ToastMaster
Limp Gawd
- Joined
- Nov 2, 2005
- Messages
- 483
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
K8L Quad-Core Tapes Outs
- Thread starter ToastMaster
- Start date
EternityZX9
Limp Gawd
- Joined
- Sep 15, 2004
- Messages
- 236
hmmm makes me wonder when dual core K8L's will come out...
Good news tho.
Good news tho.
HighwayAssassins
Supreme [H]ardness
- Joined
- Mar 2, 2004
- Messages
- 5,210
I sure hope AM2 and K8L can give core 2 duo a run for its $$!!
Im happy with my shiny new X2 though
Im happy with my shiny new X2 though
pxc said:
Historical evidence tends to support the notion of about 6-9 months from tape-out to market. We can look at clawhammer, then a little later new castle, a little after that is was venice and san diego. The most recent round was Manchester and Windsor....
All suggest 6-9 months....
Toledo being the exceptoin of about 10 months....
tayunz
[H]ard DCoTM October 2018
- Joined
- Sep 27, 2004
- Messages
- 1,343
good news.
Asian Dub Foundation
Supreme [H]ardness
- Joined
- Aug 1, 2005
- Messages
- 7,940
good for AMD
There's a difference between a respin/shrink and a new die. K8L is a new die, which is why AMD is referring to it as a next generation core, and it will be on a so far unreleased new process. I was being generous with the low end of my estimate. You can compare K8L to Toledo or 65nm to exclude the initial problems with SOI on Hammer/Clawhammer.duby229 said:
Hammer
tape out: Nov 2001
release: Apr 2003
time: 18 months
Clawhammer
tape out: ??
release: Sep 2003
time: ?? months (release was delayed several times, by more than 10 months. tape out to release was longer than 18 months)
Toledo
tape out: Jun 2004 (tape out announced on June 1, 2004, may have taped out weeks before)
release: Jun 2005 (5/31/05)
time: 12 months
K8 65nm
tape out: Oct 2005
release: Dec 2006
time: 13-14 months (scheduled)
It's a fantasy right now to still cling on to hope that K8L will be out Q1'07.
ToastMaster
Limp Gawd
- Joined
- Nov 2, 2005
- Messages
- 483
pxc said:There's a difference between a respin/shrink and a new die. K8L is a new die, which is why AMD is referring to it as a next generation core, and it will be on a so far unreleased new process. I was being generous with the low end of my estimate. You can compare K8L to Toledo or 65nm to exclude the initial problems with SOI on Hammer/Clawhammer.
Hammer
tape out: Nov 2001
release: Apr 2003
time: 18 months
Clawhammer
tape out: ??
release: Sep 2003
time: ?? months (release was delayed several times, by more than 10 months. tape out to release was longer than 18 months)
Toledo
tape out: Jun 2004 (tape out announced on June 1, 2004, may have taped out weeks before)
release: Jun 2005 (5/31/05)
time: 12 months
K8 65nm
tape out: Oct 2005
release: Dec 2006
time: 13-14 months (scheduled)
It's a fantasy right now to still cling on to hope that K8L will be out Q1'07.
People who hold on to Q1'07 are simply kidding themselves. If AMD can really push to release it and if everything just happens to go their way and no major or time-consuming problems develop, I think they could possibly see it arrive sometime in Q2'07, though almost certainly towards the end of that (or early Q3).
I can understand your misgivings... It is certainly one of those "grey" area's, however you have to take into consideration core revision of each chip....
Yes it is true that A0 was taped out in 01... But it is also true that c1 taped out on q3 02, and was launched in q2 03...
It is also true that 64nm taped out q4 05, but it is also true that it almost certainly isnt the same core revision.. although that is one of the grey area's becouse AMD hasnt released any info on how the core revisiojns for 65nm has progressed. Who knows how many of those prior revisions were incomplete?
It is also true that AMD has had several tape out of k8l this year, though none of them were production quality. This is supposedly a production quality tape out... Which means that it should bve ready for market within the next 6-9 months,most of which (at least 4 months) is ramp up..... Based on past tape out to market schedules.
