Sign in with
Sign up | Sign in
Your question

AMDs server roadmap

Last response: in CPUs
Share

Will K8L Quad be reserved for Opteron?

Total: 14 votes

  • Of course, Opteron always get the best first.
  • 22 %
  • No AMD is being killed on the desktop
  • 36 %
  • Maybe it depends on how good 65nm is
  • 36 %
  • Maybe if they need to
  • 8 %
July 18, 2006 11:58:47 PM

Phil Hester has a new roadmap "ad" where he talks about the future for the Opteron. There is little technical data but the roadmap shows that K8L is planned to have 60% greater perf/watt than current Opterons. The 2008 architecture is planned to have 150% greater perf/watt.

He explains that just like dual core Opteron will be the first K8L quad core around Q2 according to the dates shown. It maybe applied to all families (1xxx,2xxx,4xxx) to cover high end servers to high end wkstas.

Linkage!

More about : amds server roadmap

July 19, 2006 12:10:17 AM

Quote:
Phil Hester has a new roadmap "ad" where he talks about the future for the Opteron. There is little technical data but the roadmap shows that K8L is planned to have 60% greater perf/watt than current Opterons. The 2008 architecture is planned to have 150% greater perf/watt.

He explains that just like dual core Opteron will be the first K8L quad core around Q2 according to the dates shown. It maybe applied to all families (1xxx,2xxx,4xxx) to cover high end servers to high end wkstas.

Linkage!


How quaint.

Poser.
July 19, 2006 12:21:54 AM

Quote:
Phil Hester has a new roadmap "ad" where he talks about the future for the Opteron. There is little technical data but the roadmap shows that K8L is planned to have 60% greater perf/watt than current Opterons. The 2008 architecture is planned to have 150% greater perf/watt.

He explains that just like dual core Opteron will be the first K8L quad core around Q2 according to the dates shown. It maybe applied to all families (1xxx,2xxx,4xxx) to cover high end servers to high end wkstas.

Linkage!

From 5min 47sec BS, I found only quadcore as a roadmap info. What a waste of time.....
And what a typical BS from BBS.....
Related resources
Can't find your answer ? Ask !
July 19, 2006 1:04:50 AM

Isn't it cool how those smart links direct you to the presentation.

Pretty cool, huhn?

Anyway, can we please now stop the debate as to whether K8L is a server chip?

AMDs philosopy about desktops is


If you build the server they will come.


It's a good strategy considering that Intel used it for years to keep AMD out of the enterprise. With Kentsfield and clovertown and X4 and K8L, 2007 will be a FAST year. AMD may even restrict X4 to FX or the highest desktop rated proc.
July 19, 2006 1:25:30 AM

Did you take a look at the supposed 65nm wafer in the video? From what I can see that was still a 65nm single core die. I have yet to see a 65nm dual core die which is strange since the major benefit of 65nm production is to offset the capacity that the dual core transition stole.

The dies in the wafer do not look like the 65nm (K8L) single core prototype (no double FPUs) we've seen to date.

http://www.sandpile.org/impl/pics/amd/k8/die_65nm_1024k...

For one thing, the one in the video only had 3 decoders not four. The out of order buffers also appear to be less pronounced than in the blue 65nm prototype. In fact, the dies in the video wafer look virtually identical (I can't confirm for sure of course) to previous 90nm single cores.

http://www.sandpile.org/impl/pics/amd/k8/die_90nm_1024k...

This would seem to support the idea of a die shrink. Which implies that we have yet to see an actual K8L die.
July 19, 2006 1:46:52 AM

Quote:
Did you take a look at the supposed 65nm wafer in the video? From what I can see that was still a 65nm single core die. I have yet to see a 65nm dual core die which is strange since the major benefit of 65nm production is to offset the capacity that the dual core transition stole.

The dies in the wafer do not look like the 65nm (K8L) single core prototype (no double FPUs) we've seen to date.

http://www.sandpile.org/impl/pics/amd/k8/die_65nm_1024k...

