backtop

Print E-mail del.icio.us 88 comment(s) - last by overzealot.. on Dec 6 at 5:32 AM
AMD has something for everyone with its new tri-core processors, but quite a lot is still marked "tentative"

AMD picked up big headlines the day before Intel's Fall Developer Forum with the announcement of its upcoming tri-core processors. 

AMD's original release did not specify if this tri-core processor, code named Toliman, would be a totally new processor or merely a stripped-down version of the existing Agena core. The answer, it appears, is both.

In an embargoed corporate roadmap forwarded to DailyTech, details of these new triple-core oddities came to light.

The first triple-core processor, Toliman, is essentially a core-disabled version of the Agena quad-core processor. It includes a full Agena package, including the 2MB of shared L3 cache, with one core disabled.

Toliman, which will eventually herald the AMD Phenom 8000-product name, is scheduled to launch in February 2008 with mass availability in March.  AMD representatives, speaking on conditions of anonymity, confirmed the initial  2.4 GHz Phenom 8700 and 2.3 GHz Phenom 8600 tri-core processors will launch with a 95W thermal envelope.

In late 2008, AMD will shift almost all of its 65nm quad-core offerings to 45nm.  AMD will then follow up these initial quad-core offerings with 45nm dual-core and triple-core processors in 2009. 

The first of these 45nm tri-core processors, codenamed Heka, will launch with DDR2 and DDR3 support.  However, AMD guidance also details that Heka will ship with two different varieties: one with a shared L3 cache, another without.  All 45nm quad-core AMD processors incorporate shared L3 cache, with the exception of the Propus family processor.

AMD guidance goes on to state that all mainstream Phenom quad-core processors, both with shared L3 cache (Deneb) and without (Propus), shipped in 2009 will feature DDR3 exclusively.  Heka, on the other hand, will feature a mix of DDR2 and DDR3 support.

Unfortunately the answers for tri-core only raise further questions.  While Heka has a unique codename, it seems to be a combination of cut-down Deneb and Propus quad-core processors. The logical conclusion would be that Heka is merely excess or defective Deneb and Propus processors from the 2008 launch. 

Yet AMD's roadmap goes on to detail one more chip: Regor.  Regor, which has always been described by AMD as a dual-core version of Deneb, will make its debut with variable shared L3 cache and a mix of DDR2 and DDR3 support. Could it be that Regor is a core-disabled version of Heka, which is already likely a core-disabled version of Deneb/Propus?

One AMD representative declined to comment on these 45nm processors, stating that 2009 processor launches and specifications are still "tentitive."

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
Is this the road to profitability
By crystal clear on 12/4/07, Rating: 0
RE: Is this the road to profitability
By GeorgeOrwell on 12/4/2007 7:06:28 AM , Rating: 1
All AMD is shipping today is buggy chips that really shouldn't even be sent out to mainstream customers. Basic functions of the chip do not work properly.

Even the Barcelona B2 stepping -- which was supposed to boost performance and fix a lot of bugs -- ended up being really buggy itself and offered no performance increase. So there is a B3 stepping in the works that is the B2 bugfix.

Keep in mind that Barcelona, for all intents and purposes a successful prototype design but a failed production design, is being reworked for 2H08. You can call this new Barcelona, B+.

There is no compelling reason to buy Barcelona systems today unless you are an AMD partner and are developing software that takes advantage of a few things in the Barcelona design such as nested page tables (used for virtualization support).

Keep in mind that unless you have a motherboard with split plane power, it is kinda dumb to get any sort of Barcelona system anyway. The chip will not work the way it is intended to work. Barcelona is not backward compatible in any real sense.

Hence, if you need a new system between now and 2H08, Intel is the intelligent choice.


RE: Is this the road to profitability
By Chaser on 12/4/2007 9:19:13 AM , Rating: 2
And what "basic function" is not working on these "buggy" CPUs exactly?

A buggy CPU? Your points seem at least interesting but using the term "buggy" for a central processing unit is quite far fetched if not implying disaster.

I love these slippery slope "therefore go with these guys" pseudo ad campaigns.

I am sure these CPUs will function perfectly and be stable for their intended purpose. As always price/performance should be your primary consideration for you to make your "intelligent choice".


RE: Is this the road to profitability
By TSS on 12/4/2007 9:35:24 AM , Rating: 3
your a few hours behind so it's forgiven:

http://www.techreport.com/discussions.x/13721

AMD's not shipping their barcelona's anymore due to a TLB error. or so the dutch translation reads where i've read it, that was the source. so now their entire product line is grounded, not only the 2,4 ghz, because they all have it.

and even with a stable 2,4ghz, it's not better then an intel yet. and by the time it will be nehalem will be here.

i'm not a fanboy... i buy whichever processor is the best at the moment. i've had a pentium 3, pentium 4 (northwood, 2,53ghz. it was better then the xp's at the time), ahtlon 64 x2 and if i'd buy a new computer right now, it would be an intel. i simply think amd has gone "a bridge too far" with buying ati, releasing a new architecture too soon on a still not mature process (they aren't as good as intel in transitioning manufacturing processes, they need more time) and hoping for high yields.

intel's steaming ahead right now also pumping billions and billions into everything because they can easily afford to. AMD has said they are committed to intel's release scheme of ever increasing complex processors on ever more expensive node processes. they'd have only 2 years to get their R&D back on the previous generation of both.

if AMD is wise right now, they'll slow the hell down and fall back into the underdog position again. they will always have a place in the CPU market, and intel will themselves grind to an halt. their current pace is sure to win them a big market dominance but with the costs going up exponentially to keep moore's law in effect, AMD will be able to catch up eventually. but not if they run themselves aground like this.