As evidense would suggest we already have several die plotsa for K8L.... Which means that it has taped out in some form in the past, whether they are engineering tape outs or not doesnt matter... The only thing that matters is the production quality tape outs... Which this is supposedly.
Yes it is true that A0 was taped out in 01... But it is also true that c1 taped out on q3 02, and was launched in q2 03...
It is also true that 64nm taped out q4 05, but it is also true that it almost certainly isnt the same core revision.. although that is one of the grey area's becouse AMD hasnt released any info on how the core revisiojns for 65nm has progressed. Who knows how many of those prior revisions were incomplete?
It is also true that AMD has had several tape out of k8l this year, though none of them were production quality. This is supposedly a production quality tape out... Which means that it should bve ready for market within the next 6-9 months,most of which (at least 4 months) is ramp up..... Based on past tape out to market schedules.
As evidense would suggest we already have several die plotsa for K8L.... Which means that it has taped out in some form in the past, whether they are engineering tape outs or not doesnt matter... The only thing that matters is the production quality tape outs... Which this is supposedly.
EternityZX9
Limp Gawd
- Joined
- Sep 15, 2004
- Messages
- 236
As long as K8L's are offically "released" by May/June next year...meaning I can pick it up by July/August from newegg or zzf or whatever at a non-price gouging expense...I'll be a happy camper
...yeah..*crosses fingers on that one*
...yeah..*crosses fingers on that one*
1Q2007 is wishful thinking. 3Q2007 is pessimistic. I've read rumors that K8L taped out in may 2006 from a number of sources. This combined with the fact that the tape out probably happened some time in the past and is only now being announced, on the launch day of the Socket-F Opterons, means that 2Q2007 is probably right on. This also aligns well with AMD's own release time frame of mid-2007.
The only thing that interests me is that this is the quad-core tape out. AMD added a L3 cache to the quad-core K8L version to reduce the cache coherency traffic over the HT links. The dual-core version does not require this change and should be much simpler to produce. The shared L3 cache and arbitration logic should put the quad-core K8L behind the dual-core K8L. I wouldn't be surprised if we saw dual-core server and desktop K8Ls a little early; perhaps early 2Q2007 as opposed to late 2Q2007 for the quads.
The only thing that interests me is that this is the quad-core tape out. AMD added a L3 cache to the quad-core K8L version to reduce the cache coherency traffic over the HT links. The dual-core version does not require this change and should be much simpler to produce. The shared L3 cache and arbitration logic should put the quad-core K8L behind the dual-core K8L. I wouldn't be surprised if we saw dual-core server and desktop K8Ls a little early; perhaps early 2Q2007 as opposed to late 2Q2007 for the quads.
dave343
[H]ard|Gawd
- Joined
- Oct 17, 2000
- Messages
- 1,869
what does tape out mean?
It means the design is finished, polished, and sent off to the fab for manufacturing. This is similar to a game or other software "going gold". What this means is that the only stage left is the testing and tweaking involved in manufacture. This means that the quad-core K8L design is finished and the R&D teams have (for the most part) moved on to design the next generation CPU.dave343 said:
This is a huge step and implies many things. It implies that the K8L has gone through all the design changes that will take place before the first version is sold. It implies that any critical logic flaws have been sorted out and solved, and it implies that, essentially, the K8L is done.
The next step is for AMD to build masks and ramp the design. While this step isn't easy in absolute terms, it is very easy in relative terms. Compared to the time and cost required to do the core design (2-4 years), the manufaturing stage is much, much easier and less time consuming (8-12 months).
----
I always laugh at people who say AMD has been "resting" and not doing anything over the past years. These designs take a long time to make. This is a great step for AMD; they just beat Intel to tape-out of a true quad-core chip. That is no small feat.
visaris said:
I agree that AMD has certainly not been resting, the Core Duo was enough to scare them, the Core 2 Duo puts them in the rear for performance and performance per dollar in the desktop realm, their design, R&D, and marketing teams would have all had to be spending their time smoking up instead of working to not see the threat.
That being said, Intel's Quad-Core chips are set to launch Q1 of '07, and after the NetBurst debacle Intel does not want to fall behind again. Even if most people still don't trust the AMD name outside of the enthusiast market, it is the most vocal market out there, and word-of-mouth advertising is becoming more and more important. The last thing Intel would want would be for AMD to be considered a serious processor manufacturer by the general public (which face it, their not, anything you read on tech websites is read by 10% of the public at best, and that is wishful thinking).
With any luck K8L will offer significant performance increases over the regular K8 clock for clock core for core. If not, AMD has to rely on HT to give higher performance through bandwidth and Intel is still a bit away from releasing an integrated memory controller into any of their chips/chipsets.
I am all for competition, it drives both sides to work harder, I just think it is a bit early to be optimistic. I believe in real world benchmarks first (and gaming ones last on the list of real world ones, CPU technology has exceeded GPUs for the past year at least as far as that goes).
visaris said:
So let's see, sticking the word "True" in there changes nothing. Intel Showed off Quad core in Feb-06. That's like saying Current Dual Cores AMD's makes are NOT true dual cores then, right?
K8L for desktop will ship long (estimates show 6 months) after K8L for servers have been on the market. You might be able to buy K8L in June or July 07. The problem is they'll be expensive Server models, NOT desktop versions. More wishful thinking?
Donnie27 said:
Well, as much as I am an Intel fan, I hope AMD makes out with something good. We have seen E6300s hitting 3.4 with ease, and while the top of the line Core 2 Duo is what, 2.66 the top of the line Woodcrest is 3.0, and since they are basically the same design you know there is headroom left. If AMD can come up with something competitive we will see lower prices and faster Intel chips, so everyone wins.
Not at all. Intel showed off a MCM, or a single package with two dual-core dies. This hardly requires any design changes as it is essentially two unchanged Conroe/Woodcrest dies connected by the FSB. I am unbiased and perfectly fair in this reguard. If AMD came out with a chip that was simply two dual-core K8s glued together by HT and all in one package, I still wouldn't call it a quad-core chip. It wouldn't be, it would be a MCM.Donnie27 said:
No. AMD's current dual-core chips are composed of a single peice of silicon with two dies connected by the SRQ. This makes them true dual-core. Intel's Conroe is two cores on one peice of silicone connected at the L2 cache. This makes Conroe a true dual-core chip. You know this as well as I do; why do you even bother asking?Donnie27 said:
----
I make no claims about performance here. I'm only talking about the tech. Clovertown and Kentsfield may be uber-fast and may even beat AMD's solution. No one knows because the chips are not out yet. Even if they are faster, that doesn't make them quad-core chips. They are MCMs that present two loads on the FSB. I don't care how you and Intel try to spin it. Clovertown and Kentsfield are not quad-core. Intel's quad core chips are still a long way off based on their own roadmaps.
You really like to troll my posts. I can tell you get a kick out of it. I don't care about desktop parts. I run a dual Opteron 265 now, and I'm not going to step down to a one socket system any time soon. That said, a dual-socket, dual-core K8L system interests me. While I don't know the order AMD plans to introduce dual/quad parts (no one here knows for sure) it would be a reasonable assumption that AMD could make the dual-core part available sooner. With half the cores and no shared L3 cache to implement AMD could release dual-core K8L parts sooner than the quad-core parts if they felt like it.Donnie27 said:
Do they feel like it? So far reports say no. We'll have to wait and see.
NulloModo said:
Cloverton and Kentsfield are indeed MCM's.... I'm sure performance will scale well, and it will be a decent design, but they are NOT quad core.... The ARE MCM's... Period. Nuff said. No arguments...
With that sais, Intel has a much better design on there hands right now, and it will be interesting to see what K8L will do in comparison.
tHoM said:
I doubt it, but the gap will be closed and I think we'll have a fairly close race, unlike how it is today.
A lot of the changes in K8L mimic those made in Conroe (including the 4-issue core and load reordering), so that's sure to make up some ground.
NulloModo said:
The last thing I want to see is Intel feeling smug as a bug and getting lazy with Core
I already said I got my fingers crossed that K8L kicks ass when it hits the market. Yet, you still have folks playing semantics with "true" and "first" and yada yada! Frackin' Quad Core is quad core=P
If one thing I've always said, that's competition is a two way street. AMD already proved a propensity to Rape us when given the opportunity. If folks leaned NOTHING from X2, market leading Opterons and etc., that FACT was slammed that home like a Sledge Hammer, pun intended
I'm hangin' on to my A8N SLI and hope like hell K8L works on it.
visaris said:
Please try that on some one else? The Layout AMD showed is FOUR cores GLUED together using direct connect. How many L2s? They're less native or "true" than Intel two Dual Cores? Why? Intel's DUal core are more DUAL than AMD's LOL! Sorry but the industry thinks they're Dual core even when you think they're less "true" than AMD.
http://news.com.com/2100-1006_3-6038148.html?part=rss&tag=6038148&subj=news
Linkings are different, that's why
Since the AMD DC are connected to eachother with going to the crossbar, they're LESS Dual Core than C2D that can talk to eachother without going that far. Advantage of a ONE L2 vs TWO L2s. On a QUad-COre Package that 2 L2s vs what, 4? and then an L3 that eats more cycles and add latency?Their Desktop use becomes more of a Gimmick than anything. Dual core version will more than likely follow much latter. AMD needs money, not friends and fans.
Oh it's me and Intel oh WOW! Hey Poncho, I want my frackin' cut LOL! I don't know how any of this will work out either and I hope AMD is paying you for your FUD and cheerleading
? Colverton and Kentsfield is just as much Quadcore as K8L=PI don't like to troll your posts. The problem is your views of AMD is always so damed sun shiney and bright, even when reality is DIMM, that it's a shame! There's no way in hell folks really think AMD was just sitting there doing nothing while Intel worked on Core. For you to even say something like that is wild. They (AMD) saw the punch coming but couldn't duck in time. AMD might have gotten knocked silly but they're now dancing around the ring so they don't get popped again. Only Crazy folks think Intel knocked them out. WIth only a million Conroes shipping the next 7 weeks, all Intel can do is Jab!
Call it what you want, but the fact remains that Intel's architecture still is an MCM... Period...
It doesnt matter how many cores there will be... It matters how well they will scale. You can hang yourself up on semantics all you want, but you cant get past the facts that Intels MCM is not quad core... It is two dual core chips on an MCM package.... Period...
In addition to simply being two dual core chips on an MCM package, the MESI traffic is forced through the GTL+ buss... Which is already severely bottlenecked....
Besides... I think you may want to go back and take a look at the title of this thread... This is not the Intel forum, so you may want to go troll on the other side....
It doesnt matter how many cores there will be... It matters how well they will scale. You can hang yourself up on semantics all you want, but you cant get past the facts that Intels MCM is not quad core... It is two dual core chips on an MCM package.... Period...
In addition to simply being two dual core chips on an MCM package, the MESI traffic is forced through the GTL+ buss... Which is already severely bottlenecked....
Besides... I think you may want to go back and take a look at the title of this thread... This is not the Intel forum, so you may want to go troll on the other side....
LstBrunnenG
Supreme [H]ardness
- Joined
- Jun 3, 2003
- Messages
- 6,676
Donnie, I'm going to have to disagree with you on this one. P-D was two dies on a package, while Core 2 is a single die with two cores. Which would you rather have, a P-D or a Core 2? Also, what makes K8L better than Kentsfield is that, as far as I know, AMD's direct connect runs at the same speed as its cores. The two dies on a kentsfield have to communicate through the FSB, which runs much slower than the processor.
I suppose it's a bit harsh to say Kentsfield isn't quad-core, but it really is pretty much the same thing as a dual Woodcrest system on one chip. On the other hand, my X2 or your Core 2 is faster than, dual optys in my case or dual Core Solos in your case because they aren't bottlenecked by the need for external communication.
Bottom line: Kentsfield will rock, K8L will be a bit better, but it won't be out for almost a year yet, so why fret? Whether or not you call it quad-core is up to you, but I think we can all agree that when Intel or AMD engineers a quad-core chip with all the cores on one die, it will be more efficient.
I suppose it's a bit harsh to say Kentsfield isn't quad-core, but it really is pretty much the same thing as a dual Woodcrest system on one chip. On the other hand, my X2 or your Core 2 is faster than, dual optys in my case or dual Core Solos in your case because they aren't bottlenecked by the need for external communication.
Bottom line: Kentsfield will rock, K8L will be a bit better, but it won't be out for almost a year yet, so why fret? Whether or not you call it quad-core is up to you, but I think we can all agree that when Intel or AMD engineers a quad-core chip with all the cores on one die, it will be more efficient.
LstBrunnenG
Supreme [H]ardness
- Joined
- Jun 3, 2003
- Messages
- 6,676
duby229 said:
I see people (like Chemist_Slime, for example) call their Dual Dual-Core Opty systems Quad Core all the time. This is semantics. Kentsfield is quad core (literally translated, four cores), though it's not the most efficient design it could be.
LstOfTheBrunnenG said:
Its just that simple, it's one package with 4 cores so it is Quad Core=P You're right NO ONE knows the performance of the K8L and since there's nothing close to it feature wise from AMD, there's no way to even estimate how much better or whatever. For AMD, there is no Yonah before Conroe for example. Taken from a line by Coach B. Philips, "Yonah knocked on the door of the mighty X2, Conroe kicked the sumbee-youch in"!
There's also little reason for folks to get pissed at eachother. I bought AMD in Nov-05, I bought Intel in Aug-06 and if AMD K8L Dual Core rocks, I'll be looking to buy it as well. So sue me for riding the fense?
Each of the Dual Cores on the Quad Core Kentsfield have 667MHz FSB to play with. It's not like its bandwidth starved
Since these will have the same smart cache as my E6600 and smart memory acess, this means FSB and its speed or bandwidth is less important.The next Intel Server chips ship with CSI so you'll have faster and smarter buses.
duby229 said:
The end result is it is still Quad Core, it still was taped out in 2005 and demoed in Feb-06. Those are facts and are not to be confused with Fan's wishful thinking and feelings=P
You know NOTHING about GTL+ or any of its follow on iterations. If you knew what you're talking about, I wouldn't be writing this to you on my week old E6600 LOL!
LordGardenGnome
Weaksauce
- Joined
- May 5, 2005
- Messages
- 78
Well I've been reading this whole article and the Dual socket dual core? or Quad core. I'm in the navy on a submarine and well after this last patrol i'm trying to catch up on what's going on. I really perfer AMD over intel. So I was wondering if you someone can give me a good explaination. I build my own systems and was planning my next upgrade to be the K8L but i guess i don't know really what the K8L.
(cf)Eclipse
Freelance Overclocker
- Joined
- Feb 18, 2003
- Messages
- 30,027
what makes you say that? amd got it's AM2 offerings out before F, 90nm hit desktop before server, as did revE, so server first always doesn't always hold true.Donnie27 said:
Donnie27 said:
Donnie.... You really dont know what I know...
I'll just be polite and say that I have been in this business longer then GTL was released on socket7 That was a long time ago... A long time...
If anything else, lets just say it is time...
(cf)Eclipse
Freelance Overclocker
- Joined
- Feb 18, 2003
- Messages
- 30,027
i know, but whatever, donnie and i sorta have a mutal unspoken agreement not to flame each other so much, since i recall several heated debates in the past that didn't turn out well. key is in disagreeing while leaving NO room for any twisting/flame bait/trollingduby229 said:
ThatsAgood1jay
[H]ard|Gawd
- Joined
- Mar 21, 2006
- Messages
- 1,029
45 nm tech proc`s. = OMFGWTFBBQSAUCE PWNAGE.1!@!$$%$^$^&%*&*)*(()&*(*(%*#!
visaris said:
You and the AMD marketing department.
As long as a CPU with n cores uses the same socket as previous models and it is drop-in compatible it is an n-core.
For example Paxville is a single die dual core chip and IIRC single load on the FSB because it has a buffer between the 2 cores.
You use the word connected here as though it means the same thing.Only a fool would consider equal complexity in a comparison between the SRQ and a shared L2.
Not only that , but Conroe has also direct L1-to-L1 transfers.In a sense , I could easily claim that X2 isn't a true dual core simply because it far behind Conroe with regards to core integration.