Sign in with
Sign up | Sign in
Your question

K8L Architecture Presentation by AMD in Moscow

Last response: in CPUs
Share
September 24, 2006 10:44:03 PM

http://www.digit-life.com/articles2/editorial/amd-guiseppe-amato-conf-part1.html
Wow! Some new info about the mysterious K8L!



Too bad, AMD are talking only about S1207. And some dissapointening benchmarks on the next page.
September 24, 2006 10:50:14 PM

All of that was presented at AMD Analyst Day in June. I guess people are now finally starting to believe it.
September 24, 2006 10:59:46 PM

No, it was presented in Spetember 2006. Check out the slides, some of them are changed
Related resources
September 24, 2006 11:10:20 PM

Quote:
No, it was presented in Spetember 2006. Check out the slides, some of them are changed



Of course they're changed. When Analyst Day happened Barcelona had not taped out. I would moreso say they have been expanded with more hard physical data.
September 24, 2006 11:24:11 PM

Quote:
http://www.digit-life.com/articles2/editorial/amd-guiseppe-amato-conf-part1.html
Wow! Some new info about the mysterious K8L!



Too bad, AMD are talking only about S1207. And some dissapointening benchmarks on the next page.


The article was very one sided in nature, it appeared to be a Intel bashing article than a objective technology article. It also outlined what AMD spin doctors will be presenting in our market soon enough.
September 24, 2006 11:28:00 PM

What else to espect from AMD?
Becouse they can't bring something better, they are bashing what is best. Anyway, from their presentation I've concluded that I've no chance to have K8L before 2008.
September 24, 2006 11:31:27 PM

Quote:
What else to espect from AMD?
Becouse they can't bring something better, they are bashing what is best. Anyway, from their presentation I've concluded that I've no chance to have K8L before 2008.


Better than what? They have 100% or more increases in CPUs other than the one in your sig. Minus the OC.
September 24, 2006 11:32:08 PM

Quote:
What else to espect from AMD?
Becouse they can't bring something better, they are bashing what is best. Anyway, from their presentation I've concluded that I've no chance to have K8L before 2008.


Integrity and honesty actually, but its business I suppose can't play nice when billions are on the line.
September 24, 2006 11:39:46 PM

Quote:
Better than what?

Better than the new architecture of their rival, which outperforms the best of AMD.
September 24, 2006 11:44:25 PM

I think BaronMatrix is right. I flipped through the original Russian version when it was released and I didn't see anything really new there. It does basically confirm though that the first Deerhound/Barcelona version will continue using 3 HT2.0 links instead of 4 HT3.0. That was pretty much expected in order to be compatible with current platforms. They look to allow Registered DDR2 800 though which would be nice.

In regards to AMD's fairness, it's actually quite confusing. On one side they are actually favouring Intel by using a 80W TDP for Cloverton, which is for the mainstream version, instead of using the 120W TDP of the high-end Clovertons. What's more, they use a 95W TDP for their quad cores which means they are comparing their own high-end quad core to Intel's mainstream ones.

On the flip side, they are comparing their Opteron 4P system to Paxville MP instead of Tulsa so that favours Opteron instead.
September 25, 2006 12:18:12 AM

It looke like KL8 will allow for higher clock speeds with the current X2. I still think thats its too early to say what exactly will be the jump with KL8. But its looking promising.
September 25, 2006 12:26:35 AM

I don't think it's as much K8L allowing for higher clock speeds as it is the 65nm process which would be fairly mature by the time K8L launches. It is possible though that AMD added a pipeline stage or two in order to make it easier to clock higher. Right now K8/Dothan and I believe Yonah are all 12 stage while Core 2 is 14 stage. I think it would be logical for AMD to add 1 stage, if they aren't adverse to the number 13, that way they gain a bit of scaling room and it'll probably be used to help in the decode/predecode stage in order to better make use of the 32-byte decode width.

Edit: I forgot to add that with 13 stages they'd still be under Conroe's 14, which although probably only a small architectural advantage, with the adversion to long pipelines after Prescott is probably useful as a marketing advantage.
September 25, 2006 12:28:25 AM

Quote:
Better than what?

Better than the new architecture of their rival, which outperforms the best of AMD.


Since Intel took about 3 years to respond to K8, AMD has 3 years to respond to Core 2.
September 25, 2006 12:33:02 AM

Quote:
Since Intel took about 3 years to respond to K8, AMD has 3 years to respond to Core 2.


