backtop


Print 52 comment(s) - last by Justin Case.. on Feb 9 at 3:55 PM


DailyTech's AMD Quad-Core Opteron Reference
Quad-core clock speeds and model names unveiled in AMD's latest Opteron road map

AMD’s latest Opteron product roadmap unveils details on the upcoming quad-core Opteron products. Upcoming quad-core AMD Opteron processors will be available in the Opteron 8000-series, 2000-series and 1000-series.

Quad-core AMD Opteron 8000-series and 2000-series will initially use the Barcelona core. AMD will ship quad-core Opteron processors in SE, vanilla and HE models.

All Opteron 8000 and 2000-series processors include SE, vanilla and HE models share similar attributes. Each processor core will have access to its own 512KB of L2 cache for a combined total of 2MB. L3 cache is new to the Barcelona-core Opteron models -- a shared pool of 2MB of L3 cache is available to all four cores.  

Barcelona-based Opteron 8000 and 2000-series Socket 1207 processors will have three HyperTransport links, though at launch the processor will only support HyperTransport 1.0 instead of the faster HyperTransport 3.0 bus. Other notable features of Barcelona quad-core processors include support for registered DDR2-667 and AMD-V technology. Quad-core Opteron 8000 and 2000-series processors will use the same socket F as its dual-core counterparts.

At the top of the quad-core Opteron chain are six Opteron SE processors – three Opteron 8000 and three Opteron 2000-series. The AMD Opteron 8272 SE and 2272 SE are clocked at 2.6 GHz while the middle-of-the-line Opteron 8270 SE and 2270 SE are clocked at 2.5 GHz. There is also a lower speed Opteron 8268 SE and 2268 SE clocked at 2.4 GHz.

The Opteron SE models will have the highest thermal data power ratings in AMD’s quad-core lineup. AMD currently rates the Opteron SE models with 120W TDPs. It is unknown when AMD will launch its Opteron SE processor lineup, however, production is expected to begin in Q3’2007 for the Opteron 8270 SE, 8268 SE, 2270 SE and 2268 SE. The Opteron 8272 SE and 2272 SE will enter production later in Q2’2008.

Moving into the middle of the AMD Opteron line-up are eight vanilla Opteron 8000 and 2000-series processors. AMD Opteron 8266 and 2266, 8264 and 2264, and 8262 and 2262 are clocked at 2.3 GHz, 2.2 GHz and 2.1 GHz respectively. These models will enter production in June 2007 with a mid-2007 launch date.

AMD expects to launch vanilla Opteron 8268 and 2268 models clocked at 2.4 GHz like to the Opteron 8268 SE and 2268 SE models too, though the vanilla Opteron 8268 and 2268 will enter production 3-quarters later than the SE models. Vanilla AMD Opteron models will have slightly lower TDP ratings at 95W instead of the 120W of Opteron SE models.

Four Opteron HE models will be available for the low power inclined. The Opteron 8260 HE, 2260 HE, 8258 HE and 2258 HE have 2.0 GHz and 1.9 GHz respective clocks. These low power Opteron HE models have 68W TDPs. Expect the Opteron HE models to enter production in Q3’07.

AMD’s Budapest core is behind the Opteron 1000-series processors. Budapest will drop into AMD’s single processor socket AM2 platform and be compatible with current HyperTransport 1.0 motherboards and upcoming HyperTransport 3.0 compatible motherboards (socket AM2+). HyperTransport 3.0 support aside, the Budapest-based Opteron 1000-series have 4x512KB of L2 cache, 2MB of L3 and support for AMD-V technology like the Opteron 8000 and 2000-series. Unlike the multi-processor capable offerings, Opteron 1000-series processors support unregistered DDR2-800 memory.

There will not be as many quad-core Opteron 1000-series models this time around, unlike the Opteron 8000 and 2000-series line-ups. Two SE and three vanilla Opteron 1000-series models complete AMD’s quad-core Opteron lineup. The Opteron 1270 SE and 1268 SE have 2.5 GHz and 2.4 GHz respective clocks. AMD does not have a 2.6 GHz Opteron 1000-series model planned now. Nevertheless, the Opteron 1000-series SE models have 120W TDP ratings.

The three vanilla Opteron 1000-series models include the Opteron 1266, 1264 and 1262 clocked at 2.3 GHz, 2.2 GHz and 2.1 GHz respectively. These models have a lower 95W TDP rating.

