Sign in with
Sign up | Sign in
Your question

newbie question about why amd can't get ghz up?

Last response: in CPUs
Share
August 13, 2003 5:33:43 PM

If ppl can easily enought over clock athlon xp to 2.5 ghz and even up to 3 ghz, why can't amd get thier cpu's clocked speeds up themselves? I am total newbie at hardwar stuff so I don't have clue why they can't. I have been wondering about for quite along time now...
August 13, 2003 6:05:23 PM

I have only seen amd chips hit 3ghz on liquid nitrogen...

Well...i think that they have not clocked to 2.3-2.5ghz for a few reasons....first of all...they don't wan't the athlon XPs to be faster than the low end opterons/A64's

Next the chips will run alot hotter...at 2.3ghz...although possible on stock cooling the chips will have absolutly no headroom...you have to reilise that overclockers live on a much finer line than amd does...they need to at least give their chips 100mhz of headroom...for stability and heat's sake...

Also...not to many chips go above 2.2-2.3 on stock voltage...that would mean that they would be pretty much factory overclocking chips...although this has been done in the past...normally it is not the ideal solution...because they will have to beef up the cooling to support higher voltages....

How many bits are in a nibble?

If you answer this question...you must put it in terms of Pi...If not you do not win the rage pro for answering...
August 13, 2003 6:48:08 PM

Yeah I think really it comes down to marketing. Both AMD and Intel have chips ready to rock wayyyyyyy ahead of launch. However you would think AMD would release something a bit faster.

The Rig
2.4c @ 3.0
OCZ pc3500
Asus P4C800-E
4 Maxtor 250gig serial ata drives
Enermax 465watt
Viewsonic 17" tft
ATI 9800Pro @ stock
Related resources
August 13, 2003 7:46:43 PM

As Butthead puts it:
Huh Huh Huh, you said amd can't get it up.

Wanted: Large breasted live-in housekeeper. Must be a good cook, organized, and willing to pick up after me.
August 13, 2003 8:13:27 PM

Think of a Ford auto plant. Now think of a long line of assembly inside that plant. That's what we call a pipeline. Each modern CPU has many. These chains divide the task of executing what instructions and data ask, or in car analogy, dividing the different parts to assemble to complete one car (one puts the engine, the other puts the hood, the other fixes the dashboard, etc.). So imagine that car plant having 10 stages of assembly.
Each has a worker.

-- -- -- -- -- -- -- -- -- --
* * * * * * * * * *
The above is a 10 stage assembly line. Each star represents a worker in that stage.

Now, if I told my workers (I, the boss of that plant) to work faster, because I wanted for example that a new part gets passed to the next stage each 5 minutes, rather than 10 minutes, I'd have to make sure each worker is fit and that it isn't beyond his physical strength/capabilities.
This, is what the Athlon suffers from, at the moment.
It has a 10-stage Integer pipeline and 15 for floating point (decimals). Issue is, its physical limits are starting to be reached, as the stages in there are working harder and harder, that it becomes a problem.

Assume in that car plant, that each stage now is divided into two. That makes 20 stages, right? Wouldn't you agree if I told you that each stage is effectively twice easier and faster to complete? Granted that means 20 workers but this is an analogy with no constraints.

Now, with 20 workers, it means you can effectively reduce the time before new parts can pass through.

Now you know that a Hertz is a clock cycle, and it means a certain amount of beats per second. 1 GHZ means 1 billion cycles per second.

This means inside the CPU, each stage works and passes its executed tasks to the next stage a billion times a second. So here is the Athlon's problem. Since it has 10 stages, compared to the Pentium 4's robust 20, each time you up the clock speed, you reach faster the physical limit than a P4 ever will.
Theoretically a P4 would reach twice the clock speed limit than an Athlon, but this is purely theoretical and Athlons could very well not even reach half of the P4's max.