By GeorgeOrwell on 12/4/2007 9:32:55 PM , Rating: 1
Thank you for stepping in.

From what I have heard, AMD is in big trouble because they cannot make a reliable chip. AMD's partners have lost a lot of confidence in the AMD management team. It will not be until 2H08 until Barcelona is fully reworked to become Barcelona+, B+, and ready for mass adoption.

AMD did make a very solid design prototype with Barcelona, but years later than they should have. Trying to rapidly patch a design prototype and then rush it into production produces what you see on the market: buggy chips that do not work well. Chips that are so bad you see the stop shipment. Do not be surprised at all to see AMD issuing a recall for all Barcelona chips and giving customers a free replacement. If AMD doesn't do it, expect a mass exodus to Intel.

The problem with the recall is that the motherboard specs also changed based on Barcelona changes. There are quite a few motherboards out there that will not work with all the revisions of Barcelona. This fully messed up compatibility is why you do not see a Tier 1 vendor with mainstream availability of Barcelona-based products.

AMD will be smaller and humbler 8 to 10 months from now when they finally get B+ shipping in volume as the reliable chip it should have been today. But at least there will customer choice and two reasonably decent alternatives in the market.

If the cards play out, AMD may even be part of a much larger company by then, a company more capable of doing the large scale chip design work that AMD is attempting to do. AMD would also benefit from a management team that is more capable of execution vs. focusing on things that don't matter (i.e. dual dual-core vs. native quad-core).

Native quad-core will not matter much for another 10 years, until all existing operating systems have been rewritten for greater parallelism.

Barcelona's design is a case study in mistiming the market.


RE: Is this the road to profitability
By Locutus465 on 12/4/07, Rating: 0
By Locutus465 on 12/4/2007 10:29:50 AM , Rating: 2
doh, meant to reply to chaser, sorry


RE: Is this the road to profitability
By Clauzii on 12/5/2007 12:42:59 AM , Rating: 2
They actually all do ;)

It's just that some erratas are basically non-happening incidents, and therefore not as important as the current AMD one.

Being it in the Level 3 cache is of a far greater concern than a mere listed errata. And the Level 3 cache should have been 2 MB PER CORE, not per CPU :-/..

Especially when it's 10% speed potential lost!

In an already "Lumping Jack Cash", AMD MUST find a cure for their problems soon, or the market ends up with ATI being sold to Intel or something - PURE speculation from my side, though.


By Clauzii on 12/5/2007 12:50:09 AM , Rating: 2
Sorry for the reposting, but 'something' got misplaced. Heres the right text:

They actually all do :)
It's just that some erratas are basically non-happening incidents, and therefore not as important as the current AMD one.

Being it in the Level 3 cache is of a far greater concern than a mere listed errata. Especially when it's ~10% speed potential lost, as of current.

Btw. the Level 3 cache should have been 2 MB PER CORE, not per CPU :-/. Really a shame they have to, maybe, wait for 45nm for to happen.

In an already "Lumping Jack Cash" AMD >must< find a cure for their problems soon.
--

Or the market ends up with ATI probably being sold to Intel or something - PURE speculation from my side, though.


RE: Is this the road to profitability
By Misty Dingos on 12/4/2007 8:32:47 AM , Rating: 3
What I think is the issue with this product line (if you can call it that, they are failures of one sort or another) is that there will be an inherent availability issue. Let me explain.

AMD acknowledges that these are manufacturing errors or quad core processors that just didn't quite make it. So they shut off one of the cores and still sell a three core CPU. Which is smart. They are selling a product and making a profit on something that was normally thrown away as scrap.

But the guys that make the processors are going to keep trying to improve the production process and this will inherently reduce the availability of the three core processors. And hopefully their production process is good enough to produce relatively few of the failed quad core turned three core processors in the first place.

This would lead me to believe that the number of three core processors available should be very limited. If they are limited in number and people want to buy them then that will only drive up prices on them. Then they become artificially expensive and price themselves out of their spot on the market.


RE: Is this the road to profitability
By Targon on 12/4/2007 9:08:41 AM , Rating: 2
The way the processor industry operates is that if there are not enough of the slower products available due to better yields of the higher end products, those perfectly good higher end products are clocked down and sold at the slower speed. This is why overclocking is so easy for certain processors, because so many good high-end chips are clocked down and sold at the slower speeds.

Now, if yields are so good that there is a shortage of tri-core processors, the price of the quad-cores will either come down to the point of eliminating the tri-cores or working quad-core chips may have a core disabled intentionally and sold as a tri-core.

