K8L vs Conroe

C

(cf)Eclipse

Freelance Overclocker
Joined
Feb 18, 2003
Messages
30,027
all right guys. first off, keep it civil. i have the ability to lock this thread at any time, and i won't be afraid to use it. i want technical and/or factual posting only. opinions are ok, but posting negatively about someone else's is a no-no, and i will ask for that post to be removed/edited.

that said, let's talk about the performance advantages K8L will give over K8, and whether it will be enough to take on Conroe or not.

atm, i think that K8L will definitly be a good boost over K8.. with it's SSE performance boost, OOO load execution, and a few other things that should also boost performance, i'm guessing we'll see a 10-25% boost in performance over an equally clocked K8. though obviously, performance will depend on the data loading, as always.
one thing i am sorta concerned about is that it doesn't seem like the ALU units were touched. compared to conroe, K8 isn't exactly stellar when it comes to pure integer performance, which could pose a problem in some cases.

feel free to post things that we know for sure about each core, i'll be glad to keep a list in this post comparing what each core will have to compete
 
I support and I am a fan of AMD, call it blind alligence whatever you will.. its just like I am a fan of the San Francisco 49ers for life.

K8L FTW!

I am no engineer or anything, dont understand or know the real inner workings of a proc.. but AMD has never let me down since I have switch, whereas Intel has. So my money goes to AMD no matter, unless they seriously let me down somehow, and since I make my money. I have the right to decide how to spend it and it will be on a K8L and not Conroe... plus it is not always about pure raw power... cause pure raw power is useless if you dont have the OS tweaked to run it to its highest potential.. aka the average consumer having 60 billion process running at startup wondering why there latest and greastest puter is still slow.
 
yes, the operating system is indeed important, and the rest of the rig. how will each respond to faster/slower ram? my general thought is that to get through a certain amount of data, the cpu needs to be fed the information to it. K8L and Conroe chips at equal performance levels (whatever mhz that happens to correspond to) should have an equal want for bandwidth. this is where things get a little interesting.

conroe is obviously gonna have a very strong cache subsystem with gobs of L2
K8L is going to have a lot less cache, but the low latency memory controller will make up for it...

in reality, i think one of the biggest limitations we have in our computers today is the hard drive access time, and the relatively weak amount of bandwidth we can get out of it. no amount of rediculously cpu power will help loading times if the majority of the time is spent waiting for the data to be read from the disk.

ontop of that.. WTF happened with x86-64. there's an opportunity here for an easy performance gain, don't even have to go take the effort to really re-write code like you would if the program was being made multi-threaded. almost all intel and amd processors sold these days are x86-64 capable, why are the benefits being thrown away?
speaking of which, i hear rumors that intel's implementation of x86-64 is kinda odd, and very lacking. i haven't had hands on experience.. yet (i plan on getting a 631 soon), but i'd like it if someone who has used 64bit apps on both amd and intel would comment on this ;)

obviously, i'm pretty heavily biased towards amd as well, but i really do try to see both sides of things ;)
 
I'll be watching this thread with great interest mainly because I don't know much about the engineering side of either of these cpu technologies. I wouldn't begin to try to make a post here other than this one because its all "greek" to me, so to speak. I'm sure I'll get a real education, though.
I too enjoy AMD but it is good to see Intel finally waking up and putting some good tech out. We all definitely will benefit from the competition.
While the rest of you debate the merits of Conroe and K8L I'm going to be playing with my new skt 754 3200+ Venice. It should be here in my hot hands by Friday. Yeah, I'm behind the times, but I don't mind. :D
 
Your right About Intels 64 bit implementation and that will not be fixed till Wolfdale.

Now please explain from the available info on K9 how AMD is going to implement OoO I have not seen were they can accomplish this with the available info.
 
WTF happened with x86-64. there's an opportunity here for an easy performance gain, don't even have to go take the effort to really re-write code

Ya now you see the dilema that Intel is in with the Itanic. But worse yet.
 