For one thing, the one in the video only had 3 decoders not four. The out of order buffers also appear to be less pronounced than in the blue 65nm prototype. In fact, the dies in the video wafer look virtually identical (I can't confirm for sure of course) to previous 90nm single cores.

http://www.sandpile.org/impl/pics/amd/k8/die_90nm_1024k...

This would seem to support the idea of a die shrink. Which implies that we have yet to see an actual K8L die.



I thought about stopping it there but said who cares. If they don't produce it's not my dime.

I woul have told anyone that K8L would come after the die shrink. In retail terms that is.

I don't care if they don't even have the masks, why do you? I was about to post the E4200 story but it said that X2 3600+ beats it. (HKEPC)

I just wanted to clear up whether K8L would be on an early 2007 DESKTOP roadmap.

Well, I also felt it appropriate to report ANY news about x86-64 processors, but that is another story. But not another forum.
July 19, 2006 1:47:19 AM

9nm already posted it.
July 19, 2006 1:48:34 AM

Seriously do you have any sort of mental problems?
July 19, 2006 2:11:12 AM

Quote:
I don't care if they don't even have the masks, why do you? I was about to post the E4200 story but it said that X2 3600+ beats it. (HKEPC)

What HKEPC said was that the X2 3600+ and the E4200 (about the same price) were about equal, the 945D is faster (more expansive), and the 915D and 925D (cheaper) slower. Which means that all the above processors are price appropriately for the performance they give. It can still vary of course since the X2 3600+ and E4200 isn't released yet. Still, the X2 3600+ is by no means a Netburst sale killer given that the Netbursts are well priced according to their performance. Since you are concerned (legitimately) about Netburst sale viability this should be a good sign.

In any case, now that the E4200 is just equal, Intel has responded and canceled it in favour of the 1.8GHz E4300. Now doubt they will offer it at the same price. The only disadvantage is that it will be late to the party until Q1 2007. That does mean that Netburst sales will continue to be strong since their isn't any overlap with the 915D/925D/945D for a while. Intel has also advised in their PCNs for people to test those chips at 3.2GHz for a possible 935D introduction. The 935D would probably be able to directly compete with the X2 3600+ and replace the cancelled E4200 if necessary given that the 925D is slower and the 945D is faster.

Anyways, I'm going too off topic from your original thread. I wish AMD would be a bit more forthcoming about their K8L details though. We'll probably have to wait for early Q4 though, when the initial Core 2 furor cools down, and before it starts up again with Intel's quad cores.
July 19, 2006 2:12:50 AM

Quote:
Seriously do you have any sort of mental problems?



Actually I do have a few but admittin gyour problems - when discovered - is the first step to fixing them.

Could you at least point to TYROU's successes with Kentsfield on extreme systems?

Oh you're the one-liner guy.

My bad.
July 19, 2006 2:13:50 AM

WTF are you rambling on about?
July 19, 2006 2:26:46 AM

Quote:


He explains that just like dual core Opteron will be the first K8L quad core around Q2 according to the dates shown. It maybe applied to all families (1xxx,2xxx,4xxx) to cover high end servers to high end wkstas.


This is COMPLETELY INCORRECT and i recommend this thread be locked for posting blatent lies. No where in the video is K8L mentioned. Only quad core is mentioned and nothing says that the quad core will be K8L. AMD has already explicitely stated that they will release K8 baed quad core CPUs. Yes, you read that correctly, K8, and NOT K8L quad cores. Until they explicitly mention K8L in an announcement or roadmap, then dont claim for it to be K8L.
July 19, 2006 3:03:49 AM

Quote:


He explains that just like dual core Opteron will be the first K8L quad core around Q2 according to the dates shown. It maybe applied to all families (1xxx,2xxx,4xxx) to cover high end servers to high end wkstas.


This is COMPLETELY INCORRECT and i recommend this thread be locked for posting blatent lies. No where in the video is K8L mentioned. Only quad core is mentioned and nothing says that the quad core will be K8L. AMD has already explicitely stated that they will release K8 baed quad core CPUs. Yes, you read that correctly, K8, and NOT K8L quad cores. Until they explicitly mention K8L in an announcement or roadmap, then dont claim for it to be K8L.

