Sign in with
Sign up | Sign in
Your question

A64 limited to 2.4Ghz @ 130nm

Last response: in CPUs
Share
February 7, 2004 12:24:24 PM

I finally managed to get one or another link on the frequency headroom that the current A64s have. Here's a quote from <A HREF="http://www.xbitlabs.com/news/cpu/display/20040205093054..." target="_new">anandtech</A>:
Quote:
On the other hand, AMD is using its time-proven 130nm (0.13 micron) process for its current and some future processors. Even though it means that the company is not going to run into serious process-related problems with yields and costs, it also implies that there is no huge clock-frequency headroom for AMD with its current manufacturing technology. Given that 2.40GHz is likely to be the last speed bin for 130nm SOI process, AMD will have to utilize 90nm nodes for 2.60GHz and faster chips. AMD’s transfer to 130nm process was not an easy one, just like AMD’s first experiments with 130nm SOI technology. Therefore, there are quite a lot of concerns in regards AMD’s ability to successfully ramp up mass production using 90nm SOI technology in the firm’s Fab 30 in Dresden, Germany, later this year.

So, this year will probably be a messy one, as Intel cleans up the prescott core and the 90nm process in order to get Scotty up and doing what it's supposed to do (fly at high frequencies) and AMD transitions to 90nm.

The 2.4Ghz AMD processors will probably be the top performers from AMD for the better part of the year, it would seem. (3700+ and FX53)

As for the 3700+ vs 3.6Ghz, I still have a small hope that the guys at Intel get reasonable and get a 3.73Ghz Scotty out. To level the playing grounds... I don't see a 3.6Ghz defeating a 2.4Ghz AMD processor easily...

<P ID="edit"><FONT SIZE=-1><EM>Edited by Mephistopheles on 02/07/04 12:29 PM.</EM></FONT></P>

More about : a64 limited 4ghz 130nm

February 7, 2004 1:43:09 PM

It depends on which will become more popular amongst to-be-released softwares in 2004. Hyperthreading or 64bit.
From current trend it will be advantage HT.

A fine day!
February 7, 2004 1:47:19 PM

I really hope AMD come up with something like that cause I hope AMD to do well.

A fine day!
Related resources
Can't find your answer ? Ask !
February 7, 2004 2:49:06 PM

AMD will go 90nm 4th QTR.04
February 7, 2004 4:33:50 PM

That's just Anand's speculation, and what would he know ? Seriously, he is getting it backwards: he thinks 2.4 GHz is the fastest speed grade achieveable on 130nm SOI, just because AMD expects anything faster to be on 90nm. That reasoning makes no sense; there are other reasons AMD wants to move to 90nm than just his supposition (is that a word ?)that 130nm is limited to 2.4. Even if they could scale 130nm K8's to, let's say, 4 or 5Ghz, they would still never be produced, because by that time, AMD will produce 90nm or 65nm ones instead.

Mark my words, if AMD has trouble with 90nm, they will squeeze out faster than 2.4 GHz chips on 130nm. Porbably at the cost of yields, but they have always managed to do this with any core, and always beaten expectations. Think Tbred-b, Palomino, etc..

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 7, 2004 5:00:35 PM

Quote:
Mark my words, if AMD has trouble with 90nm, they will squeeze out faster than 2.4 GHz chips on 130nm. Porbably at the cost of yields, but they have always managed to do this with any core, and always beaten expectations. Think Tbred-b, Palomino, etc.

Exactly like Intel and the 3.4Ghz Northwood that never should have been, eh?...

I think it all remains to be seen... However, it is pretty clear that they can, at least, get 2.4Ghz out of current processes.
February 7, 2004 5:04:30 PM

Yeah but faster that 2.6 is close to impossible.K7 dont goes much higher either.2005 Q1 for 90NM is more realistic if you ask me.Can you get me a CPU lately that dont have been delay.

Just to show bad
February 7, 2004 5:50:45 PM

"Exactly like Intel and the 3.4Ghz Northwood that never should have been, eh?...

I think it all remains to be seen... However, it is pretty clear that they can, at least, get 2.4Ghz out of current processes."

