AMD 65nm and 45nm

[Viditor]

Originally posted by: zsdersw
That's not putting words in your mouth.

I don't think you're a "rabid AMD fanboy".

Mate, if you don't get it...you don't get it. I'm glad to hear that what you post is not what you actually think. 🙂

 

[zsdersw]

Not thinking you said something isn't putting words in your mouth.

I definitely post what I think. And while I don't think you're a "rabid AMD fanboy".. I'd definitely consider you a leading AMD advocate. There's a big difference.

 

[dexvx]

K8L is a rather misleading title. Given the name, you'd expect it to be a K8 derivative, when in facts its pretty much a new architecture.

 

[Regs]

Originally posted by: myocardia
I believe that AMD thought that the C2D was going to along the lines of Intel's last two "launches", and not be what it's turned out to be, and so they just sat on their laurels, and raked in the profits.

I don't think AMD had that thought. Especially when even here at the forum people were predicting a Pentium M type processor for the Desktop. It was only a matter of time. I saw it coming as soon as the Pentium M entered the arena. The only thing I was hoping for was something new from AMD and that never came. Who knows what AMD's problem is or what they're trying to keep secrete on.

Maybe the direction they are trying to head in is prohibiting them from manufacturing a better single threaded processor. That's my only guess.

 

[Some1ne]

because a x2 3800 that does 3ghz on average would be a tough deal to pass up.

Why? Even the E6300/E6400's can generally do 3 GHz on average, and quite often more.

a now we can fianlly which architecture can clock the highest.

I think that issue is kind of moot until AMD revises its architecture. You'd have to put a K8 at something like 3.0 GHz (or more) just to match the performance of an E6600 at stock. Factor in the fact that most E6600's will overclock to at least 3.2 GHz, and some of the luckier ones go up to 3.6 GHz, and basically you'd need to see overclocks on the 65 nm K8 parts to speeds in excess of 4.0 GHz, just in order for an overclocked K8 to be competitive with a Core 2. To beat it hands-down, a K8 chip would probably have to push 4.5 to 5.0 GHz...and it seems unlikely that a process shrink will add that much headroom to it.

 

[Zstream]

Originally posted by: Some1ne

because a x2 3800 that does 3ghz on average would be a tough deal to pass up.

Why? Even the E6300/E6400's can generally do 3 GHz on average, and quite often more.

a now we can fianlly which architecture can clock the highest.

I think that issue is kind of moot until AMD revises its architecture. You'd have to put a K8 at something like 3.0 GHz (or more) just to match the performance of an E6600 at stock. Factor in the fact that most E6600's will overclock to at least 3.2 GHz, and some of the luckier ones go up to 3.6 GHz, and basically you'd need to see overclocks on the 65 nm K8 parts to speeds in excess of 4.0 GHz, just in order for an overclocked K8 to be competitive with a Core 2. To beat it hands-down, a K8 chip would probably have to push 4.5 to 5.0 GHz...and it seems unlikely that a process shrink will add that much headroom to it.

I think you are forgetting that the K8L is a much better CPU then we currently have. So comparing it with a CPU that we have now is dumb.

Anyone know how long AM2 will last?

 

[PingSpike]

Originally posted by: DrMrLordX

Originally posted by: zsdersw
Is this not what was expected?

I was under the impression that we wouldn't see 65nm parts from AMD until Nov/Dec. Instead we get them in October. That's pretty good for AMD; they really needed to kick things into gear.

Interestingly enough, this really does help their deal with Dell. A mature 65nm process means more procs per die which helps them to feed Dell's hunger for chips. 45nm will help even more, but that won't come until a ways down the road.

Agreed. Despite the fact that conroe is clearly the better chip, if they can just get the capacity Dell wants they should be able to at least make some decent money making mid range and low end products. AMD doesn't have the cash reserves intel does, so its important they keep selling chips somewhere if they want to remain a strong player. It would at least buy them the time they need to developer a stronger competitor. We'll see what happens!

 

[Some1ne]

I think you are forgetting that the K8L is a much better CPU then we currently have. So comparing it with a CPU that we have now is dumb.