I support this suggestion, I will PM Jake now.

Good, i wasn't the only one with this opinion. I love threads created by BS..... i mean.... BM.
July 19, 2006 3:11:27 AM

Quote:
It is also not a news event, it is an advertisement -- this is in violation of the terms of agreement:

Advertisements, chain letters, pyramid schemes and other commercial solicitations are inappropriate on the Forums.
So it is. Jake will now receive a third report message.
July 19, 2006 3:13:50 AM

Quote:


He explains that just like dual core Opteron will be the first K8L quad core around Q2 according to the dates shown. It maybe applied to all families (1xxx,2xxx,4xxx) to cover high end servers to high end wkstas.


This is COMPLETELY INCORRECT and i recommend this thread be locked for posting blatent lies. No where in the video is K8L mentioned. Only quad core is mentioned and nothing says that the quad core will be K8L. AMD has already explicitely stated that they will release K8 baed quad core CPUs. Yes, you read that correctly, K8, and NOT K8L quad cores. Until they explicitly mention K8L in an announcement or roadmap, then dont claim for it to be K8L.


I will watch it again and quote the exact times where they TALK ABOUT A NEW CORE FOR QUAD CORE SERVERS THAT WILL DELIVER 60% more PERF/WAT THAN CURRENT OPTERONS.

Are you saying then that simple quad core will deliver that kind of increased perf in DATACENTERS?

The NEXT GEN SPOKEN ABOUT BY HECTOR RUIZ WILL SUPPOSEDLY DELIVER 150% MORE PERF/WATT IN Q3 08.
July 19, 2006 3:29:21 AM

It's quite simple. AMD claims they can produce a quad core with the same TDP as current dual cores. Going 65nm, I believe them since it's certainly possible. Therefore, a quad K8 with the same TDP as a dual core K8 can easily provide the 60% increase in performance/watt they are claiming for 2007.

For 2008 K8L will come in and will probably require a higher TDP with all it's features. Still since it's microarchitecture is better, a quad core K8L should provide the 150% increase in performance/watt over a dual core K8.
July 19, 2006 3:29:46 AM

Quote:


He explains that just like dual core Opteron will be the first K8L quad core around Q2 according to the dates shown. It maybe applied to all families (1xxx,2xxx,4xxx) to cover high end servers to high end wkstas.


This is COMPLETELY INCORRECT and i recommend this thread be locked for posting blatent lies. No where in the video is K8L mentioned. Only quad core is mentioned and nothing says that the quad core will be K8L. AMD has already explicitely stated that they will release K8 baed quad core CPUs. Yes, you read that correctly, K8, and NOT K8L quad cores. Until they explicitly mention K8L in an announcement or roadmap, then dont claim for it to be K8L.


I will watch it again and quote the exact times where they TALK ABOUT A NEW CORE FOR QUAD CORE SERVERS THAT WILL DELIVER 60% more PERF/WAT THAN CURRENT OPTERONS.

Are you saying then that simple quad core will deliver that kind of increased perf PER WATTin DATACENTERS?

The NEXT GEN SPOKEN ABOUT BY HECTOR RUIZ WILL SUPPOSEDLY DELIVER 150% MORE PERF/WATT IN Q3 08.

Performance per watt..... i dont think it will be 60% more, but yes, between 65nm and and quad core, it will deliver a large performance per watt. Do i think it will be NEAR 150%? No way, but thats advertising speak. When i see benchmarks i'll beleive it.

But no where did they state anything about K8L. I dont think i can be any more clear on this.

I love how you quote Performance/Watt, then you state that it will be that much PERFORMANCE! This is a MAJOR difference!
July 19, 2006 3:53:49 AM

Please Jack, refrain from feeding the troll. You are much better than this. :wink:
July 19, 2006 4:11:06 AM

Quote:
It's quite simple. AMD claims they can produce a quad core with the same TDP as current dual cores. Going 65nm, I believe them since it's certainly possible. Therefore, a quad K8 with the same TDP as a dual core K8 can easily provide the 60% increase in performance/watt they are claiming for 2007.