Yeah....(sorry for not using correct quotes above)
AMD is planning another major stepping before 90nm. Nothing says they won't also be able to improve performance/clocking. Remember when they went to 130nm? The first steppings weren't any faster than the 180nm, but 3 mon ths later they had a new stepping that scaled much better.

In the final analysis, I think much depends on how the 90nm transition goes. If it goes well and they can release chips by mid Q3 they will probably not worry about scaling the 130nm products past 2.4ghz...unless Intel comes out with a 3.8ghz product early, in which case, AMD will likely cherry pick enough to release a 2.6ghz variant on 130nm.

Keep in mind though, that just a month or so ago, most enthusiast sites were suggesting that AMD would probably keep the performance crown for all of 2004 and into 2005.

Time will tell, but this is how I think things will go.

AMD will release a 2.4ghz version in Q2 or late Q1 both FX and standard A64.
AMD will release a 2.6ghz cpu by mid Q3...perhaps even in late Q2
AMD will release a 2.8ghz chip before end of 2004.

Perhaps I am a bit optimistic, but I haven't seen anything yet that would suggest otherwise. In fact, so far, AMD has kept up with plan (as far as I know it) since A64 was released, coming out with a new 3000+ and then following with a 3400+...all in the 3 months following initial release.

Since it takes around 12 weeks from start to finish, to fab a CPU it stands to reason that as AMD tweaks the chip and process, that each quarter would see improvements in yield and speed grades.

Mark-

<font color=blue>When all else fails, throw your computer out the window!!!</font color=blue><P ID="edit"><FONT SIZE=-1><EM>Edited by zengeos on 02/08/04 09:35 AM.</EM></FONT></P>
February 7, 2004 6:00:27 PM

Quote:
AMD will release a 2.4ghz version in Q2 or late Q1 both FX and standard A64.
AMD will release a 2.8ghz cpu by mid Q3...perhaps even in late Q2
AMD will release a 3ghz chip before end of 2004.

I don't think that optimistic is exactly the right word to describe your prediction. IMO you are far beyond optimism here!
February 7, 2004 6:08:46 PM

Quote:
The 2.4Ghz AMD processors will probably be the top performers from AMD for the better part of the year, it would seem. (3700+ and FX53)

As for the 3700+ vs 3.6Ghz ...

Please correct me here if I am wrong but isn't it supposed to be FX-53 vs Prescott 3.6GHz? The 3700+ is part of their "value" line of processors, remember? I don't know about you, but I don't consider the P4EE a desktop CPU. The 3.4GHz one costs more than $1000, when the FX-53 will be quite cheaper than the FX-51 because of S939. The only reason I see for AMD to keep their prices high for that CPU is because they have no competition right now from Intel. But still, I don't think they will ask as much as they ask now for the FX-51.

And even if Intel manages to convince everyone that their top-of-the-line Prescott is actually fighting against the 3700+ (so their CPU won't be consider a clear loser), they will have a real problem when a 2.6GHz A64 will be released, since it will most probably be manufactured only for the dual-channel memory controller S939.
February 7, 2004 6:14:30 PM

Quote:
AMD will release a 2.4ghz version in Q2 or late Q1 both FX and standard A64.
AMD will release a 2.8ghz cpu by mid Q3...perhaps even in late Q2
AMD will release a 3ghz chip before end of 2004.

Even if the could, and I'm not saying it's impossible, what would be the reason? All they need to do is keep a little ahead of intel. Releasing a 3ghz before 2004 would be economical suicide IMO.

<b>wooooow <font color=red> Killer Klowns </font color=red> ... from <font color=blue>outer space</font color=blue>... HOLY SH¡T!</b>
February 7, 2004 6:14:48 PM

Quote:
It depends on which will become more popular amongst to-be-released softwares in 2004. Hyperthreading or 64bit.
From current trend it will be advantage HT.

Did you see A64's performance at media encoding, under WinXP 64bit using a 32bit application? It gained 15% just because of the OS. At first I wasn't really sure about the performance gains of AMD64 but now I am really excited about it. Imagine what will happen when a 64bit application will be used, on the final Windows version, with the latest chipset drivers and everything!

