backtop

Print E-mail del.icio.us 62 comment(s) - last by Justin Case.. on Aug 3 at 2:33 AM
AMD puts second-generation "Stars" plans on its roadmap, now with more celestial names

AMD’s latest roadmap outlines plans for second-generation Stars quad, dual and single-core processors. Five second-generation Stars processors are in the pipeline for 2H’2008. The five variants include quad-core Deneb FX and Deneb, dual-core Propus and Regor, and single-core Sargas. The new processors drop into new Socket AM3 and AM2+ motherboards for backwards compatibility.

All second-generation Stars processors have new memory controllers with DDR3 and DDR2 memory compatibility. Taking advantage of DDR3 memory requires the use of a Socket AM3 motherboard. Socket AM2+ users can upgrade to second-generation Stars processors with a memory bandwidth penalty.

Second-generation Stars processors introduce AMD’s 45nm fabrication process. AMD plans to make a quick move to 45nm. The entire product lineup will move to 45nm within 2H’2008, unlike the move to 65nm, which began Q4’2006 and is still in process.

Succeeding Agena FX and Agena processors are the Deneb FX and Deneb cores. AMD has not set any specific details such as cache configuration or HyperTransport speeds yet. However, Deneb FX and Deneb processors will carry on with shared L3 caches and HyperTransport 3.0. AMD will offer Deneb FX in Socket AM2+/AM3 and land-grid array 1207+ socket configurations. Expect Deneb FX and Deneb cores to carry the Phenom FX and Phenom X4 names.

AMD’s Propus replaces Kuma in 2H’2008. Propus shares the same feature set as Kuma such as HyperTransport 3.0 and shared L3, however, specific cache sizes are unset. Expect Propus to continue with the Phenom X2 brand.

The AMD Athlon X2 Rana processor will have a short life span. AMD expects to debut the 65nm entry-level dual-core processor in 1H’2008 and replace it in 2H’2008 with Regor. Spica will face the same fate as Rana.

The single-core Spica has a half-year life cycle before Sargas in 2H’2008 replaces it. Expect Regor to carry the Athlon X2 name while Sargas carries the Sempron name. Features of the value AMD processors should be no different from its predecessors, with the exception of DDR3 memory support.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
Show something
By Regs on 7/2/2007 4:51:00 PM , Rating: 3
*picks up phone and calls mom who spent thousands of dollars over a long time on raising me to the guy I am today*

Hi Mom! I have a baby now. I named my son Star and no you cannot see him.




RE: Show something
By TSS on 7/2/2007 5:03:18 PM , Rating: 2
lets see barcelona before we see deneb.

well, guess we know now what AMD wants to pit against nehalem. lets hope its enough because if the performance lacks too much, in combination with the move to 45nm which should be costly.... AMD might not have hit their roughest weather yet...

on the other hand, it might be intel that screws up this time, with a new architecture instead of a refresh. but i doubt it.

either way, lets hope its less of a buzz kill then barcelona has been so far.


RE: Show something
By Regs on 7/2/2007 5:07:09 PM , Rating: 2
This is actually just for penyrn. Nethlem they hope to have shanghi ironed out for it.


RE: Show something
By Regs on 7/2/2007 5:08:19 PM , Rating: 2
RE: Show something
By omnicronx on 7/2/2007 5:09:25 PM , Rating: 3
This only leaves me to believe that AMD knows Intel fucked up with with not jumping on the SOI bandwagon, and they want to get to sub 45nm levels as fast as they can. It gives me a glimmer of hope.. i just hope Barcelona is not a big flop


RE: Show something
By Treckin on 7/2/2007 5:39:47 PM , Rating: 1
If he got rated down because he said fuck, I hope DT takes away the stupid rating feature....

That was a very astute comment. IN fact however, I think this is intended to fall somewhere between penryn and nehalem. AMD was previously down about an entire architecture iteration, so they are hoping to shrink the lead to only 1.5 here. It would also explain the shitty performance of Barcelona... Unless by some second coming of the messiah Barcelona is a sleeping mamba (*preys*), we wont see any competition until this.
AMD tried both a die shrink and a arch. switch with Barcelona, as well as breaking into uncharted territory with the native quad core dealeo.
Some people I think are confused as to what penyrn is.....
Its just a C2D at 45nm.... should draw a bit less power, maybe a few percentage points higher due to less heat---> higher clocks, but IMO Barcelona will at least make it a fairly even game.
They are focusing on winning the next battle with the Stars line, having decisively lost this round of core design.


RE: Show something
By James Holden on 7/2/2007 5:55:17 PM , Rating: 1
He didn't get rated down cause he cussed. He got rated down cause his post didn't make any sense.

