Fuddo: AMD 45nm K10.5 scheduled to launch in 1H 2009

[endyen]

I'm tired of AMD promises.
When they say "H2 2008" we should expect "the last few weeks of December 2008 with token quantity and maybe defective parts". They're never honest.

Good plan! That way when they arrive in october 08, with lots of defects, we will still have something to be happy about.

 

[Halley]

I'm so not surprised. 45NM in 2008 was impossible for AMD, I'm quite fed up with their lies and deceit. Is it so hard to actually give a more realistic timeline release? Instead of empty promises which end up with delays and hurts their image even more.

I was expecting 45nm from AMD in 2009, more likely late 1H or early/mid 2H 2009.

I'm tired of AMD promises.
When they say "H2 2008" we should expect "the last few weeks of December 2008 with token quantity and maybe defective parts". They're never honest.

 

[epsilon84]

It's an improvement of xbit, but they bothe have the errata patch enabled, and use the 1.8 HTT, L3.
What's the considered loss using the patch?

No they don't, performance would be even worse with the TLB patch on. Almost every single Phenom review has been done with the patch disabled, in fact most were done right after the launch, before the patch was even available.

The 1.8GHz IMC/L3 is the design spec, what do you expect them to do, overclock it beyond specs so it pulls it closer to a C2Q?

 

[enewmen]

is the 10.5 Hekka at least a minor architecture change to the k10 (b3)?
Or NO architecture change at all?
Or is the 45nm process just for yelds ?

From what little I know, the native quad core was ambitious, even for Intel. So I hope the see the fruits appear, SOMETIME..

 

[jimmysmitty]

No, what Intel said was that soi did little for them.
Soi is only good if your problem stems from capacitance on the handle side of the wafer. Intel uses a lower v-core, so that capacitance is much lower.

Doesn't that strike you as strange? That they will release 45nm on SOI but soon after will have 45nm HK/MG?

First off, you seem to think that HK/MG will be without soi. I think not. Soitek has only just begun a new wafer manufacturing plant, to make soi wafers for AMD. I doubt they would do that if AMD is going to stop using soi wafers before that plant is even finished. On the other hand, it does seem strange that they will try to do 45nm without NiSi. I'm guessing this is the advantage of immersion lithography. The HK is not that big a thing as they will be using a median high K oxide, by icreasing the nitride level in the gate oxide.

I never stated Intel was going to use it. I simply stated that they probably researched the idea and in the end didn't look like it could go beyond 45nm without any complications. As for AMD, of course they will still use silicon but still use SOI? SOI is the process. mixing processes might be tricky so once IBM comes out with a HK/MG process I doubt AMD will just add SOI on top of it as thats where they get their technology.

Either way I doubt its going to "revolutionize" the Phenom and turn it into a Core2 killer. Maybe a little boost to match Kentsfeild and maybe even Yorkfeild. But nothing amazing.

No they don't, performance would be even worse with the TLB patch on. Almost every single Phenom review has been done with the patch disabled, in fact most were done right after the launch, before the patch was even available.

The 1.8GHz IMC/L3 is the design spec, what do you expect them to do, overclock it beyond specs so it pulls it closer to a C2Q?

I agree 100%. Yes you can OC the NB and the CPU to get close to the Q6600 but the main question is what will it do vs the Q6600 stock. Heck when you just OC the CPU(like you would a Q6600) how does it perform compared to a Q6600 @ stock?

So its great that OC'ing the NB will get it up to par with a Q6600 but then again if you OC a Q6600 it will pull away as well.

 

[LukeBird]

I agree 100%. Yes you can OC the NB and the CPU to get close to the Q6600 but the main question is what will it do vs the Q6600 stock. Heck when you just OC the CPU(like you would a Q6600) how does it perform compared to a Q6600 @ stock?

So its great that OC'ing the NB will get it up to par with a Q6600 but then again if you OC a Q6600 it will pull away as well.

One thing with your comment above, when you OC the Q6600 you are OCing the memory as well (as you're obviously going to have to bump the FSB), which in effect is probably the same as bumping the NB/IMC on a Phenom.
So OCing just the Phenom core is not the same as OCing a Q6600. To get like for like, you'd need an EE with an unlocked multi, to just OC the core speed
Although I completely agree with you a Q6600 should OC far better than a Phenom!

 

[evongugg]

Phenom = Phailnom

 

[jimmysmitty]

One thing with your comment above, when you OC the Q6600 you are OCing the memory as well (as you're obviously going to have to bump the FSB), which in effect is probably the same as bumping the NB/IMC on a Phenom.
So OCing just the Phenom core is not the same as OCing a Q6600. To get like for like, you'd need an EE with an unlocked multi, to just OC the core speed
Although I completely agree with you a Q6600 should OC far better than a Phenom!