And I know that some of you will say that it sucked at games but what's important IMO is that the A64 showed real performance gains where CPU and memory performance mattered. When it comes to games, I assume that 99% is because of the drivers (video card/AGP), which is a problem that can be "easily" solved.
February 7, 2004 7:01:12 PM

Sure.

<sarcasm>
But before that, Intel will release a prescott stepping that'll allow Scotty to go to 5Ghz by November...
</sarcasm>

I've never seen an optimist like you. You're way off optimism... you're more like a daydreamer... sorry... :eek: 
February 7, 2004 7:16:25 PM

Quote:
Please correct me here if I am wrong but isn't it supposed to be FX-53 vs Prescott 3.6GHz? The 3700+ is part of their "value" line of processors, remember?

Well, not really... It's more like the FX processors being on their "Expensive" line of processors. 3200+ and 3400+ are price-targeted against the 3.2E and 3.4E... This is not a value line at all.

As for FX-53... I don't really know if S939 FXs will be much cheaper... Why should they?... (I really don't know)
Quote:
they will have a real problem when a 2.6GHz A64 will be released, since it will most probably be manufactured only for the dual-channel memory controller S939.

Yes, but according to current estimates, AMD will only reach 2.6Ghz with 90nm processes. This means, basically, that it'll probably be late 2004... By then, Intel will reportedly have 4Ghz Prescotts out, with "wide support" - if they can get it - for high-bandwidth DDR-II... But it'll all depend on a lot of factors that we can't really predict right now.
February 7, 2004 8:41:05 PM

Reality check for everyone.

DDR2 is expect to cost twice as much as there DDR conterpart.Increasing the NB of pin on the mothersboards.Socket 939 will bring smaller A64 with 512 L2 it will just to be extra money for AMD i dont think they make a lot of money on a 100 million CPU on 130 NM SOI just the cost of SOI is huge.90nm for Intel wont be in mass production before a while and even more time before it right for big clock speed increase.For AMD 90NM micron process dont get surprise if they get delay until 2005 1H.I dont expect much before the end of the year.

Just to show bad
February 7, 2004 9:14:14 PM

Quote:
As for FX-53... I don't really know if S939 FXs will be much cheaper... Why should they?... (I really don't know)

S939 A64-FX will not require ECC ram and it will cost less to build a S939 A64-FX system.

----------------
<b><A HREF="http://geocities.com/spitfire_x86" target="_new">My Website</A></b>

<b><A HREF="http://geocities.com/spitfire_x86/myrig.html" target="_new">My Rig & 3DMark score</A></b>
February 7, 2004 9:17:25 PM

Right, I already knew that part. But there's nothing saying that it won't be priced above $700 like the extremely pricey ($750) FX-51, so I think it'll still be above my budget, for example. And I suspect it'll still be pretty expensive for most.
February 7, 2004 9:27:47 PM

DDR2 is expected to cost twice as much in Q1, but I'd expect some price reductions to the end of the year... wouldn't you?...

As for AMD's 90nm process being delayed... well, there's no indication that they're on time and there's no indication of a delay; however, if I had to guess, I'd say AMD will only really get into 90nm in 2005 too.
February 7, 2004 9:37:15 PM

I thought Newcastle was a cost-effective solution... the A64 3000+, that is.

I don't know... Truly cost-effective solutions using S939 will probably exist, once AMD64 processors have been around for a little longer and newer processors come into play. Say, when they get the 3700+ out... Then there'll be a price reduction as well.
Quote:
Maybe with 512k or 256k L2 cache instead of 1 MB.

Exactly, like newcastle: A64 3000+. It already exists and is priced at around $215±5, the lowest-priced A64...

Hopefully, however, AMD won't get into too much of a confusion with its PR ratings... 3000+ at 2.0Ghz, 512KB L2; 3200+ at 2.0Ghz, 1MB L2, 3400+ at 2.2Ghz, 1MB L2... If they want a 2.2Ghz with 512KB L2, they'll have to rate it as another 3200+... And it'll probably go on like that, and that is the problem. Newcastle will overlap with sledgehammers, and it always will. So maybe they could reinvent the name "Duron 64" for Newcastle? It would ease the confusion a bit, who knows...
February 7, 2004 11:15:02 PM

All cpu since willimete from AMD and intel been delay no exception (X86).In overall AMD release there process 1 year after intel.I see no reason to think they can do better this time.