Expect AMD Opteron 1000-series models to enter production in Q4’07 with an undetermined launch date.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Smallest performance increase ever
By photoguy99 on 2/6/07, Rating: 0
RE: Smallest performance increase ever
By lplatypus on 2/6/2007 1:50:30 AM , Rating: 3
AMD has already done this recently with the 65nm Athlon 64 X2 4800+(2.5GHz) vs 5000+(2.6GHz).

The record is probably held by Intel though: remember the 3000MHz vs 3066MGz Pentium 4's?


RE: Smallest performance increase ever
By Calin on 2/6/2007 2:43:35 AM , Rating: 2
There were also the Pentium !!! 900 and 933 (on 100MHz and 133MHz bus) to be taken into account - which is less than 4%


By Hare on 2/6/2007 3:39:04 AM , Rating: 2
Back then the bus was the bottleneck. Getting 133mhz instead of 100mhz is a more than 4% performance increase depending on the usage of the processor.


By nah on 2/7/2007 6:49:23 AM , Rating: 2
;)


By aka1nas on 2/6/2007 1:59:04 AM , Rating: 2
My guess is that they need a lower model to justify charging more for the top model. That and they probably aren't willing to go above the TDP limits they are hitting already for a server proc and they can't currently ramp up clocks higher otherwise.



RE: Smallest performance increase ever
By coldpower27 on 2/6/2007 8:39:38 AM , Rating: 2
Intel has done the 2.5GHZ to 2.6GHZ 400FSB P4 to my knowledge.

But in terms of the same product line, this may be one of the smaller increments in a long time.

This allows increased product granularity so there are more stepping stones between SKU for pricing.

As well for a heavily aware SMP app, the difference between the 2.5GHZ - 2.6GHZ Dual Core could be as much as the difference between the 2.4GHZ and 2.6GHZ Quad Core.


RE: Smallest performance increase ever
By photoguy99 on 2/6/2007 9:25:59 AM , Rating: 2
>Intel has done the 2.5GHZ to 2.6GHZ 400FSB P4 to my knowledge.

No, never by Intel before and never by AMD before has there been two cpus where the only difference is a 4% clock speek bump.

There have of course been cases <4% when cache, FSB, or architecture has also changed, but never the case where spending more money bought you only 4% clock.

Also the Intel quad you note is 2.4->2.66 rather than 2.4->2.6. I guess the 0.06Ghz starts to matter when were talking only 0.1Ghz product changes now from AMD.

The funny thing to me is 4% is close to the statistical margin of error for some benchmarks. :)


RE: Smallest performance increase ever
By coldpower27 on 2/6/2007 10:33:11 AM , Rating: 2
Never really eh, considering the existence of the 2.5GHZ and 2.6GHZ 400FSB Pentium 4's I would have to disagree with you.

http://processorfinder.intel.com/details.aspx?sSpe...

http://processorfinder.intel.com/details.aspx?sSpe...

You have never actually done any research have you.

So like I said already done by Intel with the 2.5GHZ and 2.6GHZ FSB400 Pentium 4's.


By photoguy99 on 2/6/2007 2:43:13 PM , Rating: 2
Wow, I stand corrected for correcting you. You da man (or woman possibly).

At least that now that Intel has done it I can't be marked an AMD basher for saying this: It's a stupid idea.

For the consumer it offers no value to split hairs at this level. Sure I understand from a marketing perspective it makes sense, but that benefits AMD and Intel, not consumers.





By Nanobaud on 2/6/2007 1:26:01 PM , Rating: 2
Well, how about different SKUs for the same processor binned for different TDP?


By mino on 2/6/2007 2:47:12 PM , Rating: 2
Athlon XP 2200+ 1.8GHz
Athlon XP 2100+ 1.73GHz
= 3.85%

go figure


By leidegre on 2/7/2007 5:07:46 AM , Rating: 2
Well, it's a good hint avrage for most things, but seriously, there is more to it than that.

AthlonXP (1.8GHz) == Athlon64 (1.8Ghz)?

Hmm, I recall myself upgrading from the AthlonXP to the Athlon64 processor, and gained about double the previous performance.


amd rush
By DaRkFib3r on 2/5/07, Rating: 0
RE: amd rush
By Goty on 2/5/2007 10:53:32 PM , Rating: 2
How is this a rush? It seems like a reasonable product launch to me.