$BangforThe$ said:
WTF happened with x86-64. there's an opportunity here for an easy performance gain, don't even have to go take the effort to really re-write code

Ya now you see the dilema that Intel is in with the Itanic. But worse yet.
Click to expand...
That's not entirely true. Performance gains from using 64-bit operations are not really as easy as you make them out to be. A recompile will not generally offer much of an advantage until you put some work into facilitating the implementation of 64-bit instructions. Anyways, I don't see many software developers getting excited about reworking their code for many applications because the performance gain is most likely not worth their time and money.

Anyways, this shouldn't be alarming for any of the parties involved. AMD built the K8 family to excel in a wide range of scenarios, so the tepid response to x86-64 development shouldn't be too harmful to anyone.
 
xonik said:
That's not entirely true. Performance gains from using 64-bit operations are not really as easy as you make them out to be. A recompile will not generally offer much of an advantage until you put some work into facilitating the implementation of 64-bit instructions. Anyways, I don't see many software developers getting excited about reworking their code for many applications because the performance gain is most likely not worth their time and money.

Anyways, this shouldn't be alarming for any of the parties involved. AMD built the K8 family to excel in a wide range of scenarios, so the tepid response to x86-64 development shouldn't be too harmful to anyone.
Click to expand...


Porting code to AMD64 is pretty easy. Most code will work with just a recompile. The problem comes with long-longs though, and a few other scenerios. But most of those can be quickly identified with a good debugger.

But other wize I agree. AMD64 isnt as big a deal as others make it out to be.
 
My personal opinion is that performance is becoming more and more competitive, and these two titans will clash to our benefit rather than one dominating the other. Intel was smart to ditch netburst, while AMD was smart to hit ddr2, I personally find both very exciting.
 
$BangforThe$ said:
Your right About Intels 64 bit implementation and that will not be fixed till Wolfdale.

Now please explain from the available info on K9 how AMD is going to implement OoO I have not seen were they can accomplish this with the available info.
Click to expand...
can you explain exactly what needs to be "fixed"?
and seriously. K8L. where in the title or in my post does it say K9? ;)
i don't know how OOO is going to be implemented. i'm not knowledged enough in the low level workings to predict what they will change to add it in

$BangforThe$ said:
Ya now you see the dilema that Intel is in with the Itanic. But worse yet.
Click to expand...
except that wasn't backwards compatible.. well, it was, but needed emulation, which was slow. x86-64 on the other hand, has NO penalty when running 32bit code in a 64bit environment..

jason, i'm expecting similar levels to what K8 can do now. what's the point of an IPC increase if you can't scale mhz high enough to increase performance? ;)
 
duby229 said:
Porting code to AMD64 is pretty easy. Most code will work with just a recompile. The problem comes with long-longs though, and a few other scenerios. But most of those can be quickly identified with a good debugger.

But other wize I agree. AMD64 isnt as big a deal as others make it out to be.
Click to expand...
Porting is one thing--getting a performance increase is another. Just curious, how well did you read my post?
 
Its not gonna be K8L vs Conroe, its gonna be K8L vs 45nm next-gen iteration of Conroe (Nehalem, or w/e they are calling it)
 
nobi125 said:
So subscription information for Genmay is supposed to clue me in as to how you are able to lock threads?
Click to expand...
yes. i can lock because i'm a subscriber ;)


brucedeluxe169 said:
Its not gonna be K8L vs Conroe, its gonna be K8L vs 45nm next-gen iteration of Conroe (Nehalem, or w/e they are calling it)
Click to expand...
so you're saying that intel's 45nm will be out by H1 of '07 too?
i'd like some evidence to this.
also, i know very little about nehalem. i was under the impression that it will just be a die shrink of conroe.
 
brucedeluxe169 said:
Its not gonna be K8L vs Conroe, its gonna be K8L vs 45nm next-gen iteration of Conroe (Nehalem, or w/e they are calling it)
Click to expand...
I would agree in the fact that this would crucially depend on when K8L is intorduced.

Nehalem is I beleive slated for sometime late in the 45nm process if going by Intels roadmaps, Mid 2008 maybe...

I think K8L maybe a tad earlier then that though.
 
