Sign in with
Sign up | Sign in
Your question

correction: athlon FX != Opteron 1xx

Last response: in CPUs
Share
Anonymous
a b à CPUs
October 28, 2003 9:10:52 AM

I appears along with many other posters, I've been making a mistake assuming the Athlon FX==opteron 1xx.

Look at <A HREF="http://www.amd.com/us-en/Processors/ProductInformation/..." target="_new"> this FAQ on AMD's site</A>:
Quote:

What is the difference between the AMD Athlon 64 FX processor and the AMD Opteron™ processor for 1-way servers and workstations?

A:
While they both feature a 128-bit wide integrated memory controller with registered memory, the AMD Athlon 64 FX processor is designed for gamers and prosumers who need high-performance desktop systems. The AMD Opteron processor for 1-way servers and workstations is designed for businesses that need high-performance server solutions for demanding enterprise applications.

Additionally, the processors differ in that the AMD Opteron processor features three HyperTransport links, compared to the one HyperTransport link of the AMD Athlon FX processor. They are also tested to different electrical specifications.


So there you have it. The FX has only 1 HT link, the Opteron has 3 (non of them coherent though). The implications are minimal though. AFAICS, this would mean an Opteron 1xx should work in any FX board, but not necessarely vice verca. A single cpu opteron board that uses more than 1 HT link (like one for AGP, and another for PCI-X) would not work with an FX chip, but then, I've not seen any such board so far.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
October 28, 2003 5:12:11 PM

Geeze bbaeyens. I'm gone for a while and when I come back you're <i>still</i> struggling with this and <i>still</i> using bad information to do it too? I think someone needs to take a break.

As usual AMD is bending truth to incredible lengths. And as usual you're buying it up like it was handed down directly from god or something. Ask the questions. Do the research.

Quote:
<font color=green>While they both feature a 128-bit wide integrated memory controller with registered memory, the AMD Athlon 64 FX processor is designed for gamers and prosumers who need high-performance desktop systems. The AMD Opteron processor for 1-way servers and workstations is designed for businesses that need high-performance server solutions for demanding enterprise applications.</font color=green>

Isn't this an amazing answer? It gives absolutely no information on what those design differences are. Are there any differences, or are they the exact same design? The answer is marketing. <i>That</i> is the only difference.

Quote:
<font color=green>Additionally, the processors differ in that the AMD Opteron processor features three HyperTransport links, compared to the one HyperTransport link of the AMD Athlon FX processor.</font color=green>

I wish I could say that I was surprised at how AMD bends the truth here, but I'm not surprised at all.

First of all there's a gross twisting of the facts. The K8s have two different types of HT links. One type is specifically for I/O (a slower communication to the northbridge that enables AGP, PCI/PCI-X, etc.). The other type is specifically for inter-CPU communication (enabling two, four, and eight way configurations).

<i>All</i> of the K8s have the I/O HT link. Without that the CPU would be worthless. As for which Opterons have the inter-CPU links, look at <A HREF="http://www.amd.com/us-en/Processors/ProductInformation/..." target="_new">AMD's own page on "Understanding AMD Opteron™ Processor Model Numbers"</A> to see the truth. Opteron 1xx has none of the inter-CPU HT links enabled. Opteron 2xx has only one inter-CPU HT link enabled. Opteron 8xx has all three inter-CPU HT links enabled. (It makes you wonder why there isn't an Opteron 4xx with two HT links enabled.)

So that makes the Opteron 1xx <i>exactly</i> the same as the A64FX.

This however isn't entirely true either. Initial reviews of the A64FX revealed that the inter-CPU HT links of the A64FX weren't even disabled, making the A64FXs that were reviewed equivalent to Opteron 8xxs. Maybe AMD has fixed this in their final production runs so that the A64FXs are only identical to the Opteron 1xx, but so far no word on such has been stated.

Either way though, AMD's answer is grossly warped. The A64FX should have one I/O HT link and <i>no</i> CPU HT links. The Opteron 1xx has one I/O HT link and no CPU HT links. The Opteron 2xx has one I/O HT link and one CPU HT link. The Opteron 8xx has one I/O HT link and three CPU HT links. Not all Opterons have the three CPU HT links enabled. And A64FX should have no CPU HT links enabled.

Quote:
<font color=green>They are also tested to different electrical specifications.</font color=green>

Here is the first and <i>only</i> truthful bit of information from AMD. The A64FX and the Opteron 1xx are tested differently. That's it. No hardware differences at all, just a difference in how they're tested.

Quote:
So there you have it. The FX has only 1 HT link, the Opteron has 3 (non of them coherent though). The implications are minimal though. AFAICS, this would mean an Opteron 1xx should work in any FX board, but not necessarely vice verca. A single cpu opteron board that uses more than 1 HT link (like one for AGP, and another for PCI-X) would not work with an FX chip, but then, I've not seen any such board so far.