RE: amd rush
By DaRkFib3r on 2/5/2007 11:10:13 PM , Rating: 1
Its just that if they waited a couple more months intel would of gotten more of the server market with their quad core xeon processors.

AMD are pretty behind with releases, it may have to do with its manufacturing ability.


RE: amd rush
By Justin Case on 2/9/2007 3:37:14 PM , Rating: 2
Not really, they're pretty much sticking to their roadmap. Intel released a couple of products ahead of their planned dates (by 3 months, more or less), but with extremely low availability, so they're also pretty much sticking to their roadmap. In other words, there's absolutely nothing new here.

AMD is still gaining market share in the server segment (and will continue to do so for the forseeable future). Intel's quad-core releases are aimed mainly at workstations / "power user" segment (single socket, and a subset of the dual-socket market). To regain the 4S and 4C/2S markets they need CSI; their CPUs are bandwidth limited. The faster FSB helps, but isn't enough. But 1S is where most of the money's at, so that's where Intel needs to regain (or at least stop losing) share.

Since the K7 days AMD moves from one process to another gradually, so they never have big performance jumps.


RE: amd rush
By JackPack on 2/6/2007 3:59:30 AM , Rating: 2
Look at the "Production" and the "Lanuch" dates of the first parts.

That's one hell of a rush.

Basically, AMD marketing is trying to squeeze and claim a mid 2007 launch. You can probably guess whether we'll see a paper or silicon lanuch in mid '07.


RE: amd rush
By Viditor on 2/6/2007 9:49:43 AM , Rating: 3
quote:
Look at the "Production" and the "Lanuch" dates of the first parts


That's not a rush, it's impossible...
I would bet that DT has made an error.
The second week of June IS mid 07! Maybe they meant shipping rather than production (because I believe production has already started in Jan 07).

It takes 3 months from production start to deliver the first chip...


RE: amd rush
By coldpower27 on 2/6/2007 10:35:53 AM , Rating: 2
Well Mid 2007 would be from April 1st to August 31st 2007, SO if AMD can get shipping product by Late August they are still in the clear for Mid 2007.

If production is in Early June, they should be able to make a Late August Launch.

The thing this is different from the earlier projected timetable of Q2 2007 for the Server level Quad Cores.


RE: amd rush
By JackPack on 2/6/2007 4:41:41 PM , Rating: 2
Mid 2007 is June. Nobody uses FY on these roadmaps.


RE: amd rush
By Viditor on 2/7/2007 9:13:29 AM , Rating: 2
quote:
Mid 2007 would be from April 1st to August 31st 2007, SO if AMD can get shipping product by Late August they are still in the clear for Mid 2007


If they begin volume production on June 1, the first wafer out would be August 28...then there's the packaging and testing, at least another 2 weeks...and remember they would only have the first wafers (maybe a few thousand chips?).

The roadmap DT published absolutely MUST be wrong...there is no other explanation.


RE: amd rush
By JackPack on 2/7/07, Rating: -1
RE: amd rush
By Viditor on 2/7/2007 10:05:34 PM , Rating: 2
quote:
The other explanation is that it'll simply be a paper launch


It's even too far away for a paper launch...
Also, remember that they already have bin speeds. To do that, they must already have produced numerous test wafers to find what they will be. This indicates that production has already begun.

The most logical explanation is that DT messed up when translating from the Chinese...they read "production" when it should have been "shipping". A fairly easy thing to do when translating Chinese...


RE: amd rush
By Tom Tom on 2/7/2007 11:19:25 PM , Rating: 2
I agree. Whats really disturbing is DT is not attempting to clearify the situation, probably in the name of site hits.


RE: amd rush
By verndewd on 2/6/2007 3:08:59 PM , Rating: 1
It doesnt exactly look like great progress from the consumer standpoint.
I think its yes and no rushed,somethings were rushed some were not.I think quad is late.demoing it last year wasnt really a demo;I wonder if a nano wire would have fried if you started a program :) .

They took on way too much this past year,and the planning that could have kept them up to date was used to incorporate ATI into the next few generations releases.

The pay off is late and requires a few rushed things here and now.


RE: amd rush
By jpeyton on 2/6/2007 12:01:13 AM , Rating: 2
Rushing? They are 6+ months behind Intel; how is that rushing?

Are you high?


