AMD K8L ES coming in December

Printable View

Show 100 post(s) from this thread on one page

12-28-2006, 02:49 PM
day187

Quote:

Originally Posted by Fred_Pohl

Although some people here believe that they have been around for several months under super-duper-top-secret NDA, IMO we won't see them until sometime in Q1-07 with retail availability sometime in Q3-07.

thanks
12-28-2006, 03:18 PM
Shintai

Quote:

Originally Posted by informal

The Akiba article said that (my understanding at least): AMD representatives told the press that the level 3 cache of dual core Stars family(Kuma) would bring roughly 20% increase in performance(probably average figure,dependant on the apps) over previous K8 generation X2(dual core,ie Windsor or Brisbane)!
Note that they emphasised cache in this information.
They were clearly talking about the addition of the L3 cache and the effects of it on the New Core,compared to the older K8 architecture.We can assume that the New Core benefits more from the shared level 3 victim cache due to it's largerly improved execution units and imroved design,overall.

They didn't say that Stars will be 20%(overall ) faster than K8!At least we can't conclude they said that only because of this one article(written in Chinese btw).

Tbh thats BS, 20% performance add for 2MB L3 cache? How much do we get for 512KB vs 1024 today? *HINT HINT*
12-28-2006, 03:36 PM
Fred_Pohl

I think what they probably meant was 20% faster cache performance. IMHO that would translate to ~5% better performance overall from the improved, larger cache.

Although I suppose there might be one app out there where the new cache could potentially improve overall CPU performance by ~20%.
12-28-2006, 03:44 PM
Shintai

Quote:

Originally Posted by Fred_Pohl

I think what they probably meant was 20% faster cache performance. IMHO that would translate to ~5% better performance overall from the improved, larger cache.

Although I suppose there might be one app out there where the new cache could potentially improve overall CPU performance by ~20%.

The most cache intensive application got just 10% on Core 2 2MB vs 4MB, and thats L2 and not slower L3. The average was 3.5%. But thats on Core 2, without IMC. So it might be better to compare 512KB with 1MB cache K8s and make an estimate from that.
12-28-2006, 04:16 PM
[XC] Teroedni

In 2007 Q2 Amd will Release K8l as Server chips

And in 2007 Q3 The K8l will come to the Desktop as well.
12-28-2006, 04:19 PM
Shintai

Quote:

Originally Posted by Teroedni

In 2007 Q2 Amd will Release K8l as Server chips

And in 2007 Q3 The K8l will come to the Desktop as well.

Link? I only seen Q3 for Server and Q1 2008 for desktop. Perhaps you think on regular K8 quadcores and not K8L? Or perhaps just AM2+ versions of K8.
12-28-2006, 04:20 PM
Fred_Pohl

Quote:

Originally Posted by Shintai

The most cache intensive application got just 10% on Core 2 2MB vs 4MB, and thats L2 and not slower L3. The average was 3.5%. But thats on Core 2, without IMC. So it might be better to compare 512KB with 1MB cache K8s and make an estimate from that.

If all AMD did was add a 2MB L3 cache it wouldn't add 20% performance but IIRC they are also improving the L2 cache (128bit vs 64bit) and prefetch on the stars. I agree that 20% overall improvement even in the most cache-sensitive app out there is a bit of a stretch but I think it might be possible, if unlikely, when you add all the new cache improvements up.

Edit: I meant 256bit vs 128bit.
12-28-2006, 04:24 PM
Fred_Pohl

Quote:

Originally Posted by Teroedni

In 2007 Q2 Amd will Release K8l as Server chips

And in 2007 Q3 The K8l will come to the Desktop as well.

Is that your guess or do you have a reliable source? IIRC AMD has only comitted to a stars launch (paper or hard) sometime in H2-07 and I've seen AMD roadmaps that show the first desktop stars appearing in H1-08.
12-28-2006, 04:26 PM
Shintai

Quote:

Originally Posted by Fred_Pohl

If all AMD did was add a 2MB L3 cache it wouldn't add 20% performance but IIRC they are also improving the L2 cache (128bit vs 64bit) and prefetch on the stars. I agree that 20% overall improvement even in the most cache-sensitive app out there is a bit of a stretch but I think it might be possible, if unlikely, when you add all the new cache improvements up.

But the cache is already 128bit wide...you have to go back to K7 before it was 64bit.
12-28-2006, 04:35 PM
[XC] Teroedni

1 Attachment(s)

The last one is Amd slide;)

http://www.xbitlabs.com/news/cpu/dis...721230935.html

http://www.hkepc.com/bbs/itnews.php?...me=0&endtime=0
12-28-2006, 04:50 PM
Shintai

Quote:

Originally Posted by Teroedni

http://www.xbitlabs.com/news/cpu/dis...721230935.html

http://www.hkepc.com/bbs/itnews.php?...me=0&endtime=0

Uhm..? First link aint exactly helping you there.