Not really, because the topic of this thread is the 65 nm chips (and the secific comment I was replying to was about having a 3800+ X2 running at 3.0 GHz...which is definitely a K8 model), which *are not* K8L initially (K8L is slated for mid/late 2007, if I remember it correctly). K8L may very well be competitive w/ Core 2 (and for AMD's sake, it had really better be), but a 65 nm K8 chip will not be, no matter how much overclocking ceiling it has.

 

[Viditor]

Originally posted by: Some1ne

I think you are forgetting that the K8L is a much better CPU then we currently have. So comparing it with a CPU that we have now is dumb.

Not really, because the topic of this thread is the 65 nm chips (and the secific comment I was replying to was about having a 3800+ X2 running at 3.0 GHz...which is definitely a K8 model), which *are not* K8L initially (K8L is slated for mid/late 2007, if I remember it correctly). K8L may very well be competitive w/ Core 2 (and for AMD's sake, it had really better be), but a 65 nm K8 chip will not be, no matter how much overclocking ceiling it has.

Just to clear up, K8L is due out in Q2 (according to the article and everything AMD has said). Of course the end of Q2 IS mid 2007...

 

[Hyperlite]

core names?

 

[dexvx]

Not much info, but since K8L is 4 issue, similar to Core2, its IPC should be as good as or better than Core2.

 

[Furen]

Originally posted by: dexvx
Not much info, but since K8L is 4 issue, similar to Core2, its IPC should be as good as or better than Core2.

I dont think K8L is actually 4-issue... not that it makes a huge difference, since I doubt even Conroe can hit more than 2 ops per clock.

 

[Keysplayr]

Originally posted by: Viditor

Originally posted by: myocardia
I don't see how moving to the 65nm process will be that much of an improvement for the Athlons, even if they make a much smoother transition than Intel did, which is unlikely, IMO. I believe that AMD thought that the C2D was going to along the lines of Intel's last two "launches", and not be what it's turned out to be, and so they just sat on their laurels, and raked in the profits.

Of course, I'm hoping that I'm wrong (how often do you hear anyone say that?), and AMD will at least catch C2D in performance, because the only time we, as consumers, benefit, is when both companies have similar performing products. And that applies to anything, not just processors.

Some questions...
1. Don't you think that delaying volume production for 6 months has allowed AMD to improve their 65nm process over Intel's?
2. As Intel was able to drastically reduce their power usage and increase clockspeed at the same time when they wen to 65nm, do you think AMD won't be able to do the same?
3. AMD's 90nm launch was FAR smoother than Intel's (after delaying for the same length of time)...is there a reason that you think this time will be different?

And the grandaddy of questions: What does this have to do with Intel?

 

[Hard Ball]

Originally posted by: Furen

Originally posted by: dexvx
Not much info, but since K8L is 4 issue, similar to Core2, its IPC should be as good as or better than Core2.

I dont think K8L is actually 4-issue... not that it makes a huge difference, since I doubt even Conroe can hit more than 2 ops per clock.

They are not, integer performance would be quite comparable clock for clock with K8, moderately improved through larger BTB, better OoO execution, out of order loads, and a few other smaller improvements.

Both FP and Vector will will be double issue; as opposed to Conroe only being 128bit in vector datapath. So you can expect quite a bit of improvement in terms of FP, at least through FADD and FMUL units. FMISC is not clearly explained at this point; but many speculate that FP ld/st will also be of comparable width, in order to take full advantage of the other execution units in the FP datapath.

 

[buzzsaw13]

Originally posted by: munky
I don't think they'll be making any 65nm s939 chips. Which is too bad, because a x2 3800 that does 3ghz on average would be a tough deal to pass up.

I can always hope and pray! They did make Venice S754's when everyone thought S754 was a lost cause 😛

 

[Viditor]

Originally posted by: buzzsaw13

Originally posted by: munky
I don't think they'll be making any 65nm s939 chips. Which is too bad, because a x2 3800 that does 3ghz on average would be a tough deal to pass up.

I can always hope and pray! They did make Venice S754's when everyone thought S754 was a lost cause 😛

That's about as big a longshot as it gets, mate...
939 is due to be EOL at the end of this year...