(cf)Eclipse said:
yes. i can lock because i'm a subscriber ;)



so you're saying that intel's 45nm will be out by H1 of '07 too?
i'd like some evidence to this.
also, i know very little about nehalem. i was under the impression that it will just be a die shrink of conroe.
Click to expand...
Nehalem, is bit too far off to be talking about right now. But to give you some information, it is another NGMA, though how much it differs from Conroe remains to seen, it is not a simple optical shrink with more cache.
 
(cf)Eclipse said:
yes. i can lock because i'm a subscriber ;)



so you're saying that intel's 45nm will be out by H1 of '07 too?
i'd like some evidence to this.
also, i know very little about nehalem. i was under the impression that it will just be a die shrink of conroe.
Click to expand...

No he is saying don't hold your breath for K8L to be out in 1st qt of 07 . Look for it in 4th qt of 07 . I think Amd can have K8L out by june of 07 but don't exspect to buy one til like sept. Than its only a few months to Wolf-Dale( A simple shrink it is) If K8L is OoO and reverse hyper threading and such than I would say Intel Wolfdale will beat it into the market.
 
coldpower27 said:
Nehalem, is bit too far off to be talking about right now. But to give you some information, it is another NGMA, though how much it differs from Conroe remains to seen, it is not a simple optical shrink with more cache.
Click to expand...

Its really not that far off . Till we see a 65nm die from Fab 36 who knows Nehalem might beat K8L to market .

If its the big changes that AMD says it is . They have to have a proto type taped out first. (not a hand engineered die) Than it has a valadation period . It could be 08 befor we see the K8L product that people are now talking about.
 
xonik said:
Porting is one thing--getting a performance increase is another. Just curious, how well did you read my post?
Click to expand...

Your implecation is that porting isnt worthwhile, I pointed out that wasnt the case. Performance is not really the issue.
 
$BangforThe$ said:
Its really not that far off . Till we see a 65nm die from Fab 36 who knows Nehalem might beat K8L to market .

If its the big changes that AMD says it is . They have to have a proto type taped out first. (not a hand engineered die) Than it has a valadation period . It could be 08 befor we see the K8L product that people are now talking about.
Click to expand...
so now we have anywhere from early '07 to '08 for a launch date :D

and btw, fab36 is making 65nm chips. just not any for public consumption ;)

$BangforThe$ said:
No he is saying don't hold your breath for K8L to be out in 1st qt of 07 . Look for it in 4th qt of 07 . I think Amd can have K8L out by june of 07 but don't exspect to buy one til like sept. Than its only a few months to Wolf-Dale( A simple shrink it is) If K8L is OoO and reverse hyper threading and such than I would say Intel Wolfdale will beat it into the market.
Click to expand...
the reverse hyperthreading probably won't make it into K8L.. i honestly don't even know if we'll see it in K10, but that remains to be seem.
 
duby229 said:
Your implecation is that porting isnt worthwhile, I pointed out that wasnt the case. Performance is not really the issue.
Click to expand...
That's not the whole story though. There is more to it than just porting--debugging, repackaging, updating SKUs, dealing with the logistics of distributing the recompiled version, marketing/product placement, etc. And if performance isn't the issue, why would they recompile it at all? All x86-64 chips are fully 32-bit compatible, so it's not a matter of reaching out to a market that has not been tapped. Software developers would likely waste money and time trying to spin the usefulness of a product that has no increased utility. If developers can't get increased utility out of reworking a software package for 64-bit compatibility, they just won't do it.
 
$BangforThe$ said:
Its really not that far off . Till we see a 65nm die from Fab 36 who knows Nehalem might beat K8L to market .

If its the big changes that AMD says it is . They have to have a proto type taped out first. (not a hand engineered die) Than it has a valadation period . It could be 08 befor we see the K8L product that people are now talking about.
Click to expand...
This is a very pessimistic view on when K8L will be launched, if K8L is indeed launch in H1 2008, god help AMD.

Realistic views of where AMD has major impacts with their 65nm process are in H1 2007, so H2 2007 is not an unreasonable estimate for a K8L launch.

I am looking toward a launch at around H2 2007. It would be around the right time, the 65nm process is mature by then, and they already tested it on Brisbane/Tyler. Roughly 4 years after the introduction of K8 on desktop is not an unreasonable time for a new architecture.

I consider products from the 2008 timeframe, fairly far off, Nehalem is around mid to late 2008.
 