That's retarded. AMD needs to respond as soon as they can to Core 2 so that they can turn a profit. Note that Intel has deeper pockets than AMD, so they could afford to wait three years. AMD does not have that same luxury.
September 25, 2006 12:34:39 AM

Quote:
It looke like KL8 will allow for higher clock speeds with the current X2. I still think thats its too early to say what exactly will be the jump with KL8. But its looking promising.


I would say that AMD is using 65nm to make quad core at the same power envelope as dual core. I expect them to go to about 3.6GHz with 65nm, but a 4 core Barcelona will really ouperform the dual core Opteron per clock. They will probably (IMO) come out at around 2.6GHz to start.
September 25, 2006 12:34:43 AM

Alot can happen or can be devloped in 3 years, on both sides of the aisle.
September 25, 2006 12:39:03 AM

Quote:
Since Intel took about 3 years to respond to K8, AMD has 3 years to respond to Core 2.


That's retarded. AMD needs to respond as soon as they can to Core 2 so that they can turn a profit. Note that Intel has deeper pockets than AMD, so they could afford to wait three years. AMD does not have that same luxury.

But AMD is used to running in the red. They will probably never have an unprofitable quarter again. IBM has already committed to socket royalties for 1207, with Sun and Cray right behind.

Opteron is the darling of the supercomputing world. FX owned the high end until Core 2 came out. X2 outperformed even P4 EE.

AMD has 3 years.
September 25, 2006 12:39:16 AM

Quote:
I would say that AMD is using 65nm to make quad core at the same power envelope as dual core. I expect them to go to about 3.6GHz with 65nm, but a 4 core Barcelona will really ouperform the dual core Opteron per clock. They will probably (IMO) come out at around 2.6GHz to start.


If AMD is using undervolting to keep QC in the same power envelope as DC, they probably aren't going to hit 3.6. That said, once we increase the voltage up to where it "should" be, these 65nm AMD chips should be superb overclockers.
September 25, 2006 12:42:10 AM

Quote:
I don't think it's as much K8L allowing for higher clock speeds as it is the 65nm process which would be fairly mature by the time K8L launches. It is possible though that AMD added a pipeline stage or two in order to make it easier to clock higher. Right now K8/Dothan and I believe Yonah are all 12 stage while Core 2 is 14 stage. I think it would be logical for AMD to add 1 stage, if they aren't adverse to the number 13, that way they gain a bit of scaling room and it'll probably be used to help in the decode/predecode stage in order to better make use of the 32-byte decode width.

Edit: I forgot to add that with 13 stages they'd still be under Conroe's 14, which although probably only a small architectural advantage, with the adversion to long pipelines after Prescott is probably useful as a marketing advantage.



I agree that AMD will not play clock speed games, but I think they will take advantage of superscalar functionality and produce more IPC rather than increasing stages. By averaging 4 IPC at similar clockspeeds they can immediately increase perf by 25%.
September 25, 2006 12:44:22 AM

Quote:
But AMD is used to running in the red. They will probably never have an unprofitable quarter again. IBM has already committed to socket royalties for 1207, with Sun and Cray right behind.

Opteron is the darling of the supercomputing world. FX owned the high end until Core 2 came out. X2 outperformed even P4 EE.

AMD has 3 years.


AMD used to running in the red? Their recent acquisition of ATI is going to hurt them bad financially.

Quote:
Opteron is the darling of the supercomputing world.

That would be the PowerPC chips.

Quote:
FX owned the high end until Core 2 came out. X2 outperformed even P4 EE.

Those aren't supercomputing processors.
September 25, 2006 12:44:26 AM

I see the next 3 years as being how small, how little energy you can use,how cool you can keep it, and how many core you can fit onto it. As well as the ddr3 support, but thats only a small part.
September 25, 2006 12:49:54 AM

Quote:
I would say that AMD is using 65nm to make quad core at the same power envelope as dual core. I expect them to go to about 3.6GHz with 65nm, but a 4 core Barcelona will really ouperform the dual core Opteron per clock. They will probably (IMO) come out at around 2.6GHz to start.


If AMD is using undervolting to keep QC in the same power envelope as DC, they probably aren't going to hit 3.6. That said, once we increase the voltage up to where it "should" be, these 65nm AMD chips should be superb overclockers.

Because 939\AM2 runs at 1.35V, 65nm should get down to 1.1V. That means that the circuitry will theoretically run at 1.35 safely. That should power up to 3.7GHz, while AMD will conservatively limit the ACTUAL clockspeed to around 3.4GHz (89W) stock with a higher envelope 3.6-3.8(98-120W).
September 25, 2006 1:50:57 AM

Quote:
Better than what?