In conclusion, the Athlon's pipelines work much harder than a P4's at the same speed, so heat, physical constraints all play a role in this, making it harder for it to reach clock speeds higher than Pentium 4s. AMD's K8 will have 12 stages, however the 2 extra are supposedly ADDED, not divided results. Meaning it still works at the same difficulty. Converting into 0.09m process for the K8 will help it reach higher clock speeds, along with SOI technology, but it is unclear how far it will go before any higher process change will simply not work anymore. My guess is the Athlons at the later processes, will top at 4GHZ. AMD HAS to bring forth a brand new core with 20 stages, simplifying each from the K7's 10. It will easily allow them to reach 5GHZ at later processes.


Hope this helps.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.<P ID="edit"><FONT SIZE=-1><EM>Edited by Eden on 08/13/03 04:20 PM.</EM></FONT></P>
August 13, 2003 10:02:20 PM

But the thing is...that most overclockers can get to those speeds stabily on stock voltages...2.3ghz stock Vcore is not uncommon of an athlon...

So i think as of yet the athlon has NOT reached it's physical limit...it is close though i agree...amd has been milking an old design for far to long...

However i would assume that this is why the coppermine 1.13 had problems?


Proud owner of DOS 3.3 :smile:
August 13, 2003 10:33:23 PM

there are 4 bits in a nibble

wpdclan.com cs game server - 69.12.5.119:27015
August 13, 2003 11:22:35 PM

aww god...so many people have anwered my siggy...

Anyhow...i have been forced to change...that was a very short lived siggy...




Proud owner of DOS 3.3 :smile:
August 13, 2003 11:30:22 PM

word to the wise...dont put a question or even a "?" in your siggy...


Proud owner of DOS 3.3 :smile:
August 14, 2003 3:04:57 AM

Did your new upgrade arrive?
August 14, 2003 3:10:27 AM

Quote:
But the thing is...that most overclockers can get to those speeds stabily on stock voltages...2.3ghz stock Vcore is not uncommon of an athlon...

I did not disagree nor deny that.

Quote:
So i think as of yet the athlon has NOT reached it's physical limit

I never even said something remote to that. I said it will get MUCH closer to a GHZ limit than the P4. The P4 will probably get to 10GHZ at a given process, while the Athlon will likely top out at around 4GHZ with its current heat problems.

The Coppermine was even a lie in its own name, it was never copper! If it were, it'd likely be fairly close to what Athlons scale to. Then again I am not sure of how long the FP pipeline is for the P6 core. But it WAS a very recycled core, I mean, it's been here for over 6 years and has served well the community.

Again, the Athlon will reach a physical limit faster than the P4, at both, processes and final limits. It means at the 0.13m limit, the P4 would likely top at 3.6GHZ (it already was close to 5GHZ at exotic cooling), while the Athlon at NORMAL cooling or adequate, would reach 2.4GHZ at most.

With perfected processes, I would expect a P4 to reach roughly twice the cap the Athlon would get at the same process, around 5GHZ, and 2.5GHZ for K7. Mind you the K7 has been an EXCELLENTLY scaling CPU. AMD would have likely given up the K7 and moved to K8 if they knew it would not scale well. But it was not expected that it would. Fact is, for a pipeline twice less of the P4, the Athlon is doing a damn good job making sure the P4 is not topping out clock speeds twice of the Athlon on the same process. It is likely to stay that way until the physical limit around 4-5GHZ (theoretical and assumed) for the K7 occurs and the P4 really goes right up to 10GHZ with Intel's refining.

All of this is purely theoretical mind you.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
August 14, 2003 3:55:55 AM

coppermine is the name of a town...intel names have never had anything to do with technology...i dunno why people think that it is implying that they use copper tech in their cpu's...

Yes the p6 core has its roots in the pentium MMX...

Sry...i thought you were implying with your pipeline deal that the reason amd has not "bumped it up" is because they were close to physical limits...

I do agree that amd needs some more tweaking to get more out of their cores...the atlon core (k7) is just as old as the PIII core! I also do agree that the p4 will clock higher...

Sorry for the misunderstanding...