 

[myocardia]

Originally posted by: Viditor
Some questions...
1. Don't you think that delaying volume production for 6 months has allowed AMD to improve their 65nm process over Intel's?
2. As Intel was able to drastically reduce their power usage and increase clockspeed at the same time when they wen to 65nm, do you think AMD won't be able to do the same?
3. AMD's 90nm launch was FAR smoother than Intel's (after delaying for the same length of time)...is there a reason that you think this time will be different?

Some answers:

1) Definitely not. In the technology business, getting your product to market first is all that matters, assuming that it will work as advertised once it gets there. Knowing the way that AMD does business (I have been buying only AMD chips since my 450 MHz K6-III), I think that AMD had to delay production of their 65nm chips for the same reason that it took Intel so long to actually get good at producing them-- smaller processes take time to get right, period.

2) Yes, but I don't see it being enough. Conroe/Allendale is already 65nm, and overclocks at least as well as any K8 chip will.

3) I have no idea whether AMD's 65nm launch will be smoother than Intel's. But, since that type of thing usually involves money, I don't see AMD as being "ahead" of Intel in any way, just because they've launched one process shrink better than Intel.

Originally posted by: zsdersw
and yet, he expects K8L to sail right past Conroe. Interesting...

I'm not sure you understand exactly what K8L is, man. It's not a K8 derivative, like the move from Skt. 754 to Skt. 939 was. It's actually a new chip, like the Conroe/Allendales. It was going to be named K9, but AMD thought otherwise about the name, and decided to name it K8½. The Roman numeral L=50, hence the K8.50.

 

[zsdersw]

I'm not sure you understand exactly what K8L is, man. It's not a K8 derivative, like the move from Skt. 754 to Skt. 939 was. It's actually a new chip, like the Conroe/Allendales. It was going to be named K9, but AMD thought otherwise about the name, and decided to name it K8½. The Roman numeral L=50, hence the K8.50.

Of course it's a new chip.. but that was never in dispute as far as I'm concerned.

 

[Viditor]

Originally posted by: myocardia

Originally posted by: Viditor
Some questions...
1. Don't you think that delaying volume production for 6 months has allowed AMD to improve their 65nm process over Intel's?
2. As Intel was able to drastically reduce their power usage and increase clockspeed at the same time when they wen to 65nm, do you think AMD won't be able to do the same?
3. AMD's 90nm launch was FAR smoother than Intel's (after delaying for the same length of time)...is there a reason that you think this time will be different?

Some answers:

1) Definitely not. In the technology business, getting your product to market first is all that matters, assuming that it will work as advertised once it gets there. Knowing the way that AMD does business (I have been buying only AMD chips since my 450 MHz K6-III), I think that AMD had to delay production of their 65nm chips for the same reason that it took Intel so long to actually get good at producing them-- smaller processes take time to get right, period.

You have to step back and think about WHY getting to market first can be important...
In this case, the only chip AMD would have been competing against was their own! Ask yourself, if AMD had come out with 65nm chips in Jan-Sept, would it have changed their income or strategic marketing at all? The answer is an emphatic no! In fact, they are still gaining marketshare with their 90nm chips while more than half of Intel's chips are 65nm...
You should also keep in mind that to convert to a new node is VERY expensive for many reasons. The first 2 off the top of my head are:
1. You cut short the amortization of the old equipment (can cost $100s of millions)
2. You have less time to tweak the manufacturing process thus reducing your yields (again, costing hundreds of millions)

2) Yes, but I don't see it being enough. Conroe/Allendale is already 65nm, and overclocks at least as well as any K8 chip will.

I think you missed my point...don't you think AMD at 65nm will be able to overclock at least as well?

3) I have no idea whether AMD's 65nm launch will be smoother than Intel's. But, since that type of thing usually involves money, I don't see AMD as being "ahead" of Intel in any way, just because they've launched one process shrink better than Intel.

I agree that none of us really knows one way or the other...my comment was to point out that spending money isn't really the issue. By pointing out how well AMD did on the 90nm launch, it illustrates this quite well I think...