Not always. You can OC the CPU FSB and leave the memory at the same speed, i.e. have a Q6600 running 3GHz 1333FSB and have the memory running 1066FSB. Although you wont get as good of a gain if you don't at least match the memory to the CPUs FSB.

Also when you OC the NB it has no effect on the core speed of the Phenom, or so it seems, but only the memory transfer performance. Kinda makes me cringe that the Phenom has 3 things you can OC. CPU, NB and IMC. Sounds complimicated to me

But in the end at the stock NB speeds Phenom doesn't keep up with a Q6600. OC it to what mathos, I think, has (2.6GHz CPU and 2.4GHz NB/IMC) and it keeps up. But just OC'ing the CPU to 2.4GHz seems not to help much. So I guess the lower than ES NB speed on the retail hold it back and the regular people and even the OC machines from the OEMs(probably if they OC only the CPU not the NB) it wont help again.

 

[bfellow]

The Q9300's are back at Tigerdirect.com for overpriced $300. This is the 45nm successor of the Q6600 and direct competitor to Phenoms now.

http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=3584440

 

[TechnologyCoordinator]

I'm tired of AMD promises.
When they say "H2 2008" we should expect "the last few weeks of December 2008 with token quantity and maybe defective parts". They're never honest.

I agree Halley. I'm still waiting for the processors that were launched in November to show up. I think those generous times are just to keep investors happy.

While some people here won't want to acknoledge it, AMD's stock price has been in the tank since K10's launch and often companies will do stuff just to keep their investors happy.

 

[spoonboy]

AMD will have to wait to 2nd generation of 45nm to actually improve K10's performance by a substantial margin. 45nm SOI simply won't work well for them.

I would stay away from DDR3 for now , and use them as Nehalem hits.

Your right about DDR3, it seems like the biggest waste of money e-pen1s fodder ever. I saw some test once, like ddr800 4-4-4-12 was either a tiny fraction slower or just as fast as some super expensive ddr3 kit in actual applications and games. 8-8-8-24, what use is that?

 

[Mathos]

Your right about DDR3, it seems like the biggest waste of money e-pen1s fodder ever. I saw some test once, like ddr800 4-4-4-12 was either a tiny fraction slower or just as fast as some super expensive ddr3 kit in actual applications and games. 8-8-8-24, what use is that?

Well thats because the actual memory performance hasn't really increased since the jump to DDR from regular SDRam. Since DDR1 the clocks have gone up, but the latencies are outright equivalent between DDR400 2-2-2-6, DDR2-800 4-4-4-12, DDR3-1600 8-8-8-24. So other than bandwidth and lower voltage and in turn power consumption, actual memory performance hasn't improved in a loooong time. Now if they managed to get faster speed with lower latency then that would be something to go for.

@Jimmy Neh, NB/IMC are both on the same clock/Power plane. But it's meant to be OCed otherwise it wouldn't be completely unlocked on the BE where as it upward locked on the regular processors. I'm just curious what they're gonna do with the FX-82 on the AM2+ platform. According to their charts it's slated for 2.6ghz+, I'm just wondering if the difference between the FX and the BE is going to be a higher clockable IMC for more performance. Oh, and they are going to release 2 Socket 1207+ Phenom FX chips, Phenom FX 91 slated as 2.3-2.6ghz, and the fx-92 for 2.6+, so I think 4x4 will make a come back to try and compete against SkullTrail.

 

[endyen]

No they don't, performance would be even worse with the TLB patch on. Almost every single Phenom review has been done with the patch disabled, in fact most were done right after the launch, before the patch was even available.

The 1.8GHz IMC/L3 is the design spec, what do you expect them to do, overclock it beyond specs so it pulls it closer to a C2Q?

Only benchmarks done with the original ES. All benchmarks done with a BE have been done with the patch enabled, at the request of AMD.

The 1.8GHz IMC/L3 is the design spec, what do you expect them to do, overclock it beyond specs so it pulls it closer to a C2Q?

No, the design spec called for 2ghz, the 1.8 was a last minute patch for, you guessed it the TLB eratta.

 

[endyen]

I never stated Intel was going to use it. I simply stated that they probably researched the idea and in the end didn't look like it could go beyond 45nm without any complications. As for AMD, of course they will still use silicon but still use SOI? SOI is the process. mixing processes might be tricky so once IBM comes out with a HK/MG process I doubt AMD will just add SOI on top of it as thats where they get their technology.