For 2008 K8L will come in and will probably require a higher TDP with all it's features. Still since it's microarchitecture is better, a quad core K8L should provide the 150% increase in performance/watt over a dual core K8.



Actually AMD has set the quad core TDP at 95W.
July 19, 2006 4:16:22 AM

Quote:


He explains that just like dual core Opteron will be the first K8L quad core around Q2 according to the dates shown. It maybe applied to all families (1xxx,2xxx,4xxx) to cover high end servers to high end wkstas.


This is COMPLETELY INCORRECT and i recommend this thread be locked for posting blatent lies. No where in the video is K8L mentioned. Only quad core is mentioned and nothing says that the quad core will be K8L. AMD has already explicitely stated that they will release K8 baed quad core CPUs. Yes, you read that correctly, K8, and NOT K8L quad cores. Until they explicitly mention K8L in an announcement or roadmap, then dont claim for it to be K8L.


I will watch it again and quote the exact times where they TALK ABOUT A NEW CORE FOR QUAD CORE SERVERS THAT WILL DELIVER 60% more PERF/WAT THAN CURRENT OPTERONS.

Are you saying then that simple quad core will deliver that kind of increased perf PER WATTin DATACENTERS?

The NEXT GEN SPOKEN ABOUT BY HECTOR RUIZ WILL SUPPOSEDLY DELIVER 150% MORE PERF/WATT IN Q3 08.

Performance per watt..... i dont think it will be 60% more, but yes, between 65nm and and quad core, it will deliver a large performance per watt. Do i think it will be NEAR 150%? No way, but thats advertising speak. When i see benchmarks i'll beleive it.

But no where did they state anything about K8L. I dont think i can be any more clear on this.

I love how you quote Performance/Watt, then you state that it will be that much PERFORMANCE! This is a MAJOR difference!


I'm sorry I can't offer a disclaimer you would accept. I can only try to dissemble the info presented.
July 19, 2006 4:18:23 AM

Yeah, I thought I might as well clarify.

For the 60% 2007 measurement, by keeping the TDP the same all the increase will come from performance. 50% would probably be the dual core to quad core performance increase. The other 10% would be going from Registered DDR2 667 with 5-5-5 timings to Registered DDR2 800 with 5-5-5 timings or lower. (We've discussed Socket F's DDR2 latency issues before). Revisions will also help.

For the 2008 numbers, 150% would seem very high. However, this will of course include the dual core to quad core transition. The K8 to K8L transition. The Registered DDR2 667 to FB-DIMM DDR2 667 or DDR2 1067 transition. The HT2.0 to HT3.0 transition. And also since AMD is now talking about 45nm in 2008, a 90nm to 45nm transition. For the 150% increase in performance/watt isn't all from performance, but it's also going to be from decreased watts too through 45nm if they are on schedule. Higher clock speeds are also likely.

This means that the 60% numbers are perfectly reachable with K8 quad cores, and 150% with K8L. It doesn't necessarily mean K8L in 2007 for the 60%.
July 19, 2006 4:22:07 AM

Quote:


I don't care if they don't even have the masks, why do you? I was about to post the E4200 story but it said that X2 3600+ beats it. (HKEPC)

I just wanted to clear up whether K8L would be on an early 2007 DESKTOP roadmap.


Because without the mask they cannot make the CPU, if they cannot make the CPU, then they cannot debug it, if they cannot debug it then they cannot revise it to get the next set of masks, if they cannot get the next set of masks, then they cannot release the product.

This is why it matters.

Lt. Commander Data is a pretty sharp fella', he is on the ball and knows what he is talking about. You should listen to him, you might learn something.

The tape out, debug, test, and tape out again can take up to a year or more for a new product. If they don't have the masks in yet, then it is looking like a long time before dual core K8Ls.

Also, that article you reference shows no benchmarks and discusses nothing about performance in terms of who beats who. This is FUD.

Jack

As a simile so to speak did you know the US injected A whole lot of children with radioactive isotopes during the height of the Cold War arms race(50s)?

I don't know if AMD has to inform YOU OR ME the day of every tape out. Do you know the day the R600 taped out or the day the G80 taped out or the day the Core 2 taped out?