So there you have it. You not only took AMD's lies in hook, line, and sinker like a good little fish, but you also went on to speculate invalid suppositions based on these twisted lies and partial-truths. Congrats. You're everything that AMD wants you to be. You'll make the perfect AMD customer.

The reality is that they <i>all</i> use the same slower single I/O HT link for AGP, PCI, etc. They <i>all</i> use the same single-CPU mobos. There is <i>no such thing</i> as an Opteron 1xx mobo that uses more than the one I/O HT link because there is no Opteron in existence that has more than one I/O HT link. The additional links in the Opteron 2xx and Opteron 4xx are purely for transfering data between the CPUs in a multi-CPU platform. And the Opteron 1xx doesn't even have any of those CPU HT links enabled.

<A HREF="http://ars.userfriendly.org/cartoons/?id=20031017" target="_new">Then what's your poison of choice?
Soymilk. I was raised on the stuff.
Ah. What's it made of?
Soy.
Is that soy as in soylent green?</A>
Anonymous
a b à CPUs
October 28, 2003 6:16:42 PM

>Geeze bbaeyens. I'm gone for a while and when I come back
>you're still struggling with this

No, not really. Its just I always assumed the FX was identical to the opteron 1xx, pretty much like you claimed but I've recently been pointed to information that states otherwise.

The fact that opteron 8xx, 2xx, 1xx and the Athlon FX share the same core, and wouldnt look any different under a microscope is not the issue here; they are all identical, and when still on the wafer, there is no difference between any of them. I'm not even sure if the Athlon 64 is any different, it could be, but I'm not sure. However...

>Isn't this an amazing answer? It gives absolutely no
>information on what those design differences are. Are there
>any differences, or are they the exact same design? The
>answer is marketing. That is the only difference.

Marketing, validation and microcode and/or lasering. I assume HT links are enabled or disabled through either microcode or lasering; the same for coherent HT links and non coherent links. I agree/assume all those chips are the same before binsplitting and microcoding.

>First of all there's a gross twisting of the facts. The K8s
>have two different types of HT links. One type is
>specifically for I/O (a slower communication to the
>northbridge that enables AGP, PCI/PCI-X, etc.). The other
>type is specifically for inter-CPU communication (enabling >two, four, and eight way configurations).

correct. The right terminology is (cache) coherent links. Opteron 8xx has three of them, 2xx 1 and the 1xx none. As you pointed out, coherent links can be used as interconnects between different cpu's whereas non coherent ones can be used for I/O only.

>So that makes the Opteron 1xx exactly the same as the
>A64FX.

This is not true however. Yes, they share the same die, but then I'm pretty sure any opteron/FX has the exact same die. thing is, the opteron 1xx has 3 non coherent HT links. Few if any motherboards make use if this, but it would be perfectly possible to create a motherboard that connects the AGP tunnel to one, and eg, two PCI-X bridges to each of the other HT links.

The FX on the other hand, per AMD's data, does only have 1 HT link (enabled). It could have three ondie, and two of them disabled through microcode/lasering for all I know, but AMD does not guarantee you get 3 functional HT links. In fact, I assume it only has 1 working HT link. If someone brings out an opteron 1xx board with 10 PCI slots, it will work with an opteron 1xx, but most likely not with a FX.

Bottom line; if you claim the FX is <i>identical</i> to the 1xx opteron, you may have a point, but then its also identical to the 2xx and 8xx. The only difference is applied afterwards, by disabeling (coherent) HT links. The fact AMD states the FX has only *1* non coherent HT link, while the 1xx has three, is a difference to me. Sure, same core and all that, but a functional difference nevertheless.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Related resources
October 28, 2003 10:44:29 PM

Thanks for the info bbaeyens. I have been wondering about that for a while.

I am thinking of getting the 240 Opteron, this should work in the Asus SK8N single cpu motherboard, right?
Anonymous
a b à CPUs
October 28, 2003 10:53:30 PM

yet it works just fine on that MB, at least thats what ASUS tells us :) 

Just curious, what would you buy it for ? the 240 is only 1.4 GHz, the 1.8 GHz 244 is "only" $100 more expensive on pricewatch..

= The views stated herein are my personal views, and not necessarily the views of my wife. =
October 28, 2003 11:10:19 PM

Over in the UK the 244 is $330 (USD) more expensive than the 240 :(  I think you mean $100 extra for the 144 @ 1.8GHz.

I probably should go for the 144 but I was hoping to upgrade to a dual system once the cheap dual cpu motherboards (with AGP) come out.

