 28nm parts that will replace Juniper later on, let's say in 6-8 month ?
It would fit the schedule for a pipe-cleaner @28nm for next generation.

That's just bullcrap, nVidia got thrown to hell and dragged back just to get thrown again for their naming mess which was renaming chips to "new gen names", just like renaming Juniper to HD6-series part would be.

 

[no-X]

if Barts and Cayman features UVD 3.0 and lower parts only 2.0, it wouldn't be very nice...

 

[Kaotik]

if Barts and Cayman features UVD 3.0 and lower parts only 2.0, it wouldn't be very nice...

Indeed, that too. And the "Eyefinity+", assuming it adds functionality instead of just adding 1 more DisplayPort to the cards by default

 

[MarkoIt]

That's just bullcrap, nVidia got thrown to hell and dragged back just to get thrown again for their naming mess which was renaming chips to "new gen names", just like renaming Juniper to HD6-series part would be.

Things are bit different with G92. It was first used for 8800GT/8800GTS, than for the 9800GT/9800GTX, than for 9800GTX+, than for the GTS250.
That's crap, because they reused the same chip for a higher family of product (88xx GTS-> 98xx GTX)... and they actually used it again also for the mobile market.
But Juniper is actually 2 times Redwood. Reusing it for the 66x0 series it would be to deliver almost double the performance of the 56x0 for the same price. 5770 would go EoL.
The only drawback would be the lack of UVD3, which is not a small thing for the HTPC market.

 

[thatdude90210]

That's just bullcrap, nVidia got thrown to hell and dragged back just to get thrown again for their naming mess which was renaming chips to "new gen names"...

No they didn't, except maybe in a few messageboards. In the real world, they just made a lot of money.

 

[ShaidarHaran]

No they didn't, except maybe in a few messageboards. In the real world, they just made a lot of money.

Indeed. Enthusiast perception of the market does not often correspond with reality.

 

[trinibwoy]

No they didn't, except maybe in a few messageboards. In the real world, they just made a lot of money.

You're a wise man.

 

[CarstenS]

psolord: I think that means "performance over HD5870"... anyway, 90% over 5870 or 90% of 5870 - both are strange

It's "of", so 0.9 times and HD 5870 and so on.

Regarding that stupid table with horrible name mixing just ignore it!
It says in 3DCentre article that table was compiled from FORUM LEAKS posted at http://we.pcinlife.com/thread-1516392-1-1.html

Don't worry, I wasn't referring to 3DCenters' table - I am german myself and can read that page without a translator. No, there's something else, what made me say it. "Four" is a nice number in computing, isn't it? You can evenly split a 128 bit datapath between four unit, leaving all with the same bandwidth for example. Something that has been done with SSE2 for quite a while now, if I am not mistaken. And how long is it now since AMD took over Ati? And how long would an architectures design cycle roughly be?

 

[MarkoIt]

It's "of", so 0.9 times and HD 5870 and so on.



Don't worry, I wasn't referring to 3DCenters' table - I am german myself and can read that page without a translator. No, there's something else, what made me say it. "Four" is a nice number in computing, isn't it? You can evenly split a 128 bit datapath between four unit, leaving all with the same bandwidth for example. Something that has been done with SSE2 for quite a while now, if I am not mistaken. And how long is it now since AMD took over Ati? And how long would an architectures design cycle roughly be?

3 years. RV770 was started in 2005 and released in 2008. AMD bought Ati in 2006 so Norther Islands project was started after it.

 

[CarstenS]

See?