Collection of Next-Gen AMD Processor Data

[MaGiC_MaN]

I see there is a Next-Gen topic for Intel, I think these forums could use one for AMD (make sure we stay neutral, ).

I'll start with K8L:

Quad-Core, w/ Shared L3 Cache for all 4 Cores
HyperTransport 3.0 - Providing 40GB/s Bandwidth per Link running at 5.2GHz (16x16)
Improved CrossBar and Optimizations for 65nm SoI and beyond
Support for DDR2, DDR3, and FB-DIMM's (End-User has choice of what to use)
Native support for DDR2-800 (Support for DDR2-1066 is probable)
4-6 Issue Wide Core
1/2 Cycle SSE processing (2 per Cycle) (The slide left me confused a little, correct me if I am wrong)
Additional SSE Instructions and registers
Power Consumptions of 50w or less

Now for K10:

HyperTransport 3.0/4.0
DDR3/DDR4 Support (Possible Rambus Support as well)
Quad-Core or 8-Core support

There isn't alot out there for K10, anybody with information please post so I can know. I know it's my first post, and I need to grow Tenure, but I hope it's a good start, thanks.

 

[k2000k]

Any roadmap plan of when they should be released, and the progress AMD has been making towards fulfilling that roadmap?

 

[dmdallas]

I think that this information about the k8l is pretty readily available on the internet, though I haven't heard about some of these things, for a first post, pretty good, but at least provide links to your sources so that poeple know you aren't just pulling this stuff out of your butt.

 

[MaGiC_MaN]

Any roadmap plan of when they should be released, and the progress AMD has been making towards fulfilling that roadmap?

Since AMD just announced the product that is under development, and I believe their release date is Q1 07 at the latest is pretty solid, add in Chartered SemiConductor, FAB36, and the 2 FAB's their working on now, it's safe to say K8L should be on time.

Their roadmap on their website can answer your questions.

!~Da MaGiC MaN~!

 

[MaGiC_MaN]

I think that this information about the k8l is pretty readily available on the internet, though I haven't heard about some of these things, for a first post, pretty good, but at least provide links to your sources so that poeple know you aren't just pulling this stuff out of your butt.

I figured most people know already but just wanted a place for collection. The only links available are from TheInquirer.net, but I guess it's as good as any: http://theinquirer.net/?article=31761

!~Da MaGiC MaN~!

 

[iterations]

Great! Keep up the good work, and be sure to try to provide links to reputable sources for your additions to the thread.

In particular, I'm curious where you got the 50W figure for a quadcore K8L. I did not think there was any public info on the TDP design for this future chip. Could you please provide a link?

Thanks and cheers!

 

[k2000k]

Is the K8L just a paper proc? Is it being engineered as we speak? Will there be an engineering sample soon? I would like to see one before the Conroe came out so we could get a better feel towards what we should be saving up for.

 

[BaronMatrix]

Is the K8L just a paper proc? Is it being engineered as we speak? Will there be an engineering sample soon? I would like to see one before the Conroe came out so we could get a better feel towards what we should be saving up for.

Alas, it is currently on the drawing boards -- we have no functional silicon to use. Target release date is Q1 2007 to Q3 2007, there have been contradicting reports on the roadmaps which are all based on asian vendor leaks and not directly from AMD. Hard to tell at this point, but Q1 2007 is likely in my opinion.

As more info builds in we will get more details. I am curious though, it specifically called it quad core, but AMD's roadmaps show dual cores through most of next year -- not sure, does that mean dual cores remain K8 and quads are only K8L?


Actually, it's due out to OEMs in Dec, and being the type of consumer who doesn't doubt the ability of Core 2, I don't doubt that K8 isn't dead and that with these improvements applied to it, it will be 40% faster or more.


Also, I hope they apply "L" to the dual cores, too. The desktop is not ready for quad yet. Maybe CAD/graphics workstations but not PCs.

 

[Action_Man]

I nominate 9nm. I think he can bring a lot to this project. Hes a very neutral sort of guy.

 

[Action_Man]

Yeah K8MAN would be good.

