K8L sooner then you think...

[Serge84]

http://www.translate.ru/url/tran_url...&psubmit2.y=12

K8L sooner then you think... So they showed K8L die shots in the presintation for opteron quad cores, hmm 65nm talk, L3 and so forth. More real info about K8L In the AMD's presintation. Intresting. There is no delay. Still 2007. And more info is out on the core.

[arisythila]

This looks awesome.

~Mike

[freeloader]

Lots of good information!

Nice find...

[Khenglish]

It says K8L will still have a 128KB L1 cache. If AMD doubles the bandwidth and associativity they'll have a major L1 cache advantage. This alone should raise performance a lot. Doubling the L1 cache size is the main factor what gave the pentium-m its tremendous IPC leap over the P3.

It also says they can do 4 fp instructions per clock. Hopefully there's a 4th decoder too.

I'm wondering how much of a boost the 128bit loads will give. It may be a lot since loads are by far the most common instruction (usually 33% of all instructions)

Load out of order execution should be around a 5% boost overall, and around 10% for integers.

It really seems like K8L is going to beat out Conroe. The only deciding factor left is how many decoders K8L has.

[Charles Wirth]

Oooh, more slides.

[arisythila]

Thats what im talking about Fugger!

~Mike

[STEvil]

with my luck K8L will come in laptops soon as well.

[freeloader]

The egg is going to fly when AMD releases new chips.

Fugger will be the first to get a good dose of it on his face!

No ES for you!

[DoubleZero]

Oooh, more slides.

Bottom Line, Intel needed almost 3 years to respond to K8, AMD will respond in 8-10 months to c2d, Intel then will take another year and a half to respond to K8L with nehalem, in the most optimistic projection, probably K10 then will follow suit, so yes AMD is having a lot of trouble in following Intel amazing roadmap and has only slides to show

I just hope that Intel didn't use slides to explain to its investors how can be possibly that AMD stock as already recuperated form the plunge that it took after acquiring ATI, and Intel stock despite having a new product that have taken the performance crown, showed zero improvements.

[cky2k6]

thats simple, amd is still riding the opteron wave, while intel is trying to get rid of netburst stock and cutting jobs... both sides know, you dont win with just pure performance.

[Sparky]

Very interesting. I might have a reason to upgrade sooner than I thought. But that may still depend on how far away AM3 is - I was planning to skip AM2. We'll see.

[informal]

AM3 is late '07 early '08.AM2 supports K8L(DC and QC) and coupled with good DDR2 modules can deliver enough bandwith to feed AM3 CPUs.
Later in 2008,you can sell your mobo and go AM3 all the way with good DDR3 memory.

[Anemone]

Delivery dates from AMD typically lag advanced roadmaps by about 2 quarters. That said, frankly we need K8L, for market reasons. As much as it is risky to show, we need to see results, from working chips. A lot of people could likely accept current AM2 performance, and wait for a promised upgrade if they had better understanding of what they were going to get.

I'm not sure I buy into the triple layer cache as the sure way to performance. It seems even contradictory to the basic theme of AMD architecture (looking at the on die mem controller) which is to cut out middle layers and give as many portions of the overall system as possible, direct connections.

They also mention one of the cache advantages as lowering memory latency, which, while true, doesn't seem to be much of AMD's weakness really. Maybe the creator of the powerpoint just needed something to say on the topic and picked that. It also appears that the shared cached designs are something AMD would like to mimic, while retaining L2 on the cores themselves, so possibly that's the real answer.

Direct cross connects, 3 cache levels, and out of order executions seem like a lot of things to keep straight without errata cropping up all over the place. AMD is the one to pull it off if it can be done though. Could be one expensive chip however.

I guess I'm really hoping they manage to get this pulled into late Q1 early Q2 07. Again, so this race has two more equal competitors.

[oldblue]

Charlie backtracks.

In brief (from AMD, via Charlie):

Q2 2007: K8L "Barcelona", quad-core, HT 2.0
Q4 2007: K8L "Budapest", quad-core, HT 3.0, 12xx
Q1 2008: K8L "Shanghai", quad-core, HT 3.0, 22xx, 82xx

65nm seems to be on track for end of this year.

Also, K8L = rev H.

[Serge84]

It says K8L will still have a 128KB L1 cache. If AMD doubles the bandwidth and associativity they'll have a major L1 cache advantage. This alone should raise performance a lot. Doubling the L1 cache size is the main factor what gave the pentium-m its tremendous IPC leap over the P3.

It also says they can do 4 fp instructions per clock. Hopefully there's a 4th decoder too.

I'm wondering how much of a boost the 128bit loads will give. It may be a lot since loads are by far the most common instruction (usually 33% of all instructions)

Load out of order execution should be around a 5% boost overall, and around 10% for integers.

It really seems like K8L is going to beat out Conroe. The only deciding factor left is how many decoders K8L has.

Theres even a 4th decoder in the 65nm K8's.

