AMD K8L ES coming in December

My point in bringing up AMD's overly-optimistic early estimates for Fab 36 and their 65nm shrink timeline was meant simply as a reminder that AMD, like most publically held corporations, tends to provide overly optimistic long term estimates to entice new investors and satisfy existing investors. Right now AMD is saying that they will start shipping 45nm cpus in mid-2008 and I'm weary of seeing some people take such estimates as gospel truth only to forget all about them when that timeline is later amended to H1-09 or H2-09.

As for Brisbane overclocking potential, I'm just saying not to expect 3.5-3.6GHz from the first revision. Later revisions might do that but to expect such impressive overclocks from early revisions that AMD won't validate above 2.6GHz is just setting yourself up for a big disappointment, IMO.

I didn't mean to bash AMD here. I merely meant to temper what I see as unrealistic optimism with a dose of reality.

Quote:

---

I didn't mean to bash AMD here.

---

yea , i know you better then that too

things was going good for amd , they did not see the need to go to 65nm , now they do

Quote:

---

Originally Posted by The Ghost

yea , i know you better then that too

things was going good for amd , they did not see the need to go to 65nm , now they do

---

Yeah, they probably figured to hell with profits, we'll just stick with 90nm for another 6 months while Intel kicks our ass in profit per cpu and performance per watt. :confused:

Quote:

---

Originally Posted by Fred_Pohl

Yeah, they probably figured to hell with profits, we'll just stick with 90nm for another 6 months while Intel kicks our ass in profit per cpu and performance per watt. :confused:

---

umm what makes you think it is more cost effective to make something on a smaller process? usually inorder to make things on the smaller process you need to spend even more on the wafers which may or may not make the processors cheaper to produce. It all depends on the yeilds. :stick:

Quote:

---

Originally Posted by nn_step

umm what makes you think it is more cost effective to make something on a smaller process? usually inorder to make things on the smaller process you need to spend even more on the wafers which may or may not make the processors cheaper to produce. It all depends on the yeilds. :stick:

---

Yields being the same, the smaller process will always be more cost effective. More cpus per wafer equals lower costs and higher profit margins. AMD and Intel don't spend billions on process shrinks and new fabs to lose more money.

Quote:

---

Yields being the same,

---

and how do we know that the yields are going to be the same ?

it also takes a lot longer to make the wafer , something like 3 months from start to finish

Quote:

---

Originally Posted by Fred_Pohl

Yields being the same, the smaller process will always be more cost effective. More cpus per wafer equals lower costs and higher profit margins. AMD and Intel don't spend billions on process shrinks and new fabs to lose more money.

---

That is the thing when a new process comes out the yields are LESS.
infact the manufacturing cost of Early Yonah processors were similiar to the manufacturing cost of DEC's 21064 processors despite the fact that yonah is on 65nm and the 21064 was on a 3 layer 0.75µ process :stick:
Smaller ONLY MEANS CHEAPER IF THE PROCESS YIELDS ARE GOOD.
And just a little fact for you, ALL NEW PROCESSES HAVE BAD YIELDS. Hence AMD is switching to 65nm when it suits them BEST, not before

Quote:

---

Originally Posted by Fred_Pohl

I didn't mean to bash AMD here. I merely meant to temper what I see as unrealistic optimism with a dose of reality.

---

I thought that was what you were doing too, just didn't agree with all your reasons.

:toast:

Quote:

---

Originally Posted by nn_step

That is the thing when a new process comes out the yields are LESS.
infact the manufacturing cost of Early Yohan processors were similiar to the manufacturing cost of DEC's 21064 processors despite the fact that yohan is on 65nm and the 21064 was on a 3 layer 0.75µ process :stick:
Smaller ONLY MEANS CHEAPER IF THE PROCESS YIELDS ARE GOOD.
And just a little fact for you, ALL NEW PROCESSES HAVE BAD YIELDS. Hence AMD is switching to 65nm when it suits them BEST, not before

---

Huuuhh, they've been out for long enough that you spell them correctly, it's YONAH :stick:

Quote:

---

Originally Posted by alayashu

Semprons give over 50% of AMD's revenue, so i guess (if that unoficial roadmap is true) they're using the 65nm fab fully for their best selling CPU's and not the ones that do not sell very well...