Proud owner of DOS 3.3 :smile:
a b à CPUs
August 14, 2003 4:14:06 AM

Easy to answer, LIKE INTEL, AMD IS RIPPING YOU OFF!!! Intel does this by holding back technology in order to extend developement cost over a longer period of time (thereby reducing anual development cost). This way they can overcharge you for the 3.2 currently, instead of selling the 3.2 at a reasonable cost and overcharging for, say, a 3.4.

AMD does this as well, only they're worse at the rip off game. The 3200+ can easily be overclocked by 10%. At that speed, it really IS competative with the P4 3.2C. But the problem is, it would mean their next CPU would also have to be equally better performing!

In defence of AMD, they have budgeting problems and investor problems that would prevent them from actually making money if they were honest!

So you have two companies, Intel and AMD, both ripping off their customers. Intel does it to boost profits, AMD does it to reduce losses. I'm sure I'll hear from a bunch of supporters from both companies telling me how terrible I am now.

<font color=blue>Only a place as big as the internet could be home to a hero as big as Crashman!</font color=blue>
<font color=red>Only a place as big as the internet could be home to an ego as large as Crashman's!</font color=red>
August 14, 2003 4:15:26 AM

the slocket and cpu are comming tommorrow

and the GPU is supposidly comming the 15th...but i hope FEDEX pulls a fast one and delivers it to me on the 14th...cause thats my birthday...and i desirve it!


Proud owner of DOS 3.3 :smile:
August 14, 2003 4:58:34 AM

I know they use town names, but at the time, it seemed so obvious it was implying copper technology.

I don't know if I had heard it was supposed to have it or not, but it should have IMO. It would've saved them a great deal of pain on the 1.13GHZ fiasco.

But I do imply that AMD's clock speed problems are due to physical constraints. You can't bump up 200MHZ on the AMD as easily as the Pentium 4 can do it. The Athlon has work almost twice harder for those 200MHZ, while the P4 has many units enjoying the slight increase, doing the task like a breeze.
They're not at their real physical limit, but they're getting closer more easily, and pretty soon, well, the K8 will have to give up the IPC charade and AMD will need to start focusing on clocks, ergo making a new core with an extended pipeline.

Crashman could cover up on when the P6 core began.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
August 14, 2003 5:18:11 AM

99.599999999 percent of the users don't even know copper tech will help the cpu...

99.9999 percent of the users don't even know why this effect occures (i am one of them...i am assuming less distortion therefore a higher signal strengh)...

So wheater or not intel implyed it or not...no one really knows or cares...

I totally agree that amd is architectually (spelling?) limited...and it is near it's physical limit...but not AT it...seeing how people can overclock these cpus pretty far..


Proud owner of DOS 3.3 :smile:
August 14, 2003 5:26:39 AM

Copper does help signals travel faster.

Think of how Aluminum cooling isn't better than copper in general. It's the very same concept. Athlon and Pentium III were both aluminum traced CPUs. Copper allowed the Athlon to top out at almost 1.8GHZ under 0.18m. Impressive compared to the PIII's ridiculous 1GHZ. In other words the Athlon can go as far as 80% a P3 with copper trace advantage. Pentium 4 introduced Copper straight into Wilamette I believe, but Wilamette's extremely inefficient process made it top out at 2GHZ, nearly what a 10 stage CPU like the K7 did. So the P4 at first had a very sad track record in process performance.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
August 14, 2003 5:30:22 AM

I forgot to add that adding layers generally helps a lot. That is the only way AMD could have made T-Bred work at high speeds and their 0.13m process. Regrettably 8 layers (IIRC) were not good enough, so as you can see, it's not always the process alone that can be a constraint in clock speeds, but tons of other things like layers. Once it was fixed, AMD just had to refine further its process, so that even 1.47GHZ Tbreds could reach 2.3GHZ easily. That wasn't due to the layers only, but thanks to a refined process.
AMD impressed a whole bunch with the Tbred comeback, once they added the layers. But I don't know if we had by then called them Tbred Bs. AFAIK Tbred Bs were the more recent ones with refined process, weren't they?