$BangforThe$ said:
Do you have a link to AMD producing a 65 nm die at fab 36
Click to expand...

Early samples of chips built on the company's 65nm manufacturing technology are healthy, Ruiz said. AMD is in the midst of preparing a new chipmaking factory in Dresden that will double its capacity once the plant comes online, he said.
Click to expand...

That's from January.
Linky.

You need a year or more to go from early prototypes of working parts to full production on a process.
Several years to go from producing transitors or simple circuits (like SRAMs) to full production on a process.

If AMD didn't have working 65nm parts internally right now, there is simply no way they'd hit a turn of the year timeframe for shipping production parts in volume.
 
K8L vs Conroe is so far off, its not even funny... AMD will probably release it on the server side waaay before the desktop....

So even if AMD f@nbois insist on pitting K8L vs a first iteration NGMA @ 65nm, it would be K8L vs Woodcrest.... and THAT is where I see a real dogfight..... past 4 sockets and up I bet this one is gonna be a brutal fight for the performance crown..... though for power consumption and chip price *something* in me just tells me that Intel will cost less.
 
i think conroe is going to hand K8L's ass too it, only thing the K8L is brining is shared L3 cache, (mite work well who knows) DDR2 suport and some other small optimizations, the conroe brings a brand new (Well to desktops) arctecture with mass ammounts of L2 cache, and much higher efficency per mhz and per watt (from netburst), and intel is in there 3rd generation of DDR2 memory controlers where AMD this is there first, plus not to mention its going to be uphill for AMD with 90nm vs 65nm in the begining.
 
Simpson5774 said:
i think conroe is going to hand K8L's ass too it, only thing the K8L is brining is shared L3 cache, (mite work well who knows) DDR2 suport and some other small optimizations, the conroe brings a brand new (Well to desktops) arctecture with mass ammounts of L2 cache, and much higher efficency per mhz and per watt (from netburst), and intel is in there 3rd generation of DDR2 memory controlers where AMD this is there first, plus not to mention its going to be uphill for AMD with 90nm vs 65nm in the begining.
Click to expand...

first off, spelling.

second, it's more than simply a "l3 cache addition".

0. Native quad core
1. Hypertransport up to 5.2GT/s
2. Better coherency
3. Private L2, shared L3 cache that scales up.
4. Separate power planes and pstates for north bridge and CPU
5. 128b FPUs - see 14,15
6. 48b virtual/physical addressing and 1GB pages
7. Support for DDR2, eventually DDR3
8. Support for FBD1 and 2 eventually
9. I/O virtualization and nested page tables
10. Memory mirroring, data poisoning, HT retry protocol support
11. 32B instead of 16B ifetch
12. Indirect branch predictors
13. OOO load execution - similar to memory disambiguation
14. 2x 128b SSE units
15. 2x 128b SSE LDs/cycle
16. Several new instructions

Coprocessors:
media processing
JVM/CLR acceleration
TOE, XML or SSL processing
Click to expand...

K8L will have its prefetches doubled for K8, it now has the ability to handle SSE instructions (no exact statemet as to whether just SSE, or SSE1/2/3 etc.) in a single cycle (which is one of Conroe's large advantages), new instructions, I/O virtualization, a much more easily adaptable IMC, additional FPUs, HyperTransport 3, and more.

K8L is being planned straight for 65nm. It won't be on 90nm...

DDRII is simply a means of providing larger amounts of bandwidth at higher speeds, in reality. AMD has shown that at the moment, AM2 is providing equal, if not very slightly higher performance for DDRII over DDR. This is on par with what Intel did with their change over from DDR to DDRII. If you honestly believe that "experience" in the terms you're suggesting, for DDRII implementation, is that big of a deal, than you're barking up the wrong tree...

However, if all of the above additions constitute "small optimizations", than how can you truly rate Conroe as that large of a jump over Yonah/Dothan/etc.
 
