First K8L details!

[Carfax]

Directly from AMD themselves, and Groo over at the Inq:

1) Native quad core design

2) Shared expandable L3 cache

3) Doubling of SSE/FP capabilities and extra instructions

4) More aggressive prefetch

5) RAS improvements

Some other stuff aswell!

Looks like AMD has no intention of surrendering the performance crown for any length of time

[GoThr3k]

prefetch is no 32B instead of 16B (the inquirer) so they doubled it
i would like to know the impact of that thing
shared L3 is very nice also
and doubling SSE/FP capabilities just had to be done to at least match intel a bit

nice work AMD

[Carfax]

I think AMD will have the advantage in a quad core design over Intel due to a superior platform and infrastructure.

The K8L will be battling the quad core version of Conroe (Penryn), which will most definitely have some architectural improvements, and could possibly be on 45nm by that time aswell.

[n91htmare]

A bit late IMO.... K8L isn't set out to be due til late 07. Doubt intel will sit still until late 07...

Good to see some news though but not from a site that i'd trust.

[GoThr3k]

is k8l late 07???
i thought is was begin/mid 07 to counter kentsfield

[DoubleZero]

A bit late IMO.... K8L isn't set out to be due til late 07. Doubt intel will sit still until late 07...

Good to see some news though but not from a site that i'd trust.

65nm in november 2006, and K8L late 2007? Doesn't make sense.
I think K8L will apear 1Q 2007 as been scheduled to apear for a long time.

[n91htmare]

I thought the 65nm wasn't going to hit market till early 07 and then the new K8L architecture late 07... Only thing we're expecting from AMD this year is AM2 and AM2 90nm processors..... Though i'm prolly wrong

[vitaminc]

Desktop Quadcores are not coming until 1H 2008.

Nothing to get excited about for me until AMD decide to bring out more desktop goodness.

[GoThr3k]

unless they are getting some #sswhopping from intel maybe

[cupra]

Desktop Quadcores are not coming until 1H 2008.

2008 ?
if its truth that is a slow response from amd, disappoint.

[vitaminc]

65nm in november 2006, and K8L late 2007? Doesn't make sense.
I think K8L will apear 1Q 2007 as been scheduled to apear for a long time.

AMD's first quadcore processor for desktops will be in 2008.


Quadcore Roadmap:
http://www.hkepc.com/bbs/news.php?tid=593736

[cky2k6]

yes of course, its a proven scientific fact... youre like the opposite of nn_step... nobody knows when k8l is coming out, and we wont at least until the end of 2006.

[Vapor]

yes of course, its a proven scientific fact... youre like the opposite of nn_step... nobody knows when k8l is coming out, and we wont at least until the end of 2006.

Bad. Okay.

Let's try to keep it VERY clean in here guys

[mjp1618]

2008 may be too late.

Intel's next microarchitecture Nehalem is scheduled for 2008; and native quad cores at 45nm for 2007.

However, mid-2007 K8L will shake Intel up a bit, provided they work as promised...

[cky2k6]

mid 2007 is a much more believable figure than 2008. i saw the website that posted 2008, and their roadmap was fud, they even forgot to list some present core name. amd might be behind intel in production capabilities somewhat, but theyre not that futile. 2008 would be suicide if they want to actually stay around in the desktop market, so that site seems nothing more than some intel propoganda. please dont sound so assured as if you know something we all dont, vitaminc, as it is misleading...

[vitaminc]

I don't know if the website I sited is misleading or not, but they do have pretty good track record of getting new informations, as DailyTech and Inquirer both refer to it several times.

Some cores might be missing, yes. But that roadmap is for Quadcores only. I don't know any other quadcore projects other than those.

And mind you, the first quadcore Deerhound server MP processor is due in 2H07 according to the HKEPC published roadmap. That's the same as AMD's official roadmap AFAIK.

The first quadcore DESKTOP processor in 2008 doesn't sound that bad at all.

I wouldn't call HKEPC Intel's propoganda as they are the first to publish AMD's AM2 offerings, Turion X2 offerings, Quadcore roadmaps, and now 65nm AM2 roadmaps.

[Pjotr]

"AMD desktop quad core in 2008" and server quad core 2007. Intel will also come out with server quad core during 2007.

The market for consumers getting quad core is much smaller than consumers wanting dual core. Dual core gives faster multitasking, quad core will only give benefit for applications that are multithreaded, for multitasking you won't notice anything over dual core.

[npk]

Perhaps games and a lot more apps will be multithreaded by 2008.

[mesyn191]

Actually they really couldn't do much about it, they are contract limited with Intel to not outsource more than 25% of thier total fab capacity. AFAIK Chartered is also only producing Semprons and such....

I'd agree about the die size on .65, unless AMD has done something to really cut down on logic or SRAM size this thing will be hot and huge, performance should be good though barring any major screwups of course...

If it works on AM2 THAT will be nice, though being quad core I doubt it. According to Inq sched. release date in 2007 Q3 which seems optimistic unless they've really been working on it for a long time, when were thier other projects cancelled again, mid-2005? I dunno, seems too soon for such a major revision to able to ship if they've only been working since that time.

I know other road maps have said desktop quad core in 2008 only, that may be true but remember, those release times aren't set in stone. They can be pushed back or pulled up, generally the former rather than the latter though... The difference is only a handful of months off from what thier old dates were too, I think its unlikely but Q3 2007 may be possible.

Things get interesting if you start thinking about what they'll be able to do with that HTX slot support for every CPU. Think something like an Ageia PPU but more general purpose so it can help with Folding/SETI or video encoding or rendering, with low latency access to the CPU and main memory sustainable performance could be huge, ie. Cell-ish performance for certain work loads.