Please stop flooding the board with negative vibes. It's like opening the door in a clean room.
July 19, 2006 4:26:30 AM

Quote:
Yeah, I thought I might as well clarify.

For the 60% 2007 measurement, by keeping the TDP the same all the increase will come from performance. 50% would probably be the dual core to quad core performance increase. The other 10% would be going from Registered DDR2 667 with 5-5-5 timings to Registered DDR2 800 with 5-5-5 timings or lower. (We've discussed Socket F's DDR2 latency issues before). Revisions will also help.

For the 2008 numbers, 150% would seem very high. However, this will of course include the dual core to quad core transition. The K8 to K8L transition. The Registered DDR2 667 to FB-DIMM DDR2 667 or DDR2 1067 transition. The HT2.0 to HT3.0 transition. And also since AMD is now talking about 45nm in 2008, a 90nm to 45nm transition. For the 150% increase in performance/watt isn't all from performance, but it's also going to be from decreased watts too through 45nm if they are on schedule. Higher clock speeds are also likely.

This means that the 60% numbers are perfectly reachable with K8 quad cores, and 150% with K8L. It doesn't necessarily mean K8L in 2007 for the 60%.



As I said I just watched the "whatever" after seeing it on AMDZone. By all of the indicators leading up to 2007 (including several presentations/interviews), the logical course is as stated.
July 19, 2006 4:29:04 AM

Its funny reading how much spin you put on things.
July 19, 2006 4:31:24 AM

I don't know the exact dates of tape out, but for Merom (et al), if I remember the old reports correctly, it was scheduled to tape out in July 2005, but in the end it taped out a month ahead of schedule in June 2005. I think I posted that here back then or at least mentioned it. So that's about a year from tape out to launch.

I believe the R600 was just taped out last month, but things are difference for GPUs since I think they are mostly computer designed rather than having transistors hand laid. Probably why the power consumption and heat is going through the roof. (Certainly with the R580). Anyways.
July 19, 2006 4:58:29 AM

Quote:
I don't know the exact dates of tape out, but for Merom (et al), if I remember the old reports correctly, it was scheduled to tape out in July 2005, but in the end it taped out a month ahead of schedule in June 2005. I think I posted that here back then or at least mentioned it. So that's about a year from tape out to launch.

I believe the R600 was just taped out last month, but things are difference for GPUs since I think they are mostly computer designed rather than having transistors hand laid. Probably why the power consumption and heat is going through the roof. (Certainly with the R580). Anyways.



I wasn't being argumentative. I was trying to make the point that CPU/GPU tape out dates may only make a small headline at the Inq's poor subsidiary.

I am excited about what I've heard from AMD though.
July 19, 2006 5:24:17 AM

Quote:
I don't know the exact dates of tape out, but for Merom (et al), if I remember the old reports correctly, it was scheduled to tape out in July 2005, but in the end it taped out a month ahead of schedule in June 2005. I think I posted that here back then or at least mentioned it. So that's about a year from tape out to launch.

I believe the R600 was just taped out last month, but things are difference for GPUs since I think they are mostly computer designed rather than having transistors hand laid. Probably why the power consumption and heat is going through the roof. (Certainly with the R580). Anyways.



I wasn't being argumentative. I was trying to make the point that CPU/GPU tape out dates may only make a small headline at the Inq's poor subsidiary.

I am excited about what I've heard from AMD though.

Man, you can not back up ANYTHING you say and you do not provide ANY information supporting your opinion. All you say is K8L is going to be released in early 2007. You have shown NO PROOF AT ALL to back this up. Everyone tells you that it will NOT make it that early and they offer EXPLANATIONS that support their opinion. You have not contributed ANYTHING constructive to this topic.

If AMD does not even have a first run K8L wafer out yet, then there is NO WAY that they will release K8L within a year. As ltcommander said, it took Intel about 1 year to get from tape out to release, yet you seem to think that AMD will do this in less than a year when all the data shows they haven't even taped out yet! You are dreaming! Intel did it in about 1 year, and they released it on B2, which means the silicon was EXTREMELY healthy from the beginning!
!