 

[Rommel44]

I hope those 65nm AMD CPUs will be available soon.

 

[harpoon84]

Viditor - You can't compare 65nm Conroe to 65nm K8 - different architectures, different clockspeed ceilings.
Look how much 130nm to 90nm gained for AMD over 3 years - a mere 800MHz.

I think 65nm K8 is just a dress rehearsal for the real 'next big thing' - K8L.

Even a 3GHz K8 will not be competitive with the top end C2Ds. As a rough rule, C2D x 1.2 = eqv K8

E6600 = 2.88GHz K8
E6700 = 3.2GHz K8
E6800 = 3.51GHz K8

A 65nm K8 vs Conroe is about as exciting as a 65nm Presler P-D vs AMD X2 - ie. not very.

Bring on K8L!

 

[Regs]

Originally posted by: harpoon84
Viditor - You can't compare 65nm Conroe to 65nm K8 - different architectures, different clockspeed ceilings.
Look how much 130nm to 90nm gained for AMD over 3 years - a mere 800MHz.

I think 65nm K8 is just a dress rehearsal for the real 'next big thing' - K8L.

Even a 3GHz K8 will not be competitive with the top end C2Ds. As a rough rule, C2D x 1.2 = eqv K8

E6600 = 2.88GHz K8
E6700 = 3.2GHz K8
E6800 = 3.51GHz K8

A 65nm K8 vs Conroe is about as exciting as a 65nm Presler P-D vs AMD X2 - ie. not very.

Bring on K8L!

And yeah, I would agree. There is a ceiling and there is only so many Mhz you can pump into a K8 before it has a pipeline stall and waste a good percentage of clock cycles (like the prescott would). Hopefully the K8L will be able to feed 3 GHz and more efficiently.

 

[formulav8]

While Rev 'G' cpu's won't be as 'exciting' as K8L or even Core cpu's, it 'won't' be as boring as the PD's cpu's are.

There are cpu die shots that clearly show a some-what different core. In other words, a few 'additions' to the core for Rev "G". While its certainly not 'concrete' what the new adttions are, its properly close to accurate that they include a extra Complex Decoder which will now be a total of 4x. OoO load and store buffer and OoO read/ write buffer.

The very first 65nm cpu's may be simply a Rev F shrink, but there will be Rev 'G' core changes before K8L comes around. If the die shots are accurate, the cpu performance will be increased to a certain degree before the real increase comes with K8L.

So hopefully it won't just be a boring Rev 'F' shrink.

Just my penny, I am by no means 'learned'. Just relaying some data 🙂


Jason

 

[Rommel44]

Originally posted by: formulav8
While Rev 'G' cpu's won't be as 'exciting' as K8L or even Core cpu's, it 'won't' be as boring as the PD's cpu's are.

There are cpu die shots that clearly show a some-what different core. In other words, a few 'additions' to the core for Rev "G". While its certainly not 'concrete' what the new adttions are, its properly close to accurate that they include a extra Complex Decoder which will now be a total of 4x. OoO load and store buffer and OoO read/ write buffer.

The very first 65nm cpu's may be simply a Rev F shrink, but there will be Rev 'G' core changes before K8L comes around. If the die shots are accurate, the cpu performance will be increased to a certain degree before the real increase comes with K8L.

So hopefully it won't just be a boring Rev 'F' shrink.

Just my penny, I am by no means 'learned'. Just relaying some data 🙂


Jason

Yeah thats exactly what I´m expecting.

 

[Regs]

What saddens me the most is the article on AMD's official site about the new Rev F Opteron. "next-Generation" repeated a million of times when it's nothing more than just another step up for AMD instead of the consumer. I know the K8 has been good to you guys but it's just time to let it go.

I sorry I can't find the article at amd.com but I remember saying how it will delivery better performance and I am also sure it used the term "leading". It's a complete reverse of rolls from what I can remember. Like Intel stuffing new technology into a dead architecture called Prescott.

Rev G will be more of the same. Thermal and Power management with more security features with added instruction sets that will add some small performance improvements with applications most of don't use.

What scares me most is that with all these incremental improvements what hope can we get from the K8L? The same?