Quote:

Earlier the company planned to showcase quadl-core microprocessors in the middle of the year, particularly, in August, when it is expected to introduce its new server platform and socket F infrastructure.

“When AMD rolls out dual-core processors with built-in virtualization hooks midyear, the company also aims to demo quad-core processors running on its current server platform,” Marty Seyer, a senior vice president of AMD, said in an interview.

The chipmaker now claims that the chips will be showcased by the end of the year without giving any particular timeframes, which may signal that the company currently has no quad-core chip samples mature enough for public showcase.

And the second one is before they updated the info. Aka 2.9Ghz down to 2.5Gnz and so forth.
http://www.hkepc.com/bbs/itnews.php?tid=715237
http://www.hkepc.com/bbs/itnews.php?tid=709944

Seems you are right about Q2 tho. but lets see, its been delayed before.

The slide is useless, thats been all over before and we got alot more detailed info since.
12-28-2006, 05:02 PM
[XC] Teroedni

Thanks for the last link

2007 Q2 Opteron k8l cores max=2,5ghz

2007 Q3 Desktp k8l max =2,9ghz

And the slide is from Amd themselfes. Which means it should be correct:fact:
12-28-2006, 06:16 PM
Serge84

K8L will use 256-bit wide cache. We seen the specs posted here a million and one times. Nothing amazing.

Edit: Its a new arc so whatever speculation on how it will OC is unknown period until real parts are able to be judged guilty or inicent as charged. =P
12-28-2006, 06:28 PM
doompc

How much faster would a K8 X2 be with shared L2 cache, or how much slower would a Core2 be with dedicated L2 cache??

K8L will feature 256bit L1 cache, L2 seems to be untouched (or with 20 cicle latency like brisbane...).
12-28-2006, 06:55 PM
Fred_Pohl

Quote:

Originally Posted by Shintai

But the cache is already 128bit wide...you have to go back to K7 before it was 64bit.

Yeah, my bad. I meant 256bit L2 instead of the current 2x64bit (128bit).
12-28-2006, 08:40 PM
informal

Quote:

Originally Posted by Shintai

Link? I only seen Q3 for Server and Q1 2008 for desktop. Perhaps you think on regular K8 quadcores and not K8L? Or perhaps just AM2+ versions of K8.

There is NO native K8 based quad core anywere and it never will be.The only native QC parts are the Stars family CPUs.Period.
And as for the cache,remember that New Core(notice i dont use non existent name K8L for Stars) will have wider caches,with L3 as a victim cache,and very improved architecture compared to old K8.So we shouldn't be surprised if New Core shows massive performance imrovement.It was built for that purpouse,and built for a long time(development probably started by the time of K8 introduction).

And the last thing,there are NO AM2+ versions of K8.There is only one version of K8 in 2007,AM2 one(ie Orleans,Windsor,Brisbane,Manila,server parts ...).AMD is not redesigning Brisbanes to support HT3.0...
What are you babbling about?
12-29-2006, 01:32 AM
Shintai

Quote:

Originally Posted by Teroedni

Thanks for the last link

2007 Q2 Opteron k8l cores max=2,5ghz

2007 Q3 Desktp k8l max =2,9ghz

And the slide is from Amd themselfes. Which means it should be correct:fact:

The 2.9Ghz would be a dualcore then? Else it doesnt sound likely.
12-29-2006, 01:33 AM
Shintai

Quote:

Originally Posted by doompc

How much faster would a K8 X2 be with shared L2 cache, or how much slower would a Core2 be with dedicated L2 cache??

K8L will feature 256bit L1 cache, L2 seems to be untouched (or with 20 cicle latency like brisbane...).

I think speed difference on Core 2 would be minimal in multithreaded applications with a dedicated cache. What people forget is that Core 2 dont use it shared L2 cache for snoops. But rather direct L1 links.

K8 will never get a shared cache, and I doubt the new L3 cache is much, if any faster than the crossbar. However is larger systems (more sockets) it can help abit on the coherency part.
12-29-2006, 01:41 AM
Shintai

Quote:

Originally Posted by informal

There is NO native K8 based quad core anywere and it never will be.The only native QC parts are the Stars family CPUs.Period.
And as for the cache,remember that New Core(notice i dont use non existent name K8L for Stars) will have wider caches,with L3 as a victim cache,and very improved architecture compared to old K8.So we shouldn't be surprised if New Core shows massive performance imrovement.It was built for that purpouse,and built for a long time(development probably started by the time of K8 introduction).

And the last thing,there are NO AM2+ versions of K8.There is only one version of K8 in 2007,AM2 one(ie Orleans,Windsor,Brisbane,Manila,server parts ...).AMD is not redesigning Brisbanes to support HT3.0...
What are you babbling about?