I guess price will go down but how fast and how it performe this have to be answer before we see high end OEM box come with DDR-2.

Just to show bad
February 7, 2004 11:25:16 PM

Just because some writer at anand says the a64s wont release faster than 2.4 gigs does not make it so. Initially the a64s were to be released at 1.8 gigs, but yields were too good. To say that they cant get higher clockspeeds is not realistic. This is a new chip using a new process, with a much larger die size. Many of the newer xp chips are capable of running at 2.4. Does anyone seriously believe that the a64 cant scale better than the xp?
As for the transition to 90 nanos, intel had/has a major problem with leakage current. This is mostly caused by dissimilar chages on chip. The extra stages on the P4 are a prime cause. Component positioning and layering are the answer. Amd went through that with the Tbred B chip. Amd has had working 90n chips for a few months already. They will have a much smoother transition than intel.
February 7, 2004 11:27:01 PM

You mean like Amd's newcastle?
February 7, 2004 11:50:29 PM

Quote:
Just because some writer at anand says the a64s wont release faster than 2.4 gigs does not make it so. Initially the a64s were to be released at 1.8 gigs, but yields were too good. To say that they cant get higher clockspeeds is not realistic.

Maybe you're right here, but... It won't be like that just because you're suspicious of what someone at anandtech said... For instance, THGs and Xbitlabs' A64 3400+ refused to clock at more than 2.37Ghz; none of them got to 2.4Ghz reliably. Xbitlabs commented that the current A64 stepping might not even make it to 2.4Ghz with good yields. Anandtech seems to go by that, too. AMDzone's review got the 3400+ to 2.3Ghz only, too.

So, quite honestly, this is a new chip, with a new manufacturing line, but the process it's on is the old 130nm one. So, in light of every information I've seen, yes, I believe A64 might not scale better than XP right now.

Sure, a next stepping might be underway before the switch to 90nm. But until then, it is unlikely, from the info at hand, that it will go beyond it, yes. And I take it that it's not because of 1 writer, it's because of several review sites that have had their hands on the actual processors and tested them freely.

So, from several reviews on the net, it seems that Prescott does have more frequency headroom (and I mean the current stepping!) than A64. Right now. Sure, AMD will probably get the 2.4Ghz on another stepping soon, and Intel might just change steppings too... But right now, prescott can go beyond 10% of its current speeds reasonably OK, while A64s cannot.
February 8, 2004 12:10:02 AM

If you check those sites again you will find that the poor oc is based on the board's inability to handle faster fsbs, not on the chip being able to run at higher multipliers. Better boards are due out soon. Then we can talk about ocing the a64s.
February 8, 2004 12:42:02 AM

I dont if that true but intial A64 3000+ were not new castle they were A64 3200+ with 512 KB L2 cache enable.

Just to show bad
February 8, 2004 1:58:16 AM

Is that when Troy (Opteron 2xx 90nm) is released too?

Maxtor disgraces the six letters that make Matrox.
February 8, 2004 2:02:12 AM

You're forgetting odd numbers...3300+ could be the Newcastle 2.2GHz.


Maxtor disgraces the six letters that make Matrox.
February 8, 2004 2:30:51 AM

> For instance, THGs and Xbitlabs' A64 3400+ refused to
>clock at more than 2.37Ghz

initial athlon 64's didnt overclock worth a damn either. Did that mean the K8 was limited to 2 GHz ? remember, K8 is produced on a new process (SOI), with a new sort of bus (HT) and a brand new feature (integrated MC). Who's to know what is holding back the overclocks ? While overclocking might have given an indication of clock frequency "headroom" in the past, it seems K8 doesnt follow those rules. Its perfectly possible the hypertransport bus is limiting the overclocks, and nothing else. Its also a known fact that, unlike bulk silicon, SOI doesnt like to be overvolted to achieve higher clocks.