Either way I doubt its going to "revolutionize" the Phenom and turn it into a Core2 killer. Maybe a little boost to match Kentsfeild and maybe even Yorkfeild. But nothing amazing.

No, I told you why they didn't use it, Intel just didn't need it.

SOI is the process

Say what? Soi is a type of wafer. Amd buys wafers with SOI from Soitec. That's the way they come.
Let me put it another way. AMD does not grow the soi on thier wafers. Soitec is a wafer manufacturer. They grow the Soi on the wafers, then sell them to AMD, Sony, Chartered, Hitachi and IBM, with the insulator already in place.
As to phenom being a core2 killer, not likely ever. Once people start coding for the new marchitecture, the phenoms will do much better. Facts are facts though, and Intel's process is so much better than everyone elses, that they will have speed on thier side for a long time.

Here's a nice piece about making soi wafers http://www.forbes.com/home/free_forbes/2005/0919/072.html

 

[epsilon84]

Only benchmarks done with the original ES. All benchmarks done with a BE have been done with the patch enabled, at the request of AMD.

The 1.8GHz IMC/L3 is the design spec, what do you expect them to do, overclock it beyond specs so it pulls it closer to a C2Q?

No, the design spec called for 2ghz, the 1.8 was a last minute patch for, you guessed it the TLB eratta.

No, I think you're just grasping at straws and don't want to accept reality. AMD made no such requests for reviewers. The 9600 BE being some ~13.5% slower than a Q6600 is totally plausible and falls in line with all other 9600 reviews showing it to be 10 - 15% slower than a Q6600 overall. Use some logic. The patch has a 10%+ performance hit, if the 9600 BE was reviewed with the patch on it would be closer to 25% slower than the Q6600. On the other hand, you're basically saying the 9600 BE should be neck and neck with the Q6600 with the patch disabled. Not. Gonna. Happen. 😉

1.8GHz *is* the design spec for retail chips, 2GHz is what was used on the ES in reviews. http://techreport.com/articles.x/13741/1

That original Phenom review overstated the Phenom 9600's performance for two reasons. The most obvious, of course, is the fact that we didn't have the TLB erratum patch applied. The other is that the north bridge clock on our Phenom engineering sample was running at 2.0GHz. AMD told us that was the correct north bridge speed, but our experience with production Phenom 9600 chips has proven otherwise; the correct clock is 1.8GHz. The north bridge clock is critical to performance in this CPU architecture because the L3 cache runs at the speed of this clock.

 

[LukeBird]

Not always. You can OC the CPU FSB and leave the memory at the same speed, i.e. have a Q6600 running 3GHz 1333FSB and have the memory running 1066FSB. Although you wont get as good of a gain if you don't at least match the memory to the CPUs FSB.

Also when you OC the NB it has no effect on the core speed of the Phenom, or so it seems, but only the memory transfer performance. Kinda makes me cringe that the Phenom has 3 things you can OC. CPU, NB and IMC. Sounds complimicated to me

But in the end at the stock NB speeds Phenom doesn't keep up with a Q6600. OC it to what mathos, I think, has (2.6GHz CPU and 2.4GHz NB/IMC) and it keeps up. But just OC'ing the CPU to 2.4GHz seems not to help much. So I guess the lower than ES NB speed on the retail hold it back and the regular people and even the OC machines from the OEMs(probably if they OC only the CPU not the NB) it wont help again.

Ok, my apologies! I thought bumping the FSB dragged the memory speed up with it regardless!
You're right though, it does leave you with technically more to OC!

 

[Mathos]

1.8GHz *is* the design spec for retail chips, 2GHz is what was used on the ES in reviews. http://techreport.com/articles.x/13741/1

Actually Barcelona/Agena spec was adjustable NB/IMC on a seperate power plane, that would run at or within 200Mhz of the cores clock speed. That was done to prevent the uneven divider issue with memory that had plagued the K8,as well as allowing C&Q 2.0 to throttle the NB/imc while the system was idle for better power savings. I don't remember exactly which review page I read that off of, or the name of the article, which is why I haven't linked it yet. With limited time on my hands, trying to find an old article without remembering the name is a pain. At this point i can't remember if it was THG, Anands, or one of the others.

 

[jimmysmitty]

Actually Barcelona/Agena spec was adjustable NB/IMC on a seperate power plane, that would run at or within 200Mhz of the cores clock speed. That was done to prevent the uneven divider issue with memory that had plagued the K8,as well as allowing C&Q 2.0 to throttle the NB/imc while the system was idle for better power savings. I don't remember exactly which review page I read that off of, or the name of the article, which is why I haven't linked it yet. With limited time on my hands, trying to find an old article without remembering the name is a pain. At this point i can't remember if it was THG, Anands, or one of the others.

