New AMD CPU Architecture Announced!

[Ika]

Originally posted by: cmrmrc

Originally posted by: MDme
Now the problem is.....

Do I buy Conroe or wait for K8L? (sigh)

I originally wanted to upgrade my S754 Athlon64 @2.6 Ghz to an X2 939...but then AM2 made me wait....but seems like AM2 is not worth it so Conroe seemed like a good idea since it's coming out soon.....but now this??? So do I wait for 2007? when is it coming out?? 1H, 2H? aaaarrgh!!!

you will always wait if you think of it this way...thing is don't wait until something good is not on the horizon so that you can buy a system...buy the system that fits you at that moment and be happy with it

when you'll wait for K8L...you'll probably gonna be hearing details about intel's new core nehalem..and then if you wait again..k10 may be leaked...and so on..

or, make up your mind to buy a new computer at a certain time (for me, after Vista and DX10 cards are released and benched). Either that, or wait so long that you can't play any good games anymore and need to get a new comp just to do the stuff you want to do.

 

[MDme]

Well actually I anticipated these responses to my post and quite frankly you guys are right...BUT the difference in my situation is, I still feel I have enough power to do the things I need to do (except high-end gaming - since I have an AGP 6800GT OC). So there is no NEED for me to upgrade other than the annual itch to get more cutting edge.

So I guess if Conroe is really good and the price is right then I'd get Conroe now then K8L in 2007 if it is really better.

If AMD was smart, they can make the AM2 socket compatible with K8L so that you can buy AM2 now then upgrade to K8L when it is available, because i think the AM2 X2's and 65nm should be enough to hold us until K8L.

 

[Hard Ball]

Originally posted by: Fox5


Will first gen Am2 motherboards support K8L?

Not sure, but I doubt it; If K8L are to have FBDIMM and DDR3 support, it's hard to see how AM2 infrastructure will support it. And with the additional voltage control elements, I'm sure the current generation of platform (including AM2 and S939, which are derived from S940), it is improbable that it can be implemented on current platform, with the current arrangement of power and ground pins. And AM3 is probably just around the corner, and would probably have similar pin count to that of Socket F (1207), and LGA.

 

[dmens]

Originally posted by: zephyrprime
Having out of order loads will be interesting if it really has it.

Like the original P6? Sweet!

 

[Hard Ball]

Originally posted by: dmens

Originally posted by: zephyrprime
Having out of order loads will be interesting if it really has it.

Like the original P6? Sweet!

I doubt that he is referring to loads into register file; it's referring to speculative loads into L2/L3.

 

[dmens]

Originally posted by: Hard Ball

Originally posted by: dmens

Originally posted by: zephyrprime
Having out of order loads will be interesting if it really has it.

Like the original P6? Sweet!

I doubt that he is referring to loads into register file; it's referring to speculative loads into L2/L3.

the inq reporter is seriously confused, because he talks like he is referring to juggled loads, and that has been around forever. nobody refers to speculative loads (aka prefetching) as "out of order loads".

in any case, this is interesting info, but unfortunately there is no word on machine depth/width, assuming K8L is still of K8 style, most likely it is. that's what is really important!

 

[Furen]

Originally posted by: dmens

Originally posted by: zephyrprime
Having out of order loads will be interesting if it really has it.

Like the original P6? Sweet!

Like the K7 and K8 DON'T have...

Now, according to my understanding rev G (Brisbane, AMD's first 65nm chip) was supposed to be K8L. I dont know if this has changed but considering that we've been hearing about 65nm AMD chips coming out (for OEMs) before the end of this year K8L could be closer than most think, either that or initial 65nm chips could be a dumb shrink of rev F. I figured a major revamping of the front end would be necessary to keep the K8L working at a decent pace but I wonder how these enhancements will affect non-FP/non-SSE performance. I'm tempted to say that I'm a bit skeptical about what these improvements will yield (performance-wise) but I said the same thing about Conroe and it does look like this last one will increase performance over Yonah quite a bit.

 

[Hard Ball]

Originally posted by: dmens

Originally posted by: Hard Ball

Originally posted by: dmens

Originally posted by: zephyrprime
Having out of order loads will be interesting if it really has it.

Like the original P6? Sweet!

I doubt that he is referring to loads into register file; it's referring to speculative loads into L2/L3.

the inq reporter is seriously confused, because he talks like he is referring to juggled loads, and that has been around forever. nobody refers to speculative loads (aka prefetching) as "out of order loads".

I think that's a matter of opinion, I have heard several of my past systems / architecture profs referring to "out of order loads" along with LRUL and eviction in the same sentence; and some of them are very well known quantity in the field.

I don't know which school of engineering you went to / are going to, but you might have to consider that terminology at different firms or research institutions are different.

in any case, this is interesting info, but unfortunately there is no word on machine depth/width, assuming K8L is still of K8 style, most likely it is. that's what is really important!

Right, AMD is not telling much of substance yet (such as this "speculative load" is not as explicitly explained as CMA memory disambiguation), although there are some veiled hints at what maybe coming; but I think Henri Richard sometime in the last month promised some details at Computex.

 

[coldpower27]

Originally posted by: Furen

Originally posted by: dmens

Originally posted by: zephyrprime
Having out of order loads will be interesting if it really has it.

Like the original P6? Sweet!

Like the K7 and K8 DON'T have...

Now, according to my understanding rev G (Brisbane, AMD's first 65nm chip) was supposed to be K8L. I dont know if this has changed but considering that we've been hearing about 65nm AMD chips coming out (for OEMs) before the end of this year K8L could be closer than most think, either that or initial 65nm chips could be a dumb shrink of rev F. I figured a major revamping of the front end would be necessary to keep the K8L working at a decent pace but I wonder how these enhancements will affect non-FP/non-SSE performance. I'm tempted to say that I'm a bit skeptical about what these improvements will yield (performance-wise) but I said the same thing about Conroe and it does look like this last one will increase performance over Yonah quite a bit.