[Vapor]

The market for consumers getting quad core is much smaller than consumers wanting dual core. Dual core gives faster multitasking, quad core will only give benefit for applications that are multithreaded, for multitasking you won't notice anything over dual core.

I wouldn't dismiss the power of change that software can under-go in a short timespan. Look at games, no, the actual software hasn't changed, but the drivers have been made to make use of the unused core and speed things up that way. That's only in the VERY short term.

In another year or so, most games made will actually be multithreaded and most multimedia applications already are. Pretty much anything that will be CPU intensive will be multithreaded.

Another thing to think about is that a lot of what we do now, stresses one core. The multitasking ability comes from that second core still being free to do stuff and not slow down the original task. In the near-future, things we'll be doing will take up more than one core's CPU-time. Yes, they'll be able to back down to 1, but that'll slow it down. So by then we'll have the tech to have four cores and have them only back down to 3, barely slowing them down.

Also, the dual core 'scene' wasn't great even 3 months after AMD brought them to shelves. People were still pessimistic about them. Now it's rare for a recommendation to not include a dual core proc (and it's only ~7-8 months later).

The age of multitasking and multithreading is still young....I wouldn't be doubtful if by 2k7 we'd be asking for even more cores.


K8L as an architecture looks great...it really does. These aren't things that are getting patched, these are raw improvements, VERY much akin to Yonah -> Conroe/Merom. It'll be interesting to see what AMD might be able to do in the meantime. I do not think they'll be able to match Conroe/Merom in 2006, but if they bring something to market that can at least fill a niche that Conroe/Merom can't fill, they'll still be okay and not lose a huge portion of marketshare or consumer confidence.

[mesyn191]

Yea, right now multi-threading is still being worked into most programs so releasing quad core on desktop would be silly for most people, but in late 2007 it'll make alot of sense I bet.

[mesyn191]

K8L as an architecture looks great...it really does. These aren't things that are getting patched, these are raw improvements, VERY much akin to Yonah -> Conroe/Merom. It'll be interesting to see what AMD might be able to do in the meantime. I do not think they'll be able to match Conroe/Merom in 2006, but if they bring something to market that can at least fill a niche that Conroe/Merom can't fill, they'll still be okay and not lose a huge portion of marketshare or consumer confidence.

I think on the desktop market they're screwed til' K8L comes out, so about a year of financial/market share reaming without lube barring any mis-steps in thier product or processes, in which case it'll be even worse...

Server market they've got a toe hold though, and for quad and up machines I think they'll still sell well so I don't think that they'll be in as dire financial straits as they were back when it was Thoroughbred vs. 2.4C Northwood but they are gonna be hurting.

Only way I can see them compete is on price in the desktop market, and that hurts AMD more than it hurt Intel, it'll be nice having some good* cheap dual core options available.

*I ain't saying that those PD 805's are bad mind you, just that even at around 3.8-4Ghz even a X2 3800+ still holds up pretty well against them. If AMD gets thier dual core prices to somewhere around there and they OC as well as they currently do it'll be well worth it to go AM2 (though not at launch though, price hikes.... ).

[Carfax]

Here's a more comprehensive list of all the improvements, quoted from David Kanter who was kind enough to list all the improvements so my lazy self wouldn't have to

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

The only problem AMD may have is with the cache. If AMD doesn't extensively rework the cache of the K8L, then they'll run into bandwidth problems because the execution units won't be optimally fed.

Thats one area where Intel has always excelled. They definitely know how to make an impressive cache!

Intel will own the rest of 2006 to mid 2007 it seems with Conroe and Merom.

AMD has a good chance of taking back the performance from with the K8L, but Intel won't be sitting idly either.

I'm sure that they plan to introduce 45nm parts next year.. Penryn will be on 45nm, and will certainly hit 4ghz (for the desktop model), along with other architectural improvements.

The computing world will be VERY interesting through to 2007 guys

[vitaminc]

Server market they've got a toe hold though, and for quad and up machines I think they'll still sell well so I don't think that they'll be in as dire financial straits as they were back when it was Thoroughbred vs. 2.4C Northwood but they are gonna be hurting.

QFT. Server markets have much longer product life cycle compare to desktops (just think how much CPU have you burn thru in the past year OC'ing).

But the most advantage I see from AMD is HT.

Memory controller is a design choice. And All other nifty features AMD has in K8L are probably going to appear in Intel's real quadcores. And vice versa. (Intel has first to market shared cache architecture over AMD, mind you)

The sustainable differentiater is HT.

Only way I can see them compete is on price in the desktop market, and that hurts AMD more than it hurt Intel, it'll be nice having some good* cheap dual core options available.

*I ain't saying that those PD 805's are bad mind you, just that even at around 3.8-4Ghz even a X2 3800+ still holds up pretty well against them. If AMD gets thier dual core prices to somewhere around there and they OC as well as they currently do it'll be well worth it to go AM2 (though not at launch though, price hikes.... ).

I am really expecting a huge price cut from AMD because AM2 processors at their current price point isnt attractive for me.

I sold my Athlon XP 1800+ for $50 bux on ebay. My A7V333 mobo for $60 bux on ebay. What can I say? AMD shizzles are overpriced at the moment.

Here's a more comprehensive list of all the improvements, quoted from David Kanter who was kind enough to list all the improvements so my lazy self wouldn't have to

I think he missed the independent core voltage/multiplier turn down for power savings. That's important imo.

[mesyn191]

Overpriced? Eh, I dunno, they've got the performance lead @ stock ATM as Conroe aint' out yet so they've got some justification for it.

Expensive? Hell yea, they are definitely that ATM.

BTW, carfax ty for posting the list of stuff.