IF AMD is going to be doing all this to the Desk top K8L we won't see that die till 08 thats a die due over and AMD just put those additions into their K8L This is a major core change AMD isn't going to get it to market any faster than intel did Woodcrest . So thats an 08 chip Its won't be up against Conroe . Until Amd shows a working chip bekind closed doors thats 9 months out from that date At least. Till than its vapor ware Also Intel is 4 +1 Issue CPU. I don't see how you can even compare the 2 That were Intel gets their performance with really wide pipes.
 
I'm shocked it's civilized so far... threats work...lol

Hey E, you have that 631 yet? You could get a dual for slightly more. Though it's probably more for benchies and testing.

What DDR2 u looking at btw? I needed an MATX board and there really arent any good OCers in that category so thats why I didn't have a good experience with Intel. That Vdroop is killer!

AMD is probbaly out till DDR3 comes out which is a ways off. DDR2 wasn't used by video card makers becasue DDR3 was THAT much better. Do't know if it applies to system ram as it does VC but I'msure AMD is probably done being king for a while.
 
$BangforThe$ said:
IF AMD is going to be doing all this to the Desk top K8L we won't see that die till 08 thats a die due over and AMD just put those additions into their K8L This is a major core change AMD isn't going to get it to market any faster than intel did Woodcrest . So thats an 08 chip Its won't be up against Conroe . Until Amd shows a working chip bekind closed doors thats 9 months out from that date At least. Till than its vapor ware Also Intel is 4 +1 Issue CPU. I don't see how you can even compare the 2 That were Intel gets their performance with really wide pipes.
Click to expand...

Um, first of all, AMD's been working on K8L for a *very* long time. It's not like the engineers at AMD were all sitting around a table, a few days before the AMD show, and said "Hey,let's throw together a powerpoint and put in bits of info about what we want to add".

Right now, AMD has K8L listed for early 2007 (first half of 2007,with some of those roadmaps suggesting early '07). Even 6 months ago and longer, the earliest that a supposedly modified K8 (in more than just dual core/quad core/etc) was rumored to be released was late '06. Thus, as far as I see it, AMD really isn't that "behind" on their plans. If that diagram that they displayed is a true die of an early K8L, as I would suspect it is, than that means that they are father far along in the processor design process. Thus, a 1H 07 release is not as unlikely as you seem to think. I love how you automatically assume that AMD is only *now* starting K8L, and thus it can't obviously release earlier than 08.

Second, while the extra issue does help Conroe, it's already been stated by many people who are more knowledgeable concerning processor design than you and I, that the #1 current reason for most of the huge differences between Conroe and K8 is the fact that Conroe can handle SSE instructions in a single cycle.

I take this from Anandtech's architecture breakdown:

"Another impressive area is Core's SSE multimedia power. Three very powerful 128-bit SSE/SSE2/SSE3 units are available, and two of them are symmetric. Core will outperform the Athlon 64 vastly when it comes to 128-bit SSE2/3 processing.

On K8, 128-bit SSE instructions are decoded into two separate 64-bit instructions. Each Athlon 64 SSE unit can only do one 64-bit instruction at a time, so the Core architecture has essentially at least 2 times the processing power here. With 64-bit FP, Core can do 4 Double Precision FP calculations per cycle, while the Athlon 64 can do 3."

Now, they're taking what already exists with K8,and tacking on two more SSE units, although units that can handle single-cycle 128b instructions. Thus, while Conroe still has 3 128b units compared to the two being added to K8L, K8L will still have the current 128b SSE instruction set as well that decodes into two 64b instruction sets, and thus it'll be much more equalized situation.

It also appears as if AMD is implementing a modular IMC, prefetch instructions similar to Conroe, and while they may or may not be implementing larger amounts of L2 cache, they're going to implement L3 cache. If they make use of ZRAM for the L3 cache, and they could very well given the timeline for K8L, then you could very easily see a *huge* L3 cache reserve.

Does it mean that K8L will smash Conroe's architecture? Who knows. It might. It might simply give roughly equivalent performance. We'll see come early/mid 07. Either way, it's great to see you Intel fans rushing to descredit anything AMD does.
 
$BangforThe$ said:
Memory mirroring is something intel can do also with fb -dimms.
Click to expand...

That is a feature of the MC inside the chipset.AFAIK , Xeon servers from the big guys have had it since 2002 at least.
 
You must log in or register to reply here.
Back
Top