Well if going by past AMD expereinces is any indicatio expect Brisbane to be a dumb 65nm shrink of Windsor with K8L derivatives in the second half of 2007, when AMD gets more of a hang of the 65nm process. There isn't any indication to suggest they have deviated from this slow and steady do it right approach.

 

[Viditor]

Originally posted by: coldpower27

Originally posted by: Furen

Originally posted by: dmens

Originally posted by: zephyrprime
Having out of order loads will be interesting if it really has it.

Like the original P6? Sweet!

Like the K7 and K8 DON'T have...

Now, according to my understanding rev G (Brisbane, AMD's first 65nm chip) was supposed to be K8L. I dont know if this has changed but considering that we've been hearing about 65nm AMD chips coming out (for OEMs) before the end of this year K8L could be closer than most think, either that or initial 65nm chips could be a dumb shrink of rev F. I figured a major revamping of the front end would be necessary to keep the K8L working at a decent pace but I wonder how these enhancements will affect non-FP/non-SSE performance. I'm tempted to say that I'm a bit skeptical about what these improvements will yield (performance-wise) but I said the same thing about Conroe and it does look like this last one will increase performance over Yonah quite a bit.

Well if going by past AMD expereinces is any indicatio expect Brisbane to be a dumb 65nm shrink of Windsor with K8L derivatives in the second half of 2007, when AMD gets more of a hang of the 65nm process. There isn't any indication to suggest they have deviated from this slow and steady do it right approach.

Well the roadmaps are fairly clear that the K8L will be in H1 07, but I wouldn't be surprised if we see an initial release at the end of this year that is a dumb 65nm shrink as well...
They'll probably sell them with the same model numbers as the 90nm versions (which they did for a few months when the 90nm shrink launched) until the K8L versions start shipping.

 

[Griswold]

For what it's worth. No idea how credible this is, but we'll find out soon enough I guess.

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

From:

Real World Technologies

 

[HopJokey]

Originally posted by: Griswold
For what it's worth. No idea how credible this is, but we'll find out soon enough I guess.

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

From:

Real World Technologies

On 65nm how big will the die size be? I guess that depends on the cache size, but from the looks of that list, it will pack a lot of transistors for the quad core version.

 

[lsman]

K8L 2008.....

http://www.hkepc.com/bbs/viewthread.php?tid=600827&extra=page%3D1
http://www.hkepc.com/bbs/viewthread.php?tid=600836&extra=page%3D1

- K8L support DDR2, DDR3 and FB-DIMM....
- K8L support RAS

 

[Viditor]

Originally posted by: lsman
K8L 2008.....

http://www.hkepc.com/bbs/viewthread.php?tid=600827&extra=page%3D1
http://www.hkepc.com/bbs/viewthread.php?tid=600836&extra=page%3D1

- K8L support DDR2, DDR3 and FB-DIMM....
- K8L support RAS

K8L 2007 H1

Link

 

[PentiumIV]

Did AMD tape out the K8L and do they have some live silicon in their labs?
It should take ~1 year from A-0 step to production....

 

[Viditor]

Originally posted by: PentiumIV
Did AMD tape out the K8L and do they have some live silicon in their labs?
It should take ~1 year from A-0 step to production....

They have announced neither, but then again they've stopped releasing long term roadmaps as well...AMD has been very cagey about their tech for almost a year now, they release solid info only when it's close to launch.

 

[Duvie]

Awhile back they announced they had working 65nm for some time in the labs...I would assume it likely was more then just a 65nm AM2 version...

They definitely are keeping their cards close to them....Makes me wonder if they are going to try to play the spoiler later on....

Who knows!!! Either way it is getting exciting again.....

 

[kknd1967]

different updated info from the same HKePC:
2007H2 and 2008
Maybe all just rumors...

The last paragraph
"AMD???2007?????????????????Deerhound,??????????,????2008??????????K8L???,???????Greyhound???????Cadiz???????Zamora?"
"AMD plans to launch 4-core server processor Deerhound in 2007H2, which is not based on the new architecture. It is not until 2008 H1 that products based on K8L will show up, including desktop processor Greyhound, workstation processor Cadiz and server processor Zamora."

Originally posted by: Viditor

Originally posted by: lsman
K8L 2008.....

http://www.hkepc.com/bbs/viewthread.php?tid=600827&extra=page%3D1
http://www.hkepc.com/bbs/viewthread.php?tid=600836&extra=page%3D1

- K8L support DDR2, DDR3 and FB-DIMM....
- K8L support RAS

K8L 2007 H1

Link

 

[Viditor]

Originally posted by: kknd1967
different updated info from the same HKePC:
2007H2 and 2008
Maybe all just rumors...

The last paragraph
"AMD???2007?????????????????Deerhound,??????????,????2008??????????K8L???,???????Greyhound???????Cadiz???????Zamora?"
"AMD plans to launch 4-core server processor Deerhound in 2007H2, which is not based on the new architecture. It is not until 2008 H1 that products based on K8L will show up, including desktop processor Greyhound, workstation processor Cadiz and server processor Zamora."

Originally posted by: Viditor

Originally posted by: lsman
K8L 2008.....

http://www.hkepc.com/bbs/viewthread.php?tid=600827&extra=page%3D1
http://www.hkepc.com/bbs/viewthread.php?tid=600836&extra=page%3D1

- K8L support DDR2, DDR3 and FB-DIMM....
- K8L support RAS

K8L 2007 H1

Link

They are only talking about the quad core K8L server chips, not all K8L there...though I have to admit that I did several double takes when I read it as well.