Better than the new architecture of their rival, which outperforms the best of AMD.


Since Intel took about 3 years to respond to K8, AMD has 3 years to respond to Core 2.

But in two years Nehalem will be emerging with 45nm node, so AMD needs to work in a much smaller time table.
September 25, 2006 1:53:13 AM

Quote:
But AMD is used to running in the red. They will probably never have an unprofitable quarter again. IBM has already committed to socket royalties for 1207, with Sun and Cray right behind.

Opteron is the darling of the supercomputing world. FX owned the high end until Core 2 came out. X2 outperformed even P4 EE.

AMD has 3 years.


AMD used to running in the red? Their recent acquisition of ATI is going to hurt them bad financially.

Quote:
Opteron is the darling of the supercomputing world.

That would be the PowerPC chips.

Quote:
FX owned the high end until Core 2 came out. X2 outperformed even P4 EE.

Those aren't supercomputing processors.


AMD will have around a year before they need to start paying back loans. AMD now has plenty of partners who would put up cash for preorders and royalties. Plus the deal is set to complete at the end of October which means that along with fab bills, they will get ATi s revenue. A complete platform will allow them to sell a percentae of their chips to ODMs for laptops, which is a real growth industry right now.


Not to mention the ability to proide high end server chipsets for Torrenza.

Cray has some of the fastest per socket supercomputers using Opteron.
September 25, 2006 1:55:11 AM

Quote:
It looke like KL8 will allow for higher clock speeds with the current X2. I still think thats its too early to say what exactly will be the jump with KL8. But its looking promising.


I would say that AMD is using 65nm to make quad core at the same power envelope as dual core. I expect them to go to about 3.6GHz with 65nm, but a 4 core Barcelona will really ouperform the dual core Opteron per clock. They will probably (IMO) come out at around 2.6GHz to start.

I have to say 2.6 is really pushing it for the K8L, I see it being more conservative 1.6-2.4 maybe even 2.2. Additionally I think AMD will miss that thermal sweet spot they claim they will have in the early revision of the core.
September 25, 2006 2:04:02 AM

I disagree. The highest 90nm dual core is at 2.8 and 65nm will give them enough headroom to start at 2.6GHz with good yields. The curren arch revision runs at 2.0GHz min so 600MHz isn't too farfetched of an increase.
September 25, 2006 6:05:55 AM

Quote:
Better than what?

Better than the new architecture of their rival, which outperforms the best of AMD.


Since Intel took about 3 years to respond to K8, AMD has 3 years to respond to Core 2.

But in two years Nehalem will be emerging with 45nm node, so AMD needs to work in a much smaller time table.

Your nuts bro K8L arrives next year they dont need 3 years to outperform core2 and besides 4x4 will Be the answer ....about nahalem it will arrive in 2009.Intel must work on their quad cores because they had no match to AMD's True Quad core design compared to a two dual core in 1 package of Intel offering
September 25, 2006 6:11:49 AM

Quote:
Your nuts bro K8L arrives next year they dont need 3 years to outperform core2 and besides 4x4 will Be the answer

Your nuts bro, 4x4 won't be competitive with Kentsfield.

By the way...
4x4 is not an answer to Core2 Quadro... it's a hacked together job by AMD. Compared to C2Q, it's slower, hotter, and more expensive, and requires a motherboard replacement.

Quote:
Intel must work on their quad cores because they had no match to AMD's True Quad core design compared to a two dual core in 1 package of Intel offering

There is no performance penalty of using a double core vs a dual core. One of Anand's articles found that there was little difference in scaling between the PD and the K8, latency was even roughly the same when reading from each core's cache (the K8 does not BENEFIT in LATENCY from the memory crossbar!).

I expect K8L to outperform Core 2, but it's still in the air whether K8L can beat Core 2's 45nm shrink in mid 2007.

EDIT: found the article, it's Xbitlabs :-)
http://www.xbitlabs.com/articles/cpu/display/dualcore-d...
September 25, 2006 6:47:14 AM

Quote:
Better than what?

Better than the new architecture of their rival, which outperforms the best of AMD.


Since Intel took about 3 years to respond to K8, AMD has 3 years to respond to Core 2.

Has it?

Intel did not had debt higher than market cap these 3 years and had better average net margins with Netburst than AMD ever achieved with K8.

Now Core2 forced AMD to drastically reduce margins, if they do not release performance leader with high margins soon (that means sooner than in 2009), they are in trouble.