RE: amd rush
By verndewd on 2/6/2007 3:12:12 PM , Rating: 2
how do you arrive at 6 months?

go back and check the release dates of 65nm and the announced planning for it.6 months is FUD.32 bit core duo launched in dec 05,65nm chips were released months before that.google it.


something fishy is going on
By mino on 2/6/2007 10:32:38 AM , Rating: 2
IMO 2.3 is too low for a mature 65nm process. Unless..

Unless cities cores are REAL monsters (>10% faster than c2 cores).

This summer will be interesting to the least.

As for HT3 going first with desktop/single socket server. The simple reason is that 2S/4S solutions will not be properly evaluated by that time so HT3 was not an option.




RE: something fishy is going on
By coldpower27 on 2/6/2007 10:49:10 AM , Rating: 2
These are Quad Core CPU's. So they have higher TDP's and can't clock as high compared to Dual Core equivalents.

We will have to see won't we on how fast K10 vs Core Architecture is won't we..


RE: something fishy is going on
By mino on 2/6/2007 2:36:27 PM , Rating: 1
IMO the title of the article says Quad-Core suficiently for me to comprehend. :)

Problem is 2.3 Barcelona is REEALLY low. C2D is no Netburst...

2.3 (and nothing above until Q4) will be disaster for AMD. Period.
2.5 is a must in <4S landscape for them to be competitive.

I don't see a way for K8L core beeing more than 10-15% faster to Conroe in general.

I case 2.5 does not make it to the june launch in wolume at 89TDP, expect 2.5 or 2.6 with extreme prices and SE moniker to make PR. Anyway I do not believe 2.4-2.5 at 95W and 2.6-2.7 at 125W should be a problem.

Remember, current 65nm trade huge transistors(compared to standard 65nm ones) for time to market. Therefore they are skew limited.
As such only power envelope shoul prevent AMD going 2.8 or 3.0 at launch.
However, expect 3.0/125W and 2.7/95W on the dawn of 2007 at the latest.

Well, let's see.


RE: something fishy is going on
By mino on 2/6/2007 2:37:53 PM , Rating: 2
WOULD be disaster for AMD. Period.


By coldpower27 on 2/6/2007 9:14:30 PM , Rating: 2
We'll have to see, 2.3GHZ will likely hold it's own in it's intended TDP target range against the Xeon DP E5345 2.33GHZ of 80W. 95W to 80W is fair as when you factor FB-DIMM's as well as the chipset level memory controller, the power consumption will normalize in AMD favor. Even if the Bareclona core only has equal IPC compared to the Clovertown derivative, I expect Barcelona to fair a bit better once you take in scaling into account in the server arena.

Nice emphasizing, but Barcelona at these clock rates will be pretty competitive with Intel's offerings, not to mention they will beat Intel to Tigerton on the MP front.

When your talking about Servers you have to factor in scaling and from what we have seen K8 scales, a bit better compared to Core 2 when you in more then 1 Socket territory. Since K10 seems to be an considerably enhanced version of that, it will do fine.

Speculative clock frequencies aside, this looks to me a fine line up for AMD in the server front, they will survive just fine.

There is nothing in the roadmaps for speedbumps on the Server front from Intel so what they have now will have to last them until the Harpertown derivatives arrive. Which isn't till Q1 2008.



By Justin Case on 2/9/2007 3:39:41 PM , Rating: 2
> Unless cities cores are REAL monsters (>10% faster than c2 cores).

They are. On memory intensive applications, they're actually a lot faster than that. But expect Intel to have an answer to those about 4 months after they're released. They did learn something from the P4 fiasco.


Timetable
By JackPack on 2/6/2007 3:52:08 AM , Rating: 2
It shows the first Barcelona products being produced in "June '07" and being lanuched in "Mid 2007."

Uh, June is already mid-2007. They expect to get the parts out of the fabs and packed in a month? Yeah, right.




RE: Timetable
By retrospooty on 2/6/2007 11:50:18 AM , Rating: 3
Uh... Mid 2007 can be anywhere from April to September (even later if need be), its a general term. It is not a clear date that states it must be within one month of production.


RE: Timetable
By JackPack on 2/6/2007 4:58:42 PM , Rating: 1
That's clearly not true for this roadmap. They would have stated Q3 if that were the case.

The "Production" targets are very specific, e.g. June '07, yet the "Launch" date becomes so ambiguous? Come on.


Mistake?
By DeepThought86 on 2/6/2007 8:36:03 AM , Rating: 1
How could quad-socket parts (the 8000 series) be released with HT1? These are the ones which need HT3 the most. I think something's screwed up in this roadmap.