Isn't C&Q 2.0 only available for AM2+?

 

[uguv]

How many boards that support Q6600 will also support Nehalem this time next year? How many that will support Penryn will? Many Intel enthusiasts here are used to switching out boards for their Intel builds every time they upgrade their CPU, while at the same time kvetching about the motherboard companies lack of support for 690 upgrades to Phenom, but when all is said and done; a 780G board and a B3 is the perfect budget upgrade to quad core this spring because it will be supported when Deneb arrives.

The answer is none of course, but then again even some 975X boards can support Penryn. Those boards have lasted quite a while huh? How many AM2 boards will support 45nm AMD procs?

Sure, right now an AM2+ board will (in theory) last you a couple of upgrade cycles, assuming the mobo manufacturers release bios updates to let them run the chips in the future, but Intel boards generally last a few upgrade cycles too. Nehalem will require a new motherboard but it's foolish to say that Intel CPU upgrades always require a new motherboard.

 

[jimmysmitty]

The answer is none of course, but then again even some 975X boards can support Penryn. Those boards have lasted quite a while huh? How many AM2 boards will support 45nm AMD procs?

Sure, right now an AM2+ board will (in theory) last you a couple of upgrade cycles, assuming the mobo manufacturers release bios updates to let them run the chips in the future, but Intel boards generally last a few upgrade cycles too. Nehalem will require a new motherboard but it's foolish to say that Intel CPU upgrades always require a new motherboard.

You should have asked how many S939 boards support Phenom, hehe. S939 is just a broken promise. Intel has had only 2 sockets in the past what 5-6 years? Not too bad.

 

[Amiga500]

You should have asked how many S939 boards support Phenom, hehe. S939 is just a broken promise. Intel has had only 2 sockets in the past what 5-6 years? Not too bad.

How many of the chipsets allow you a clear upgrade path?

 

[jimmysmitty]

How many of the chipsets allow you a clear upgrade path?

Well considering on my PGA478 845PE, which was one of the first in that series, chipset I could run any Pentium 4 from the first one available for it all the way up to the Pentium 4 EE on the 90nm process with 1MB cache, a dam good upgrade path for 4 years.

For LGA775, depending on the mobo maker, there are some P965 chipset mobos that will run Yorkfeilds. Thats on Asus but they are very reliable in terms of mobo updates. Now this applies to AMD as well. This has happened with Phenom and AM2 as not every mobo maker added updates for support.

But my main point is AMD was touting clear upgrade path and such back in S939 and then bam. Threw out AM2, which was supposed to be faster but wasn't. Then they just left S939 in the dust. Kindad like their 4x4 platform they just decided to throw away.

 

[Amiga500]

Then you've got luck. I've come across numerous **** chipset limitations with Intel.


At least with AMD, a S939 mobo will support an S939 CPU, no further questions needed. Unfortunately they moved to DDR2, and abandoned the old format too quickly. However, with the IMC, there is no question of having DDR1 and DDR2 chips so AMD made the right call in having a new socket to avoid confusion. Their error was not supporting the old path further.



Unfortunately AM2 to Phenom has become fudged and I'm not sure if things work as well. :-(

 

[epsilon84]

Then you've got luck. I've come across numerous **** chipset limitations with Intel.

Its a mixed bag, I agree. Earlier S775 chipsets during the P4/PD era (with the exception of the 945) can't run C2Ds.

Things have been going quite well since the P965, many are capable of running Wolfdales and Penryns with a BIOS update and its coming on 2 years since the chipset was launched.

 

[enigma067]

That would have been the best scenario. If a 2.2 GHz Phenom performed the same as a 2.4 GHz Q6600. Whole new world then. Obviously AMD can't push the clocks as high as Intel (especially now at 45nm) so they really needed that advantage and didn't get it.

You're comparing a Multi Chip Module with a quad core processor? That's laughable!
And you failed miserably in supplying us with links that prove your statement!

I seriously doubt that a Multi Chip Module that has four 64 bit FPUs and suffers from a memory bandwidth bottleneck
can out muscle a Quad Core Phenom that has four 128 bit FPUs and does not suffer from a memory bandwidth bottleneck.

The Q6600 is nothing more then regurgitated technology! I shouldn't have to remind you how OLD the 8086 architecture is? It came out in the late 1970's!

BTW... a Barcelona owner reported some time ago on Newegg.com that he is over clocking the Barcelona to 4.1 Ghz air cooled and 4.7 Ghz. liquid cooled. Yes AMD can push their clocks higher! Check the forums for over clocking Barcelona.