---

Bingo, when AMD made the move to 90nm with winchester they released 3000+, 3200+, and 3500+ processors because there was more volume there. While not all of the winchesters clocked great i had a very early 3200+ that hit 2.77ghz stable, (i believe) before the fx-57 even came out. Just because they arent releasing at high clock speeds doesnt mean the process sucks. It makes a lot more sense to have your high volume parts be on your smallest process because you can make a whole lot more of them. Also, since your early process will be spotty with parts that clock and some that clock low you can focus on your mature process for your highest speed low volume parts.

Now realistically, i think that with the early parts youre gonna see a lot of variation but you will definitely see a good amount hit in the 3.2-3.3 range, not amazing, but it barely happens with current silicon/process.

Quote:

---

Originally Posted by informal

Perkam,there is NO RevG QC!MCM isn't possible with K8 since two cores can't share one mem. controler...
RevH is the only QC design and the rivals for Kentsfield will be 4x4 systems(at least while there is no K8L in those configurations)

---

It's absolutely possible I'm afraid :stick: . I remember numerous S940 Server mobos which was equipped with 2 CPU sockets and 4 DIMM slots laid out only to first CPU. In this situation all memory data requested by CPU2 was redirected to IMC from CPU1 in same way like it's done when you have data in another CPU memory pool (NUMA).
But TBH I don't think AMD will made quad core RevG.

Quote:

---

Originally Posted by nn_step

umm what makes you think it is more cost effective to make something on a smaller process? usually inorder to make things on the smaller process you need to spend even more on the wafers which may or may not make the processors cheaper to produce. It all depends on the yeilds. :stick:

---

Your right, even the 90nm SOI of today makes the 65nm's of intels look bad. Since the 65nm SOI's with Girminanium, Cu, COMS, DSL, and afew other processes alone in normal production will blow intels 65nm processes away. All intel uses is COMS Cu 65nm. Nothing else so its sad and the reason why 90nm AMD's are still better then the 65nm intels.

Thats why a 90nm 35w amd upto 2.4ghz is much better at performance per watt then conroes upto a E6600. When they come out, because currently it goes upto a 2.2ghz at 35w now, but is still quite impressive consittering its 90nm and not 65nm. Just because the process is smaller doesn't automatticly make it better in anything. Theres more to it then just die size people.

You have to consitter the fab processes themselfs with the arc desine silly.

screw performance per watt, when amd can make processors that can clock to 5.5ghz, please get back to me. this isnt green peace systems, this is xtremesystems... by the way, low leakage transistors which constitute these 35w chips can barely clock over stock, so no thanks, ill take an x6800 that can clock to 4.5ghz on water and it can eat all the power it needs to get to 4.5ghz...

Quote:

---

ill take an x6800 that can clock to 4.5ghz on water and it can eat all the power it needs to get to 4.5ghz...

---

i hope you do , because you are not going to be happy watching the K8L when it comes out

so then i figure you already tried k8l?

It is mostly because K8L sips at power because it will have a slightly thicker gate oxides, which will slightly reduce switching speed, in exchange for virtually no current leakage and a brand new Record for performance Per watt. they could in theory reduce the gate Oxide thickness in exchange for more speed but that would go against their goal of performance per watt. With the side benefit of making it significantly more reliable than any processor made by Intel.

I honestly wouldn't mind if they made an enthusiast version that ignored heat issues and that just went balls out for speed. I have no problem running large fans or water cooling or for that matter phase change cooling, this is however impracticle so I guess we'll just have to make due with whatever they give us and see how far we can OC it.

Quote:

---

Originally Posted by nn_step

That is the thing when a new process comes out the yields are LESS.
infact the manufacturing cost of Early Yonah processors were similiar to the manufacturing cost of DEC's 21064 processors despite the fact that yonah is on 65nm and the 21064 was on a 3 layer 0.75µ process :stick:
Smaller ONLY MEANS CHEAPER IF THE PROCESS YIELDS ARE GOOD.
And just a little fact for you, ALL NEW PROCESSES HAVE BAD YIELDS. Hence AMD is switching to 65nm when it suits them BEST, not before

---

Quote:

---

Originally Posted by kemist

Bingo, when AMD made the move to 90nm with winchester they released 3000+, 3200+, and 3500+ processors because there was more volume there. While not all of the winchesters clocked great i had a very early 3200+ that hit 2.77ghz stable, (i believe) before the fx-57 even came out. Just because they arent releasing at high clock speeds doesnt mean the process sucks. It makes a lot more sense to have your high volume parts be on your smallest process because you can make a whole lot more of them. Also, since your early process will be spotty with parts that clock and some that clock low you can focus on your mature process for your highest speed low volume parts.