Or did I create an extra scenario? Hmm, now that I think about it, they were probably Tbred Bs, but we used that nomenclature more often once they switched all their line to Tbred Bs, from the XP1700+. I MUST be mixed lately. :eek: 

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
August 14, 2003 6:21:46 AM

EDEN! Pull it together man...unmix yourself!

Are these copper traces that they put in the pcb of the cpu? So the pins directly connect to the core via a copper traces...not Aluminium ones?


Proud owner of DOS 3.3 :smile:
August 15, 2003 12:48:53 AM

Well, I am not aware of any PCB inside the CPU, it's the traces connecting the layers and all. I am not professional in chemistry, but it seems to me copper traces on silicon will logically help current much more than aluminum.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
August 15, 2003 1:05:56 AM

Exactly! Those are copper interconnections. The pins connect to the outmost layer of them. And the innermost layer connects to all individual transistors.
For P4, there are 6 copper layers (same starting as early as P2 era); for athlon, it has evolved from 6 to 9 (barton). More layers you have, the more speedpaths you can arrange to increase operational frequency (like shortcut for signals), but at same time your manufacturing cost increases linearly and yield drops.
August 15, 2003 1:17:46 AM

hmm from reading all this seems like AMD is holding back a bit. I would think that pushing more near limit would be better for AMD so they can get more performance out of it...
August 15, 2003 2:14:12 AM

That does it. I'm not getting ripped off any more. I'm going to buy Cyrix chips from now on.
August 15, 2003 3:06:11 AM

They could, but heat issues would arise. Also that would require increasing voltage, and packing better heatsink fan combos in the Retail boxes.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
August 15, 2003 3:08:06 AM

Ah, that puts it better. I had forgotten copper layers and copper interconnects were pretty much the same.

Actually this might be what the P4 needs, to be able to linearly beat the Athlon by a 2x ratio in clock speed at the same process. It has 6 layers. At 9 layers as the Tbred had, it could significantly rise to 4GHZ easily. Intel has the ability to pay for it anyways, if AMD could.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
August 15, 2003 6:16:57 AM

thank you...

yes eden i was taling about the pcb under the PGA...

I don't think copper and transistors mix too well...


Proud owner of DOS 3.3 :smile:
August 15, 2003 4:07:26 PM

It is true copper can kill transistors, and that's why they cannot touch each other directly. Between 1st copper layer (Metal 1) and transistors, tungsten contact plugs are used to connect them electrically.
August 15, 2003 8:35:52 PM

They are still however inside the chip itself and not on the die (anything around the core or island), correct?

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
August 15, 2003 9:49:36 PM

Umm...

I am not processor expert...but the only use i see for copper in a cpu is to connect the pins to the core...other than that i just can't see why you would want an excellent conductor of electricity to be mixed with a semiconductor...it would defeat the purpose i would think...

However i am using logic and not knowledge...so if you have any ideas on what copper can do to help..i would be very interested...


Proud owner of DOS 3.3 :smile:
August 15, 2003 11:58:41 PM

Again, on P4 there are 6 layers of copper wires on the die - the 128mm2 piece of silicon - inside the chip package. They are used to connect those 55million transistors, route the signal, and eventually lead to 478 pins.
On Athlon, there are 7 Cu layers; on Palomino, there are 8; on Barton, there are 9.
August 16, 2003 1:14:26 AM

i relise that...

I just wanted to see what eden was thinking...thats all...


Proud owner of DOS 3.3 :smile:
August 16, 2003 2:11:28 AM

I too don't know much, but from the charts showing the layers, it seems it is all in the die, not the layer above the pins only. I've done a Science project on CPUs and had some silicon facts on how chips are made (Intel's website has a side that teaches students about it and the chemical background).

What you should know on base, is that the chip is not one flat surface, but a serie of many many others. I recall there were almost 200 layers during the production of the silicon at the fab, however I don't know if the copper layers are the real amount of layers.
What is sure, is that between layers and circuits lies the copper interconnection, which, logically, allows even smoother driving of electrons, and heat as well. Just look at the Athlon vs PIII at 0.18m.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.
!