RE: Mistake?
By highlandsun on 2/6/2007 5:42:08 PM , Rating: 2
Yes, considering that all the current chips use HT2 why would they go backwards?


RE: Mistake?
By lplatypus on 2/6/2007 11:40:27 PM , Rating: 2
I guess you're thinking of inter-socket bandwidth? Some people care most about I/O bandwidth. A dual or quad socket opteron system can use 4 HT links for I/O (2 per socket for dual and 1 per socket for quad), whereas the single-socket 1000 series only has 1 HT link for I/O, so you could argue that the single-socket system is most in need of HT3.


SE and not SE
By AnotherGuy on 2/5/2007 10:22:11 PM , Rating: 2
what is the difference between the 1000 series SE and vanilla?

The 1000 series might actually be exciting for the non server market




RE: SE and not SE
By Anh Huynh on 2/5/2007 10:24:29 PM , Rating: 2
125W vs 95W TDP


confused about diff between 8000 and 2000 series
By RamarC on 2/6/2007 9:47:40 AM , Rating: 2
amd's websites says
quote:
2000 Series (up to 2P/4-core)... 8000 Series (4P/8-core to 8P/16-core)


does this mean the processors are identical performance-wise for each series, but the 8000 series supports more processors per system?




By kenji4life on 2/6/2007 10:25:54 AM , Rating: 2
I'm not positive about the performance but I believe your assertion is correct.


They need a few more model numbers in there :)
By 13Gigatons on 2/7/2007 6:20:43 AM , Rating: 2
Do they need so damn many model numbers ?

Do they need so many socket types ?

At least intel has milked 775 for a good amount of time, AMD has gone 754, 939, AM2, etc in the desktop market.




By Justin Case on 2/9/2007 3:55:04 PM , Rating: 2
You're kidding, right?

Why do you think AMD changed sockets (939 -> AM2) when it moved from DDR ro DDR2? So people wouldn't stick a CPU with a DDR2 controller into a motherboard with DDR, of course. What kind of advantage would there have been in keeping the same socket? Just a lot of dead (or at least non-working) chips.

The 754 / 939 / 940 distinction was mainly a commercial one, so people would have to pay a premium for dual-channel memory access and SMP support. You can't stick a pair of Pentium-4s on a Xeon board and have them work in SMP, either, so that's no different from what Intel does.

Sure, it was nice when you could use Athlon XPs on an MP board, but I don't see those days coming back.


Thanks
By Regs on 2/5/2007 10:07:14 PM , Rating: 1
I almost cried when I saw barcelona on a AMD road map like it actually existed.





RE: Thanks
By tuteja1986 on 2/5/2007 11:38:44 PM , Rating: 2
Why you don't think AMD will have barcelona out by this year ?


eh hm
By ElJefe69 on 2/5/2007 11:49:44 PM , Rating: 2
Does this mean that a socket AM2 will have a quad chip soon? would a 1207 board have any advantage if just using one chip? (typical enthusiast setup)

sounds schweeet. Next Christmas would be a great time to upgrade to quad core. 2.2 ghz dual core is fine enough it seems for anything I through at it. r600 ati card and quad core with 4 gigs ram. that would be some clean specs for a while I imagine?




I/O Bandwidth
By lplatypus on 2/6/2007 12:07:22 AM , Rating: 2
So a single socket system will have 30% more I/O bandwidth than a dual socket system which uses the hypertransport link of both CPUs... that's quite a reversal.

We heard that RD790 Socket F+ reference boards were already being sent to manufacturers:
http://www.dailytech.com/article.aspx?newsid=5647

I wonder what CPUs will be able to run on them? Will Athlon FX's support socket F+ before Opteron 2xxx's? This roadmap shows dual socket Opterons being made only for Socket F well into 2008.




120 Watts
By Lexington on 2/6/07, Rating: 0
RE: 120 Watts
By mrdoubleb on 2/6/2007 2:50:53 AM , Rating: 2
You do know that it's a quad core, right? So that's a 30W TDP per core MAXIMUM for the whole product line , under maximum load and power draw.

Per core this thing could go as mobile CPU.

And BTW, Intel has a similar TDP for their quad cores, only it's without the memory controller.


"When an individual makes a copy of a song for himself, I suppose we can say he stole a song." -- Sony BMG attorney Jennifer Pariser














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki