AMD K8L ES coming in December

Quote:

---

Originally Posted by brentpresley

The contact I mentioned a few posts above (DOD) has hinted to me that the 65nm ES chips they have do not OC well at all. Although he won't come out and just say it or put it in writing b/c of the NDAs.

NOW - I don't know how early on the chips he has are, or how much AMD has improved things since then. So keep that in mind.

But to expect that a dumb shrink (Rev G) of K8 is going to overtake Core 2 (20% faster IPC) is just wishful thinking.

Intel has had a YEAR to perfect their 65nm SHIPPING products. AMD is still perfecting 65nm FOR RELEASE. Simply put, Intel has had MUCH longer to work with the process technology and improve both the yields and the speed their chips can hit.

Additionally, it looks like we are seeing further improvements in Intel's 65nm process tech. Early release Core 2 chips had a lot of difficulty getting above 3.5GHz (unless you had a REALLY cherry-picked CPU). Now it seems that hitting 3.7-3.8GHz on decent air cooling is not that hard, if you have a motherboard that is up to the task of the high FSBs.

For Reg G(2, 3, whatever) to MATCH the performance of a 3.5GHz Core 2, AMD would have to get the clock speed up to 3.5GHz * 1.2 = 4.2GHz, assuming a 20% IPC lead for Core 2. I won't say it is impossible, but in a SHIPPING product it is going to take QUITE A WHILE.

No AMD's answer to the Core 2 performance crown is Rev. H. Period.

---

The performance in games is much closer than you would think Brent.

Perkam

Quote:

---

Originally Posted by brentpresley

The contact I mentioned a few posts above (DOD) has hinted to me that the 65nm ES chips they have do not OC well at all. Although he won't come out and just say it or put it in writing b/c of the NDAs.

NOW - I don't know how early on the chips he has are, or how much AMD has improved things since then. So keep that in mind.

But to expect that a dumb shrink (Rev G) of K8 is going to overtake Core 2 (20% faster IPC) is just wishful thinking.

Intel has had a YEAR to perfect their 65nm SHIPPING products. AMD is still perfecting 65nm FOR RELEASE. Simply put, Intel has had MUCH longer to work with the process technology and improve both the yields and the speed their chips can hit.

Additionally, it looks like we are seeing further improvements in Intel's 65nm process tech. Early release Core 2 chips had a lot of difficulty getting above 3.5GHz (unless you had a REALLY cherry-picked CPU). Now it seems that hitting 3.7-3.8GHz on decent air cooling is not that hard, if you have a motherboard that is up to the task of the high FSBs.

For Reg G(2, 3, whatever) to MATCH the performance of a 3.5GHz Core 2, AMD would have to get the clock speed up to 3.5GHz * 1.2 = 4.2GHz, assuming a 20% IPC lead for Core 2. I won't say it is impossible, but in a SHIPPING product it is going to take QUITE A WHILE.

No AMD's answer to the Core 2 performance crown is Rev. H. Period.

---

Well we all know tha AMD ES usually suck when compared to shipping cpus.This has been the regular case in the past.
So,i said if we compare out of the box products that will come Q1 07 and IF AMD can ramp up clock on Brisbanes,3.4GHz would be enough for C2D to match it in 90% of the cases.
Remember one thing: usual everyday peple don't know or care about OCing.They can read some PCMagazin or whatever and if they see the chips are on par and the prices are on par also,majority of folks will look at TOTAL cost of the platform and decide what to buy

this should be interesting :D

Quote:

---

Originally Posted by brentpresley

Possible, but my gut tells me that if Intel got even a HINT that AMD was going to try to shoot for the performance crown I bet we would "magically" see some 3.2 - 3.4GHz Core 2 Extreme editions pop up FS. I wouldn't be surprised if Intel has binned a few dozen thousand chips that clock well just for a situation like this.

Keep in mind, we are not dealing with the same Intel from the past 10-15 years that was pretty comfortable sitting on it's laurels and letting AMD have x% of the market.

Netburst was a mistake. A mistake that cost Intel the performance crown for about 3 years. Intel is PISSED about the loss of face AND the loss of marketshare. I don't think we will see them make a MAJOR screwup like Netburst again for a LONG LONG TIME.

Have no doubts, we are witnessing the first salvos in a war of attrition. Intel will take the lead, and them AMD will take it back. And Vice Versa for YEARS to come.

The ONLY winners are going to be us, the users.

Welcome to the NEW ERA of dirt cheap super-chips.

(now if we could just get this kind of war in the GPU arena we would be set).

---

Very true...we'll have Super computer FPU performance in 2008...on a $500 PC.

Definitely is interesting :)

Perkam

Quote:

---

Originally Posted by brentpresley

Have no doubts, we are witnessing the first salvos in a war of attrition. Intel will take the lead, and them AMD will take it back. And Vice Versa for YEARS to come.

The ONLY winners are going to be us, the users.

Welcome to the NEW ERA of dirt cheap super-chips.

(now if we could just get this kind of war in the GPU arena we would be set).

---

Well said, it's a beautiful time for us end users, and pc enthusiasts. :p:

Quote:

---

Originally Posted by perkam

AMD and Intel have both confirmed that their nex gen native QC designs are launching Q3 '07...

All this talk about AMD "will be behind" is crap...it'll on par with the year...Conroe allowed Intel's crown to remain for 1 year, while AMD's lasted 3.

Perkam

---

1. Intel has the lead over AMD in x86 CPU for ages.
2. Netburst was a tragic mistake that Intel used for the past 6 years.
3. Intel's product development cycle is much shorter now with their new strategy. There's only a slim chance for AMD to keep up.
4. The best part of AMD's business will now be its GPU, aside from its digital TV and cellphone coprocessors.

What do we know today?
1. Server K8l Deerhound Opteron will be announced in Q2 2007
2. FX socket F+ (with HT 3.0 support) for 4x4 (in fact it's the same Opteron, remarked) and X4 am2+ (also with HT 3.0 support) - in Q3 2007
3. Dell will buy a great quantity of AMD procs :)
4. In several months AMD will show us the system with K8L and proove its high performance. Maybe this will be octal-core 4x4 :)
5. 65nm AMD procs can operate at clocks not less than C2D B2, B3
6. There is no quad-core K8 in AMD plans at all, quad-core is only K8L

brentpresley, look at overclock abilities of 90nm Dothan and K8 - frequency limit (temperature 20C) is just the same
why OC limit for 65nm short-stage intel and amd cpus must differ? (sorry for bad engl. :) )

Quote:

---

Originally Posted by vitaminc

3. Intel's product development cycle is much shorter now with their new strategy. There's only a slim chance for AMD to keep up.

---

I wouldn't bet the farm on that ;)

http://www.xtremesystems.org/forums/...d.php?t=118091

Quote:

---

Originally Posted by MAS

brentpresley, look at overclock abilities of 90nm Dothan and K8 - frequency limit (temperature 20C) is just the same
why OC limit for 65nm short-stage intel and amd cpus must differ? (sorry for bad engl. :) )

---

So it's speculation, not an fact.

The first few batch of 65nm AM2 CPU probably won't clock so well. It's a new process node for AMD afterall.

Quote:

---

Originally Posted by LowRun

I wouldn't bet the farm on that ;)

http://www.xtremesystems.org/forums/...d.php?t=118091

---

GP GPU has nothing to do with CPU. Sure GPUs (ATI > nVidia) has :banana::banana::banana::banana:loads of FLOPs, but all of the GPU will just choke so hard when running x86 style instructions in x86 CPU thermal envelopes.

Hopefully we could use a 200W PSU for a CrossFire or SLi high end gaming rig someday.

Quote:

---

Originally Posted by vitaminc

So it's speculation, not an fact.

The first few batch of 65nm AM2 CPU probably won't clock so well. It's a new process node for AMD afterall.

---

True, however the first few batches are about 1 Month of Time. After that you have to deal with AMD's Rapidly evolving process, realtime process changes combined with a group of highly skilled and motivated Individuals working as a team. At the same time for the first 65m chips (provided that are K8) From AMD would clock comparative to Yohan (Conroe's father) and shortly after it'll start to clock faster and faster

Quote:

---

Originally Posted by nn_step

True, however the first few batches are about 1 Month of Time. After that you have to deal with AMD's Rapidly evolving process, realtime process changes combined with a group of highly skilled and motivated Individuals working as a team. At the same time for the first 65m chips (provided that are K8) From AMD would clock comparative to Yohan (Conroe's father) and shortly after it'll start to clock faster and faster

---

We could debate AMD's APM vs. Intel's Copy Exact strategy all day long, but the fact is that yields defect density will only go down with time.

You can never state the overclockability of a chip as a fact before you have them on hands with actual data. Everything now is speculations under NDA smoke screens.

Quote:

---

AMD's chips made EARLY on in a process shrink (90nm, 65nm, etc.) don't overclock very well. They continue to improve the process and release faster and faster chips. It just takes some time, but they eventually get VERY good with working with a particular process level.

---

some times i think that amd does rthis stuff on purpose , just look at the mhz that the 4x4 cpu's are going to be , amd has no other cpu running that fast

Quote:

---

(they have the luxury of having an ENTIRE EXPERIMENTAL FAB - D1D in Oregon),

---

i read some where that amd has , as you would say it , a experimental lab at fab 30 and fab 36

Quote:

---

Originally Posted by The Ghost

some times i think that amd does rthis stuff on purpose , just look at the mhz that the 4x4 cpu's are going to be , amd has no other cpu running that fast


i read some where that amd has , as you would say it , a experimental lab at fab 30 and fab 36

---

Actually ALL AMD fabs include the ability to experiment and improve process performance and yeilds

Quote:

---

Originally Posted by The Ghost

i read some where that amd has , as you would say it , a experimental lab at fab 30 and fab 36

---

They don't. They jointly develope with IBM. They do have a "test" line for future processes (not a whole dedicated fab like Intel).

Quote:

---

Originally Posted by nn_step

Actually ALL AMD fabs include the ability to experiment and improve process performance and yeilds

---

The improve process performance/yields are result of APM (software fab control and other magics). Has nothing to do with their process development for 45nm or 32nm.

Quote:

---

Originally Posted by vitaminc

GP GPU has nothing to do with CPU. Sure GPUs (ATI > nVidia) has :banana::banana::banana::banana:loads of FLOPs, but all of the GPU will just choke so hard when running x86 style instructions in x86 CPU thermal envelopes.

Hopefully we could use a 200W PSU for a CrossFire or SLi high end gaming rig someday.

---

Looks like you didn't even read the article :rolleyes:

Quote:

---

Originally Posted by LowRun

Looks like you didn't even read the article :rolleyes:

---

skim through it. nothing interesting that i havent heard of, except it does not address a lot of important concerns such as power envelope, process integration, etc.

Quote:

---

Originally Posted by perkam

The performance in games is much closer than you would think Brent.

Perkam

---

So will the performance in wordpad. Just goes to show that games that are out now do not utilize CPU's well. Though I think they will in the future, what with physics simulation and all.

Also Perkam,i totaly forgot about this : Did you ever thought that S.Meyer of AMD might be talking about 4x4 as non-native quad core solution?I did,and it goes well in line with what H.Richards said at Digitimes interviews when asked about that.
So i still am 99.9% sure there never will be MCM K8 based non native QC.As i said it previously,how is it possible that both chips share one mem. controler(be it on first or second core) and still use the same socket?OTOH,I can see that two sockets can make non native=almost MCM like solution= and that is what AMD publicly stated as their plan to combat Kents(which may fair well against since there could be some serious cache trashing -during heavy multitasking- in Kentsfield's shared cache structure and FSB bottlenecking,no matter if it is not full 8MB of shared L2,but 2x4MB)

Slightly OT: As for Daily tech credibility lately,look at this one :http://images.dailytech.com/nimage/2...07_roadmap.png
Tell me they are serious,since they have been wrong with their "roadmaps" (custom made ? :p: ) when they spoke about "roadmaps" showing AMD only getting to X2 5600+ in Q3'07...

Quote:

---

Originally Posted by vitaminc

skim through it. nothing interesting that i havent heard of, except it does not address a lot of important concerns such as power envelope, process integration, etc.

---

Must be your way of saying you've only read the title otherwise you wouldn't say something like

Quote:

---

GP GPU has nothing to do with CPU. Sure GPUs (ATI > nVidia) has loads of FLOPs, but all of the GPU will just choke so hard when running x86 style instructions in x86 CPU thermal envelopes.

---

Even without reading the article i can't see why one would want to run x86 style instructions on a GP-GPU that sits on a die with 4 x86 cores by it's side.
They won't put it there for you to run x86 instructions :slap:

Quote:

---

Originally Posted by informal

Also Perkam,i totaly forgot about this : Did you ever thought that S.Meyer of AMD might be talking about 4x4 as non-native quad core solution?I did,and it goes well in line with what H.Richards said at Digitimes interviews when asked about that.
So i still am 99.9% sure there never will be MCM K8 based non native QC.As i said it previously,how is it possible that both chips share one mem. controler(be it on first or second core) and still use the same socket?OTOH,I can see that two sockets can make non native=almost MCM like solution= and that is what AMD publicly stated as their plan to combat Kents(which may fair well against since there could be some serious cache trashing -during heavy multitasking- in Kentsfield's shared cache structure and FSB bottlenecking,no matter if it is not full 8MB of shared L2,but 2x4MB)

---

According to HKEPC (Hong Kong based site), Yorksfield will be Intel's 45nm single die quad core with Conroe core in 2H07.

FSB problem still presists, but it's not as bad as people made it out to be.

Quote:

---

Originally Posted by informal

Slightly OT: As for Daily tech credibility lately,look at this one :http://images.dailytech.com/nimage/2...07_roadmap.png
Tell me they are serious,since they have been wrong with their "roadmaps" (custom made ? :p: ) when they spoke about "roadmaps" showing AMD only getting to X2 5600+ in Q3'07...

---

Well, if you think about it, AMD has always been focusing on architecture improvement since the K6 days, thus the clocking of X2/X4 are not that big of a deal IMO. They could always do it Intel P4 style and clock it really high, but that's not "green" computing. :p

Quote:

---

Originally Posted by LowRun

Must be your way of saying you've only read the title otherwise you wouldn't say something like



Even without reading the article i can't see why one would want to run x86 style instructions on a GP-GPU that sits on a die with 4 x86 cores by it's side.
They won't put it there for you to run x86 instructions :slap:

---

So they only taught you reading but not thinking in school?

"X86 style" instructions meaning codes that's not highly parallel like graphics.

1. You don't think that all of the current HPC softwares will be recompiled away from the current x86/SPARC/PPC codes and translate into DX9/DX10 based GPU code, do you?
2. Running a couple X1950XTX 512M cards with 200W power dissipation each under load just all of the sudden makes a whole lot of sense, especially with AMD's compaign of green computing.

Brent is right about 65nm

Even AMD says the same, they silently add new tech bits to their sauce. Early on in a shrink the process is almost exactly the same, and basically results in only a power and die size shrink. Further out when they begin adding the new tech is when the clocks start to ramp consistently. (i did have an early winchester that clocked to 2.77ghz which was great for the time, there will always be a few exceptions)

Quote:

---

Originally Posted by vitaminc

According to HKEPC (Hong Kong based site), Yorksfield will be Intel's 45nm single die quad core with Conroe core in 2H07.

FSB problem still presists, but it's not as bad as people made it out to be.

---

I was refering to Novembre release of Kentsfield and the AMD's response to that(in my opinion,it's not MCM K8 like hybrid-gotta love that term:p:).
As for Yorksfield,i think it will be somewhat better than Kents ,BUT if it has no IPC improvements,i don't know how well it will fare against K8L,especially in HPC segment(in all honesty I believe that server variant of K8L will be better,and by a wider margin)

Quote:

---

Originally Posted by vitaminc

Well, if you think about it, AMD has always been focusing on architecture improvement since the K6 days, thus the clocking of X2/X4 are not that big of a deal IMO. They could always do it Intel P4 style and clock it really high, but that's not "green" computing. :p

---

I was speaking of Brisbanes arrival dates in that DailyTech "roadmap".It's said that they will arrive in Q1 07,but other sources (HKPEC for instance) are saying Decembre.As for clock speeds on that slide,they are in line with what we know by now,my only objection was the fact that dailytech wasn't upto the mark,at least when roadmaps for AMD are in question