Uhm...I really really doubt K8L aka. Stars was started just after K8. In that case its an insanely slow development team for an evolutionized design. The only wider cache seems to be the L1. And even if the L2 is, then we gotta consider the new 20 cycle latency standard they introduced. Also very improved? You gonna be hold up on that. I would call Core 2 very improved over Core 1. K8L only holds a small handful of enchancements when compared to K8.

So unless you got something to add to the list of improvements I don´t see them anywhere.

Lets try put up a list here, for Core 2 to be around 20% faster than Core 1 it needed:
1 extra simple decoder
1 extra issue port
Bigger buffers here and there
128bit vs 64bit paths
Memory disambiguation
1+1 faster decode rate
1 extra ALU
2 extra SSE units
Single cycle SSE
Macro fusion (32bit only tho)
4 FP units vs 1 (Tho 4 simpler)
Larger L2 cache
Direct L1 links

Now, lets look on K8L vs K8:
Memory disambiguation
128bit vs 64bit paths
Single cycle SSE
Bigger buffers here and there
L3 cache

Anything you want to add? :fact:
12-29-2006, 01:54 AM
[XC] Teroedni

2 Attachment(s)

Quote:

Originally Posted by Shintai

The 2.9Ghz would be a dualcore then? Else it doesnt sound likely.

Its seems your a little Confused;)

The Opterons based on K8l will be released first. They will come in Q2 with a maximum of 2,5 ghz.
http://www.xtremesystems.org/forums/...1&d=1167386642

Then in Q3 Amd will come with Higher clocked K8L for the Desktop
http://www.xtremesystems.org/forums/...1&d=1167386642
12-29-2006, 02:19 AM
Shintai

Quote:

Originally Posted by Teroedni

Its seems your a little Confused;)

The Opterons based on K8l will be released first. They will come in Q2 with a maximum of 2,5 ghz.
http://www.xtremesystems.org/forums/...1&d=1167386642

Then in Q3 Amd will come with Higher clocked K8L for the Desktop
http://www.xtremesystems.org/forums/...1&d=1167386642

No not really. Now its time to apply logic. If your "4x4" CPU can reach 2.9Ghz with a 125W TDP. Why would your precise same Opteron peak at 2.5Ghz at 120W. Your desktop one is from october, and at that time people said 2.9Ghz for the server CPUs aswell.

Also there seems to have been name changes aswell
http://www.hkepc.com/bbs/itnews.php?...me=0&endtime=0

The 2.9Ghz also sounds more like the new Brisbane clock scheme.
12-29-2006, 02:38 AM
NapalmV5

so, less than 3 days left.. anything?
12-29-2006, 02:52 AM
alayashu

Quote:

Originally Posted by Shintai

Lets try put up a list here, for Core 2 to be around 20% faster than Core 1 it needed:
1 extra simple decoder
1 extra issue port
Bigger buffers here and there
128bit vs 64bit paths
Memory disambiguation
1+1 faster decode rate
1 extra ALU
2 extra SSE units
Single cycle SSE
Macro fusion (32bit only tho)
4 FP units vs 1 (Tho 4 simpler)
Larger L2 cache
Direct L1 links

Now, lets look on K8L vs K8:
Memory disambiguation
128bit vs 64bit paths
Single cycle SSE
Bigger buffers here and there
L3 cache

Anything you want to add? :fact:

sure, it will be slower than C2D :rolleyes:
any comparisions like the one you try to do are only useful to
entertain public forums like this one :P
12-29-2006, 03:53 AM
[XC] Teroedni

Quote:

Originally Posted by Shintai

No not really. Now its time to apply logic. If your "4x4" CPU can reach 2.9Ghz with a 125W TDP. Why would your precise same Opteron peak at 2.5Ghz at 120W. Your desktop one is from october, and at that time people said 2.9Ghz for the server CPUs aswell.

What logic yours?:ROTF:

Unless you can show link to back your claim your only talking BS

Now please back up your facts, or dont comment. This thread is not supposed to be about Amd vs Intel which you obviously thought in the locked 65nm thread:stick:
12-29-2006, 04:14 AM
Shintai

Quote:

Originally Posted by Teroedni

What logic yours?:ROTF:

Unless you can show link to back your claim your only talking BS

Now please back up your facts, or dont comment. This thread is not supposed to be about Amd vs Intel which you obviously thought in the locked 65nm thread:stick:

Just tell me why, a desktop product would be clocked alot higher than a serverproduct for about the same TDP since its the same dies and design. And check the newer desktop roadmap. Kuma is new, the other desktop (besides the altair is gone). So you got something that you forgot to look at. Since the product names dont even match, its sure to asume that your older list is outdated :stick:

The 65nm was locked, I presume, due to people making up crap instead of sticking to reality and starting to flame one another personally.

Show 100 post(s) from this thread on one page