Me? If someone else wants to help me I'll do it otherwise I'd have to think about it if its just me.

 

[Pippero]

4-6 Issue Wide Core

Where did you get this?
I have doubts if they'll do 4 issue, and 6 issue doesn't make any sense without some sort of hyperthreading.
There just isn't that sort of parallelism available in the programs..
But i don't really think that 4 issue is what you need to match Conroe's performance.. even Netburst was wider than K7/K8, being able to issue and execute 4 integer instructions (simple, but common ones like ADD) per clock, if they were in the trace cache.
Yet they had to use hyperthreading to take advantage of the wide execution engine...

1/2 Cycle SSE processing (2 per Cycle) (The slide left me confused a little, correct me if I am wrong)

My understanding, is that they'll be able to process 128bit wide in a single cycle, so 2x the current K8, and as fast as Conroe.
But the above statement is a bit ambiguous, as it sounds like it'd be possible to process 2 128bit instructions in a single cycle, so a 4x speedup compared to now, which i find kinda unlikely.
What would be interesting to now, is if they're gonna change the length of the pipeline.. cause at the moment, it looks like K8L can match Conroe in IPC (instructions per clock), but will be at a disadvantage in terms of clock frequency.

 

[zarooch]

Here is AMD's roadmap...

AMD Roadmap

 

[jkflipflop98]

Here is AMD's roadmap...

AMD Roadmap

If that is accurate, that puts the K10 @ 1H'08. And Rev.G aka K8L hitting the streets 1H '07. If they don't make that timeline, then someone needs to take that horse out behind the barn.

 

[zarooch]

Here is AMD's roadmap...

AMD Roadmap

If that is accurate, that puts the K10 @ 1H'08. And Rev.G aka K8L hitting the streets 1H '07. If they don't make that timeline, then someone needs to take that horse out behind the barn.

well i can't gurantee its accuracy but... if AMD doesn't follows that roadmap they'll be completely left behind Intel... anywayz... time will tell.. BTW why isn't this post being updated by AMD Next-Gen Info...?

 

[clue69less]

Yeah K8MAN would be good.

Me? If someone else wants to help me I'll do it otherwise I'd have to think about it if its just me.

Specifically what help do you want?

 

[ethernalite]

That is certainly true - AMD needs to have the 65nm K8L out in Q1 '07 or else they are in serious trouble. I'm not a fool that says AMD is going to go out of business because of a few consecutive quarters of poor revenue, but 65nm is key for AMD.

AMD fanboys claim that AMD doesn't "really" need to deploy 65nm, because they are simply perfecting the 90nm tech they have now, and there is no point in incurring extra costs when they don't need to. But that is a falsehood. Even if 65nm provided no performance gain, it is still very important, just like any other process shrink. A process shrink from 90nm to 65nm results in a die (roughly) half the size of a 90nm version. For the same silicon waffer, you can now get around two times the number of processors as before if you make no changes to the die size.

This is simple economics. This is how Intel is now able to slash their prices on the Presler and related cores and leave AMD in the dust. Intel uses a lower cost (see JumpingJack's fab manufacturing posts), smaller process technology. If Intel and AMD get into a price war, process size is key, and Intel will win it very time.

AMD used to only be about a year behind Intel in process shrinks, but that gap just increased to 18 months at the very least. Intel will be deploying the 45nm process in 2H 2007.

If AMD slips on K8L release, or it hits late 1H 2007, they are certainly in trouble. As I said, it isn't like they are going to go out of business. They operated in and out of the red for the past decade. However, they do have some major costs coming up - namely the depreciation of their new fab. AMD simply doesn't have the profit margins Intel has. Intel, even in the "atrocious" last 4 quarters, had a 22% net profit margin. That's not gross margin, that's NET. They made over 8 billion dollars last year. AMD, in their "excellent" first quarter had only a 14% profit margin, and that was without the depreciation of the new fab. If that had been included, the profit margin would have been around 3-5%. AMD simply can't deal with a price war, especially if they continue to lag so far behind Intel in the process size department. Simple as that.

 

[Mardark]

Here is AMD's roadmap...