SOI was to help clock speeds ramp in the single core days. It worked for IBM, 4.7 ghz on PPC. Why didn't it work for AMD? Cause the architecture won't get above 2.0ghz right now, let alone 3. Intel doesn't need SOI, they're not over 3.0ghz either.


RE: Show something
By plewis00 on 7/2/2007 8:04:28 PM , Rating: 2
Penryn Core 2s will also carry 6Mb cache on the Duos (Wolfdale) and 12Mb cache on the Quads (Yorkfield) with a 1333FSB. I'm pretty sure they also have SSE4 support and are approximately 5-10% faster clock-for-clock than the current generation. It's not just a die-shrink, there are some minor architectural differences but this is all pretty normal if you look back over previous chips, a new architecture, then a die-shrink after approx. year with some minor changes then another year with the same die-size but new architecture (I think it's because you wouldn't risk a whole new architecture straight out with an immature process) and so-on (this is all vaguely speaking).


RE: Show something
By smilingcrow on 7/2/07, Rating: 0
RE: Show something
By kamel5547 on 7/2/2007 6:26:50 PM , Rating: 3
ANd here I though AMD was simply trying to get onto the same manufacturing process as Intel rather than constantly being a generation behind.

I don't think the rush to 45nm this means anything besides the fact AMD would like to level the playing field as much as possible.


RE: Show something
By vignyan on 7/3/2007 7:18:40 AM , Rating: 4
Dude... you really think Intel does not know about SOI when you do? there are PhD's working in their research labs not people like you and me.

The fact is that SOI yeilds are better but at the same time, it has multiple limitations. SOI is a very costly process. The retooling of a fab itself takes around 2-3bn USDs.. That pushes your margins down. There is a minor difference in the yields and reliability. Still these can be covered in the design.

Apart from being costly, as some one commented in this thread, it does not scale well. Thats why intel reached 3GHz in 2003-2004 ish and AMD reached it in 2006 ish (core clock not the 3000+ thing)...

Also at the lower nodes, (65nm, 45nm, 32nm...) there is an effect called as Antenna effect which affects the SOI process based products to a large extent due to the electron entrapment during the SOI process.

So it all depends on what you call superior.... the finances and the future options also contribute to that "superior technology" brand! :)


RE: Show something
By Justin Case on 7/3/2007 3:40:49 PM , Rating: 2
Barcelona is having problems scaling past 2 GHz or so in volume. But some 2.3 GHz units have been floating around for a while, and they're quite impressive (we're talking 50% faster than a 2.66 GHz Clovertown, in FP, and around 15-20% faster in integer).

If AMD can fix the speed scaling issues quickly, Intel is in for some serious competition. That is a big "if", however. With the current yields on parts above 2 GHz, they're just not economically viable.


RE: Show something
By TomZ on 7/3/2007 5:15:08 PM , Rating: 1
50%?!? I'll believe that when I see public benchmarks on AnandTech, et al.


RE: Show something
By excrucio on 7/3/2007 5:29:00 PM , Rating: 2
where da hell u been there was a 3rd party benhcmark not too long ago in this site


RE: Show something
RE: Show something
By AvidDailyTechie on 7/5/2007 9:20:25 PM , Rating: 2
Did I miss something? How does that link support Justin Case's comment? I think it's the opposite: "The Kentsfield Xeon was 58% faster with a 50% higher clock frequency for Cinebench"

Even accounting for clock differences, wouldn't it still actually be slower... clock for clock???


RE: Show something
By Justin Case on 7/7/2007 2:48:10 PM , Rating: 2
Cinebench is a worst-case scenario for AMD (both with Barcelona and older models); it makes heavy use of SSE3 and very little x87 FP (unlike more advanced software, such as Maya, 3dsmax, etc.). All but the latest revision of Barcelona has issues with SSE3. The latest revision fixes those (about 10-15% faster than the previous ones), but they still can't get enough chips to run at high enough clock speeds to make models above 2 GHz commercially viable.

Until that happens, Barcelona will only beat Intel in 4-socket configurations and above (and it more or less matches them on 2S configurations); it clearly loses in 1S, which is still where most of the sales are.


RE: Show something
By KristopherKubicki (blog) on 7/3/2007 5:26:54 PM , Rating: 2
I've not read or heard anything that would indicate anyone has a chip at 2.3 GHz.


RE: Show something
By Justin Case on 7/26/2007 9:05:44 PM , Rating: 2
Well, I take this article...

http://www.dailytech.com/AMD+Demonstrates+30+GHz+Q...

...to mean that now you have read or heard something about chips running at 2.3 GHz (and higher)...? :)

Like I said, they're just not economically viable (yet). But some select partners have had access to them for a while.


RE: Show something