BTW, you called another poster that predicted 3 GHz K8s by the end of the year, and 2.6 on 130 nm "overly optimistic". Now think... K8 is a new core, with a slightly longer pipeline, and definately designed for higher clocks than K7. Its produced on a process (SOI) that allows lower leakage and/or higher frequency than K7 (bulk silicon). K7 topped out at 2.33 GHz, and you seriously think 2.4 is the highest K8 could ever do ? Doesnt sound reasonable to me. Have you seen the specs on the latest 3400+ stepping ? Cool & Quiet allows it to operate at 1 GHz@ 22W (I think), compared to 800 MHz@35W for the previous stepping. Sure sounds like A64 has *plenty* of headroom, since power consumption/leakage can usually be traded for clockspeed.

As for the 3 GHz claims by the end of the year; it may be optimistic, but let me tell you its what AMD is expecting. Could be a month or so earlier or later, but its still what is on the roadmaps. Stop making the mistake of expecting intel to do better than its roadmaps, and at the same time, expect AMD to do worse.

Finally, you state there is no evidence about AMD's 90nm status; there is, and so far, everything seems to be going smooth. 90nm hammers have been sampling for some time, and AMD seems to be *very* happy with the results. There are no guarantees of course, but things do look good, unlike intel's 90nm process/prescott design which was rumoured/known to have issues for more than 6 months.

expect those 3 GHz hammers by the end of the year, and if not, early Q105.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 8, 2004 3:58:11 AM

A got some question for how can HT hold back the chip frequency.Now how can SOI improve clock speed.How AMD can decrease the disparity in clock speed in % with Intel by that much.

Just a cheapo theory
NW 20+ stage 3.4 GHZ 3.4/20=0.17
K8 14 stage 2.4 ghz 2.4/14=0.17xxxxx


Just to show bad
February 8, 2004 4:30:12 AM

SOI decreases current leakage, which reduces heat.
February 8, 2004 5:47:28 AM

once again SOI dont bring faster chip.
Just to show bad<P ID="edit"><FONT SIZE=-1><EM>Edited by juin on 02/08/04 02:49 AM.</EM></FONT></P>
February 8, 2004 9:41:43 AM

Duh yup just cause a chip runs cool dosn't mean you can clock it faster. Look at scotty, its a toaster, but it will clock 2 more gigs right.
February 8, 2004 9:57:52 AM

less heat might allow clock increases, but it isn't always a certainty; even supercooled chips can't go beyond X.X Ghz

"Some mice have two buttons. Macintosh has one. So it's extremely difficult to push the wrong button." - Apple ad. circa 1984.
February 8, 2004 11:46:47 AM

OK, you're right, odd numbers could be a choice, too... A good one at that.

Besides, going from 2.2Ghz to 2.4Ghz already gave them an odd number (and for those who actually think about what they read... I mean odd hundreds...)... It implies a 9% increase in clock, which gives them a rating of ~3710 if scaled linearly from the 3400+...
February 8, 2004 11:58:49 AM

Maybe you're right about A64's overclocking capabilities not indicating the frequency headroom appropriately either... It's not as if there's abundant information on the topic. So there's little way of really knowing for sure.

As for reaching 3Ghz this year, <A HREF="http://www.c627627.com/AMD/AthlonXP/" target="_new">these roadmaps</A> are relatively optimistic for AMD, and they still do not assume reaching 3Ghz this year... But it could all be wrong too...
February 8, 2004 12:07:24 PM

>A got some question for how can HT hold back the chip
>frequency

Hu ? Unless you have multiplier unlocked samples, the only way to overclock A64/FX is by overclocking the HT bus AFAIK. People used to think the Athlon classic had no headroom because the KT133 chipset wouldnt overclock more than a few percent...

>Now how can SOI improve clock

Here is a quote from IBM: "SOI adds a layer of oxide material between a transistor and the silicon it rests on inside a chip. The oxide insulates the transistor from the silicon, reducing the amount of electrical energy absorbed by the silicon. <b>The transistor, therefore, can run faster </b>or it can be tweaked to consume less power at a given clock speed

That is how.

>How AMD can decrease the disparity in clock speed in % with
>Intel by that much.

A better secret sauce ? Seriously, how did they do it in the K7 days ?