AMD Roadmap

If that is accurate, that puts the K10 @ 1H'08. And Rev.G aka K8L hitting the streets 1H '07. If they don't make that timeline, then someone needs to take that horse out behind the barn.

From what I read, Rev. G is just the die shrink from 90 to 65 nm, its not K8L. But I could be wrong. Sometimes my reading comprehension is lacking. Can't find the article where I read that, or would provide a link.

 

[Pippero]

If that is accurate, that puts the K10 @ 1H'08. And Rev.G aka K8L hitting the streets 1H '07. If they don't make that timeline, then someone needs to take that horse out behind the barn.

That's wrong.
Rev.G != K8L (source Dailytech)
http://www.dailytech.com/article.aspx?newsid=2489
Rev.G is only a die shrinked version of K8, with support for quad-core and possibly some (yet unknown) improvements for the high end line (Opterons).
So, if that roadmap is correct, we won't be seeing K8L until 2008.
As for K10.. it's still high in the clouds.

 

[clue69less]

If Intel and AMD get into a price war, process size is key, and Intel will win it very time.

Right now that is true. The longer AMD stays in the business, the more that gap could close.

AMD used to only be about a year behind Intel in process shrinks, but that gap just increased to 18 months at the very least. Intel will be deploying the 45nm process in 2H 2007.

From what I hear, the commercialization of 45nm at Intel is by no means a done deal. Roadmaps are in reality goal sets - they slide and/or are abandoned from time to time.

 

[K8MAN]

Yeah K8MAN would be good.

Me? If someone else wants to help me I'll do it otherwise I'd have to think about it if its just me.

How about we wait until after June 1st when we supposedly will get the scoop.

 

[ethernalite]

If Intel and AMD get into a price war, process size is key, and Intel will win it very time.

Right now that is true. The longer AMD stays in the business, the more that gap could close.

AMD used to only be about a year behind Intel in process shrinks, but that gap just increased to 18 months at the very least. Intel will be deploying the 45nm process in 2H 2007.

From what I hear, the commercialization of 45nm at Intel is by no means a done deal. Roadmaps are in reality goal sets - they slide and/or are abandoned from time to time.


That gap can't close as long as AMD gets progressively farther behind in process technology, and Intel doesn't make a major goof with their architecture like they did with Netbust.

As far as Intel missing their 45nm ramp - I doubt it. Intel is usually pretty good at beginning new process sizes on schedule, once every two or so years. Remember, they've been doing 65nm since mid-2005. The only issue will be with how good their yields are. One has to wonder, considering that they will still be using 193nm dry litho, the same litho tech they used on 4 process generations ago. Then again, according to Intel (so take it with a grain of salt), their 65nm ramp produced the best yields they had ever seen with a new process, so who knows.

 

[MrsD]

That gap can't close as long as AMD gets progressively farther behind in process technology, and Intel doesn't make a major goof with their architecture like they did with Netbust.

Farther behind??
Let me remind you fanboy that Intel has yet to release anything that can touch an FX60. Just when you were about to blow your load huh?

 

[Action_Man]

As if you're not LMM.

 

[ethernalite]

That gap can't close as long as AMD gets progressively farther behind in process technology, and Intel doesn't make a major goof with their architecture like they did with Netbust.

Farther behind??
Let me remind you fanboy that Intel has yet to release anything that can touch an FX60. Just when you were about to blow your load huh?


I was referring to the theoretical price-performance gap. Even now, low end Intel chips are very competitively priced. While we spend a lot of time looking at the high end chips here, one can't forget that the vast majority of the dekstop market is in sub-$200 chips and Intel was never that far behind AMD in terms of price/performance in the lower price ranges. And since I was referring to the theoretical price/perf ratio, Intel could have easily reduced the price farther without running into the red ink. AMD couldn't.

That was my point.

Maybe you should learn to read before you start randomly flaming people? And I'm hardly an Intel fanboy, considering the fact that I am responding to you on a box that has an X2 processor on it. Good try, though.

 

[Action_Man]

Any help would do. What we could do, I could start a new thread and work with that and people could help by posting or PM links and info.