Now realistically, i think that with the early parts youre gonna see a lot of variation but you will definitely see a good amount hit in the 3.2-3.3 range, not amazing, but it barely happens with current silicon/process.

---

nn_step and kemist for president!

Quote:

---

Originally Posted by nn_step

It is mostly because K8L sips at power because it will have a slightly thicker gate oxides, which will slightly reduce switching speed, in exchange for virtually no current leakage ...

---

Are you on drugs or something ? :stick:

Quote:

---

Originally Posted by savantu

Are you on drugs or something ? :stick:

---

no.. that's what thicker oxide gates do. reduce leakage at the expense of switching speed ;)

Well if you have your electric bill paid for you then that's fine, but i don't, performance per watt FTW! :toast:

I think a certain hardcore fanbois should relax abit. They got the ball and running for an endzone that doesn´t exist. K8L is an EVOLUTIONARY upgrade of K8. The only performance enhancing thing in K8L is absicly its ability for the single cycle SSE like Core 2. However things like macro fusion and such is still out. Speed is also somewhat hyped upwards. Specially when looking on the number we see from AMD. Also K8L gonna fight 45nm Core 2 and a year after Core 3. :fact:

But some L3 cache, single cycle SSE and some improved branches dont give 40%. Nor will the power consumption drop to crazy low levels. Its the same 65nm process, and that same process AMD says 65W for smaller regular K8s. :fact:

This sounds just like Intel fanbois saying Core 2 would be 50% faster overall than K8. Reality gonna be a biatch when it hits :D

So evolutionary, not revolutionary. This aint the K9 or K10. Its a easy hotfix to K8 to make it last a year or 2 longer until they are ready with something vastly better. The K8L is more like the P4D when Intel had to do soemthing about the X2. So the big game aint this turn, but the next one.

Quote:

---

Originally Posted by Shintai

I think a certain hardcore fanbois should relax abit. They got the ball and running for an endzone that doesn´t exist. K8L is an EVOLUTIONARY upgrade of K8. The only performance enhancing thing in K8L is absicly its ability for the single cycle SSE like Core 2. However things like macro fusion and such is still out. Speed is also somewhat hyped upwards. Specially when looking on the number we see from AMD. Also K8L gonna fight 45nm Core 2 and a year after Core 3. :fact:

But some L3 cache, single cycle SSE and some improved branches dont give 40%. Nor will the power consumption drop to crazy low levels. Its the same 65nm process, and that same process AMD says 65W for smaller regular K8s. :fact:

This sounds just like Intel fanbois saying Core 2 would be 50% faster overall than K8. Reality gonna be a biatch when it hits :D

So evolutionary, not revolutionary. This aint the K9 or K10. Its a easy hotfix to K8 to make it last a year or 2 longer until they are ready with something vastly better. The K8L is more like the P4D when Intel had to do soemthing about the X2. So the big game aint this turn, but the next one.

---

For me K8L is revolution in evolution!
Applying your logic, C2d is a Piii upgrade!!
Go and try to add 1 useful transistor to complex design like K8 or Core then we can debate!

Quote:

---

Also K8L gonna fight 45nm Core 2 and a year after Core 3.

---

core 2 45nm is just a core size reduction with no improvements , K8L has many improvements

Quote:

---

Originally Posted by The Ghost

core 2 45nm is just a core size reduction with no improvements , K8L has many improvements

---

Doesnt matter much, if a 45nm Core 2 is 3.733Ghz and a 65nm K8L is 3.2Ghz. Aswell as the cost of production. Its factors that cant be left out either. So any clock to clock advantage gets eaten up. And currently Core 2 leads the clock to clock and clockspeed advantage. So process size matters alot.

Also 45nm Core 2 does have improvements in form of SSE4 parts and more cache, plus maybe a few minor other parts.

I wanna see K9/K10, K8L is just a bump to try and close the gab abit.