>NW 20+ stage 3.4 GHZ 3.4/20=0.17
>K8 14 stage 2.4 ghz 2.4/14=0.17xxxxx

And here is a different way of looking at it:
K7 12 stages 2.33 GHz 2.33/12=0.194
K8 14 stages x GHz x=0.194x14=>2.7 GHz

Since K8 is a lot more like K7 than NW, if there is any credibility in your theory (there is none), my number would be closer.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 8, 2004 12:22:00 PM

>As for reaching 3Ghz this year, these roadmaps are
>relatively optimistic for AMD, and they still do not assume
>reaching 3Ghz this year... But it could all be wrong too...

Damn those model numbers. The one I had seen didnt have clockspeeds,but had a 4400+ model and FX57 for the turn of the year. I assumed that to be a 3 GHz model, but then again, it probably is not. Still, 2.875 Ghz (if true) is pretty close.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 8, 2004 12:30:14 PM

Damn these stupid PR numbers! AMD need to do something about this 3400 not being anywhere near 3ghz. I mean even IF you know what the basis for the numbering system is - and most that buy pcs dont, then it still is confusing, and could be seen imo as false advertisinng.

XP2000, 256ddr 2100ram, GF4 MX440, XP Pro
February 8, 2004 12:37:14 PM

I accidentally skipped a speed grade and thus ended up at 3ghz instead of 2.8ghz. My apologies...and I have corrected the error in my original post.

Mark-

<font color=blue>When all else fails, throw your computer out the window!!!</font color=blue>
February 8, 2004 12:40:22 PM

Lets not restart that discussion *again* please..

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 8, 2004 7:05:24 PM

The PR numbers can be quite confusing as to the case of Athlon XPs vs P4b/c.
If AMD is willing to allow their PR rating to be more flexible (together with price) then I have nothing against it. But look at the current XP2500+, it is slower than P2.4c with FSB800 and dual ddr.
This is not good for their PR for sure cause people tend to
remember bad experience better than good.

A fine day!
February 8, 2004 8:57:57 PM

It's just a name! Can you get over it?
February 8, 2004 10:13:27 PM

Amd uses a linear rating system for thier chips. Intel has a,b,c,&d chips as well as celeries and new celeries coming out. Talk about confusing. Which 2.8 is faster, and do you have one of those or one of those. Personally I am in favour or a universal PR system so people dont get sucked in by the letters.
February 8, 2004 11:50:37 PM

Hu ? Unless you have multiplier unlocked samples, the only way to overclock A64/FX is by overclocking the HT bus AFAIK. People used to think the Athlon classic had no headroom because the KT133 chipset wouldnt overclock more than a few percent...

I think dont think so most mothersboards offer the same performance on overclocking and Nvidia that there 3 chipset with ht link that not something new for them.

On pure transistor performance 90Nm from intel is about 2 faster that anything on the market right now.That dont mean the chip can go faster.The electron speed cannot be change.The interconnect delay wont chage if there are SOI or FD-SOI.


A better secret sauce ? Seriously, how did they do it in the K7 days ?

They never did it, when the P4 was release it was 200 mghz faster than the tbird.Now it well over 1 ghz.I never said my theory was right it just a constation.

Just to show bad
February 9, 2004 12:57:37 AM

Well you are right that the scotties run 1 gig faster than the A64, too bad the scotty cant keep up. As far as SOI goes, and this may be hard for you to understand, if a chip is not as badly affected by leakage current, and the heat that causes, it will be able to recognize gate changes more quickly, so the chip can run faster.
February 9, 2004 4:35:52 AM

So i your mind less heat mean gate change quicker.Also thank you to teach me that gate switch i dint know that.Where is the gate by the way to your world.

Just to show dad<P ID="edit"><FONT SIZE=-1><EM>Edited by juin on 02/09/04 01:36 AM.</EM></FONT></P>
February 9, 2004 4:41:32 AM

IPC or MHz don't mean squat!

The AMD side is "locked" on the IPC-argument.
The Intel side is "locked" on MHz-argument.

I only care about preformance/design/scalability...

Terra

Don't pretend - BE!
February 9, 2004 4:45:19 AM

Ok what is your definition of IPC.Give it a try

Just to show dad
!