I am waiting on the 2.8GHz and above Phenom processors to see how well they will perform. The extra cores may be nice for newer titles that use multi-threading, but I still like playing too many older games to make the lower clock rate a good thing.


By bradley on 12/4/2007 9:33:34 AM , Rating: 4
Yeah, exactly. This is standard practice in the chipmaking industry, including Intel. AMD is simply trying to maximize their profits based on yields. I'm surprised so many people on a tech savvy site wouldn't realize this.


By Hawkido on 12/4/2007 2:02:51 PM , Rating: 3
I kinda like your post... but you forget one thing.

What if all four cores work at the 2.4 Ghz speed. but 3 of the 4 will run within specs at 2.8 Ghz.

Whould it be smarter to sell it as a 2.4 Ghz Quad Core or as a 2.8 Ghz Tri-Core?

If it is not in a server enviroment, do you really need the extra core...

Plus 3 cores will be more compatible for Xbox 360 ports as the 360 has 3 cores.


Multi cores aren't the solution
By JoeBanana on 12/4/2007 12:07:35 PM , Rating: 3
No average user would use more than 4 cores. Multi cores are just a market move. A much better solution is one faster CPU.

Programs are executed in serial sequence and as such they don't gain much from multi cores. One may argue that we have thread programing. Yes but usually the one part of the program requires data from previous par t of the program.

Though it is true that if your OS supports multi core processors that each core can run a separate application but the average person runs low CPU intensive applications (OOo, MSN, FF, uTorrent, xmms...) which can be executed in serial in real time. But one could get some use in max of four cores at this point.

The fact is that the frequency border has been reached because processors produce too much heat. There is no room in assembly line time. The only real solution then is in next generation materials, new gates, wires...

Of course multi cores are good for specific operations, like servers that give access to hundreds of people.




By amanojaku on 12/4/2007 2:38:41 PM , Rating: 2
That's true for today's programs, but tomorrow's programs might be optimized for multi-threading.

But look at it from another point of view: should I run 60 processes on one CPU or 15 processes per four CPUs for a total of 60? A lot of OS inefficiency is due to the context switching between processes; by adding more processors and an efficient multi-threaded kernel performance might increase. The communication between cores needs to be efficient, as well, otherwise the bottleneck would just be changed from context switching to interprocess communication, which is worse.


RE: Multi cores aren't the solution
By TomZ on 12/4/2007 5:19:04 PM , Rating: 2
quote:
No average user would use more than 4 cores. Multi cores are just a market move. A much better solution is one faster CPU.

That would only be true if your computer could only run one thread and one process at a time.

In reality, most applications run a number of threads, and those threads in turn rely on operating services (e.g., reading a file) which also run some number of threaeds. So even a single application can keep 2+ cores busy.

And that doesn't even count background processes and device driver threads. For example, there's no reason that incoming and outgoing Etherhet packets can't be processed at a very low level by a different core, thus offloading the core running your "single" app and letting it run a little faster.

I'm running on a bit, but I hope you see my point.


By JoeBanana on 12/5/2007 5:38:44 AM , Rating: 2
Of course it's true that we can run more threads and more processes at a time. But does the average computer user need more than 4 cores.

One core can more than easily take care of all the background processes and also run a few low intensive applications in the background(and as far as I know the average user can be focused on 2-6 programs at the time the rest are sleeping). It's not like an average user will code video on one processor, code sound on other, run server on third, write something in OOo...

The term that if you double CPUs you double your program execution isn't true for an average user. Actually if you change from two core to more you wont even notice the difference.


RE: Multi cores aren't the solution
By Strunf on 12/4/2007 6:49:40 PM , Rating: 2
The application you point out could be run on a P3, so what's the point of having faster CPU for them?
Dual, Quad and more are "meant" to people that need more processing power, like video editing, games and others, this kind of programs can "easily" be multi-threaded.

And ray tracing is in the works which could be a major burden for CPU (and very multi-threaded).


I don’t get the point of this 3 core CPU
By Strunf on 12/4/2007 5:15:57 AM , Rating: 3
People speak of them as a “failed” quad core that instead of being thrown in the bin will be used, but the thing is that if let’s say 10% of the quad cores fail and become triple cores then there won’t be many of them out there to be of any significance and the lack of supply may hinder the image of AMD, if on the other hand AMD is making real triple cores then it’s a waste of money since these production lines could be used to make more quad cores and have them compete with Intel dual cores.
Also in 2009 dual-cores will be the budget CPU, the quad core will be mainstream intermixing with CPU with 8+ cores, that’s what I think Intel is planning to, and I don’t read anything about 4+ cores on the AMD side does that mean will get another 3 (or something) years of desert crossing?

IMO if AMD keeps going this way I don’t see them making a comeback anytime soon. This thing of buying ATI in the end just added another “rival” to AMD, and when a company is already taking a beating better not make more enemies, I could be wrong though and ATI be in the end the key for the AMD revival but as it is AMD will keep getting a beating from both Intel and nVIDIA for the next few years, or at least till the next generations of CPU and GPU see the light.