Yep, this is not fair. But bussines is not about being fair.
September 25, 2006 7:26:13 AM

I don't think he was arguing that the Power PC chips were in the top, I think it was more along the lines of the number of sockets rquired for those computers. Although, only #1 is that much higher than the other top 10's. Sure the #1 computer is fast, but it also used 131k cpus. But who cares?
September 25, 2006 7:41:49 AM

Quote:
The top supercomputers use PowerPC chips.
Link: http://top500.org


Burn. The Itanium's even on there before the Opteron.
September 27, 2006 6:31:14 AM

I fail to see how any of this is relevant. Who really cares what cpu the fastest super computer uses? Go G4!!
September 27, 2006 8:20:18 AM

You're logic is HUGELY flawed. Think through this carefully for me.

The AMD vs. Intel Anti-trust lawsuit is about the (supposed) monopolistic behaviors Intel used to keep AMD from growing as much as it could have had Intel not (supposedly) limited their growth. If AMD is correct and Intel did limit them, then AMD should have had MORE partners, MORE market share, and HIGHER margins now than it does.

You're saying that AMD has to be ALONE, have no support, and be losing market share to win this case. That's like saying you sprain the leg of an NFL quarterback a few days before a game, but they still kick your ass all over the field because you're just not as good as them.

Foul play needs to be punished. You don't have to be losing to have been a victim of foul play.
September 27, 2006 9:25:49 AM

You smell like troll and like sock puppet.....
September 27, 2006 2:26:09 PM

Quote:
Bravo. :) 
As I recall, K8 --- a great product, but at release did not blow out the competition quite the same way C2D has performed. The performance gap really began to widen between K8 and Netburst once AMD had 90 nm fully in gear (Summer 2005) and were able to bin out the higher clock speed.

Jack

And it will be the same when they outperform C2D; they don't have the power for a massive blow. They really need to tighten the roadmaps to have a photofinish with C2D because Intel in not going to make the same mistake again, they won't sit and wait.
September 27, 2006 10:12:38 PM

Quote:
In regards to AMD's fairness, it's actually quite confusing. On one side they are actually favouring Intel by using a 80W TDP for Cloverton, which is for the mainstream version, instead of using the 120W TDP of the high-end Clovertons. What's more, they use a 95W TDP for their quad cores which means they are comparing their own high-end quad core to Intel's mainstream ones.

On the flip side, they are comparing their Opteron 4P system to Paxville MP instead of Tulsa so that favours Opteron instead.




(Legend: Seems like kitchen chemistry without the ingredients).

Anyway, it sure looks strange to attribute the same TDP (38W) to different Intel chipsets (namely, on what concerns Dempsey & its MCH); and, last but not least, strange again when, in AMD's original display, http://www.amd.com/us-en/assets/content_type/DownloadableAssets/sales_sheet_opteron_rev_f_v8.pdf#search=%22clovertown%20northbridge%22, the link refers to nothing less than the Tech Report, on what concerns Intel data (had AMD nothing better to link from?!).
I mean, from an admirable company such as AMD, one would expect a bit more care towards the [interested] customers, on what regards fairness & the accuracy of the presentation, especially now, when such a [smaller but] resourceful company going through harsh times, could take advantage of its desirable humility, as a PR & Marketing weapon.
So much adjectivation might be pertinent, in the present case, since AMD is missing the point & loosing a chance: They can't argue against [Intel] facts; and, they seem to have upcoming interesting products, marketed the way others [Intel] like it to be: ill marketed.


Cheers!
September 28, 2006 4:33:50 AM

Not 'boohoo'ing and being the victim of Anti-trust are two different things. If you can run AMD better than them, go take over :p 
a b à CPUs
September 29, 2006 2:40:23 PM

Quote:
Better than what?

Better than the new architecture of their rival, which outperforms the best of AMD.


Since Intel took about 3 years to respond to K8, AMD has 3 years to respond to Core 2.

But in two years Nehalem will be emerging with 45nm node, so AMD needs to work in a much smaller time table.

Your nuts bro K8L arrives next year they dont need 3 years to outperform core2 and besides 4x4 will Be the answer ....about nahalem it will arrive in 2009.Intel must work on their quad cores because they had no match to AMD's True Quad core design compared to a two dual core in 1 package of Intel offering

Not wanting to offend but I believe you're incorrect in your assessment that K8L is to arrive next year. AMD's roadmap clearly calls for Brisbane and Barcelona to be AMD's new offerings in the 2nd half of 2007. K8L (Greyhound) is set to arrive in the 1st Half of 2008 (Jan-June 2008).
September 29, 2006 8:31:43 PM

Nice of AMD to make a presentation of the CPU at the one and only place it will be able to run without catching fire :lol: 
!