[ozzimark]

Theres even a 4th decoder in the 65nm K8's.

really? i was under the impression that revG was going to be a dumb shrink

[Serge84]

really? i was under the impression that revG was going to be a dumb shrink

http://129.15.202.185/athlon_rev_g/6...hancements.jpg

Here is rev G and its changes shown here have been added. Rev F is different then rev G, compare it with rev F die shots.

http://129.15.202.185/athlon_rev_g/a64_rev_f.jpg

This is rev F. See the difference? Only 3 decoders. And some parts missing compared to a rev G that has improvments. Also the new SiGe will be used in them with a few new fab prosesses in the mix. It will allow AMD to hit higher clocks. FX-64 and 66 will use SOI-3 as well to improve transistor speed but stay on 90nm for awile.

http://www.c627627.com/AMD/Athlon64/

[dinos22]

if AMD work out Pacifica so that i can use the PC as HTPC and gaming machine separately & independently i'll be all over it like a rash because Intel VT ain't up to it

[dinos22]

Bottom Line, Intel needed almost 3 years to respond to K8, AMD will respond in 8-10 months to c2d, Intel then will take another year and a half to respond to K8L with nehalem, in the most optimistic projection, probably K10 then will follow suit, so yes AMD is having a lot of trouble in following Intel amazing roadmap and has only slides to show

I just hope that Intel didn't use slides to explain to its investors how can be possibly that AMD stock as already recuperated form the plunge that it took after acquiring ATI, and Intel stock despite having a new product that have taken the performance crown, showed zero improvements.

it is still TOO EARLY to make such claims........let's see these products out on XS with people testing them. Only then will we know how good or not they really will be

[Zardokk]

Yeah. We really don't know how the real-world performance of K8L will be. It's too soon to make any claims. Once we see these chips out and about, being tested, then we'll know the truth. My prediction is that K8L will give about 15-20% performance boost over Conroe in real-world situations, maybe quite a bit more in benchmarks. We simply don't have enough apps that utilize multiple cores...maybe if AMD hurries up with their "anti-hyperthreading."

[dinos22]

Yeah. We really don't know how the real-world performance of K8L will be. It's too soon to make any claims. Once we see these chips out and about, being tested, then we'll know the truth. My prediction is that K8L will give about 15-20% performance boost over Conroe in real-world situations, maybe quite a bit more in benchmarks. We simply don't have enough apps that utilize multiple cores...maybe if AMD hurries up with their "anti-hyperthreading."

i'd love to know how you could come up with sucha conclusion

if indeed 65nm added another decoder it should merely make it even with performance of Conroes depending on the L2 cache and RAM performance.........What will give them a leap ahead..........i could understand such prognosis had there been an inclusion of L3 cache in large size but don't see how it is possible otherwise.......

[The Ghost]

i'd love to know how you could come up with sucha conclusion

he gave his prediction , not a conclusion

i would hope that he is close to his prediction and it becomes a conclusion

[Serge84]

i'd love to know how you could come up with sucha conclusion

if indeed 65nm added another decoder it should merely make it even with performance of Conroes depending on the L2 cache and RAM performance.........What will give them a leap ahead..........i could understand such prognosis had there been an inclusion of L3 cache in large size but don't see how it is possible otherwise.......

Please don't make me post another topic on explaining what the K8L arc can do. Physicly the die shots show a 2nd FPU giving it 2x the FPU performance and more ALU's giving it about 25% to 40% performance. Its more then just that, it will use dual 128-bit processes and wide pipes. The Cache is 256-bit. All this simply allows conroe like processes. It will be faster then conroe period. K8L will give 50% to 80% more performance then K8's. If you want to get all the info on why this is possible look at my past posts on all my K8L responces. I don't feel like making another speech.

Look at die shots and look at the ones that tell exactly what everything is. Nothing is if anymore. AMD showed the die shots and specs. It will do what we think it will who can understand what the parts do just by looking at the die. It will perform 2 128-bit instructions in one cycle so it will match conroe right there not including the other parts. The only thing that can compete is DUO3.

[dinos22]

Please don't make me post another topic on explaining what the K8L arc can do. Physicly the die shots show a 2nd FPU giving it 2x the FPU performance and more ALU's giving it about 25% to 40% performance. Its more then just that, it will use dual 128-bit processes and wide pipes. The Cache is 256-bit. All this simply allows conroe like processes. It will be faster then conroe period. If you want to get all the info on why this is possible look at my past posts on all my K8L responces. I don't feel like making another speech.

Look at die shots and look at the ones that tell exactly what everything is. Nothing is if anymore. AMD showed the die shots and specs. It will do what we think it will who can understand what the parts do just by looking at the die.

that's what i thought

[marauder16]

Yeah. We really don't know how the real-world performance of K8L will be. It's too soon to make any claims. Once we see these chips out and about, being tested, then we'll know the truth. My prediction is that K8L will give about 15-20% performance boost over Conroe in real-world situations, maybe quite a bit more in benchmarks. We simply don't have enough apps that utilize multiple cores...maybe if AMD hurries up with their "anti-hyperthreading."

there is no such thing, that was just a rumor...