Hmmmmm.....
October 29, 2003 1:47:36 PM

Quote:
No, not really. Its just I always assumed the FX was identical to the opteron 1xx, pretty much like you claimed but I've recently been pointed to information that states otherwise.

No offense, but if that's the culmination of all of your "information that states otherwise" then it's hardly conclusive.

Quote:
Marketing, validation and microcode and/or lasering. I assume HT links are enabled or disabled through either microcode or lasering; the same for coherent HT links and non coherent links. I agree/assume all those chips are the same before binsplitting and microcoding.

My guess would be that a lot of those coherent HT links are disabled using lasering because they failed validation at some point. Downbin the CPU to lower multi-processor capabilities much in the same way that CPUs are downbinned according to clockspeed validations. Obviously some 'good' ones are downbinned too in order to meet demand, but that's more or less how I figure it's done. It'd be interesting to know if AMD started using microcode instead of lasering though. But in the end it's really just all about marketing. It's the same CPU, just at some point after production someone came along and castrated them to various degrees. It's no different setting a clockspeed. It's still the same chip.

Quote:
the opteron 1xx has 3 non coherent HT links. Few if any motherboards make use if this, but it would be perfectly possible to create a motherboard that connects the AGP tunnel to one, and eg, two PCI-X bridges to each of the other HT links.

The FX on the other hand, per AMD's data, does only have 1 HT link (enabled). It could have three ondie, and two of them disabled through microcode/lasering for all I know, but AMD does not guarantee you get 3 functional HT links. In fact, I assume it only has 1 working HT link. If someone brings out an opteron 1xx board with 10 PCI slots, it will work with an opteron 1xx, but most likely not with a FX

If you have any proof whatsoever of this I'd love to see it. So far it's all just conclusions based on shoddy information from AMD as far as I can tell. I haven't read a single thing yet that states that Opterons are guaranteed to work with more than one I/O HT link, nor have I read anything that definitively states that the A64FX has one and only one.

Quote:
Bottom line; if you claim the FX is identical to the 1xx opteron, you may have a point, but then its also identical to the 2xx and 8xx. The only difference is applied afterwards, by disabeling (coherent) HT links. The fact AMD states the FX has only *1* non coherent HT link, while the 1xx has three, is a difference to me. Sure, same core and all that, but a functional difference nevertheless.

If you can actually prove that all of the the A64FXs released have one and only one non-coherent HT link and that all Opterons released have three non-coherent HT links that <i>all</i> actually work then you have a point. In so far as there is no hardware available to even prove any of this yet though, that point is pretty meaningless.

So in the future it may be a valid point. But then in the future the A64FX will also be Socket939 instead of Socket940. And I'd be willing to bet half of my soul that <i>if</i> the A64FX actually has hard specs to adhere to, it'll do so only for Socket939 and until then the Socket940 A64FX is identical to an Opteron1xx in every way but name and specs on paper (but not in practice).

Two final thoughts:
1) Doesn't it bug you of the terminology that AMD is using? Coherent? Shouldn't the opposite of that be <i>in</i>coherent? Why would anyone want an incoherent link in their processor? ;) 

2) Doesn't it bug you even more how little information is actually available on the <i>exact</i> details of the A64FX and the Opteron and what their differences <i>really</i> are, even if it's just paper differences and not actually adhered to by the hardware?

<A HREF="http://ars.userfriendly.org/cartoons/?id=20031017" target="_new">Then what's your poison of choice?
Soymilk. I was raised on the stuff.
Ah. What's it made of?
Soy.
Is that soy as in soylent green?</A>
October 29, 2003 2:48:15 PM

Quote:
Few if any motherboards make use if this, but it would be perfectly possible to create a motherboard that connects the AGP tunnel to one, and eg, two PCI-X bridges to each of the other HT links.

If this is true it gives a little credence to the whole desktop vs workstation argument. Should be easy enought ot test as well. As long as a 1xx proccesor will work in a dual CPU mobo, get a motherboard with a few PCI-X slots and see if the slots work. Then try a A64FX.

It's not what they tell you, its what they don't tell you!
October 29, 2003 3:05:16 PM

So, basically, it's a opt 1 that fails the hypertransport link test... basically a form of speedbin'n or an intentional cut like some of the celery chips cache.

Shadus
October 29, 2003 3:46:15 PM

I have to admit the HT system really is interesting, in how flexible it is and how it standardizes the communication system.
I wish we could have hardware that actually DEMANDS the HT bandwidth to get better results. (aside from inter-CPU communications)

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>This just in, over 56 no-lifers have their pics up on THGC's Photo Album! </b></font color=blue></A> :lol: 
Anonymous
a b à CPUs
October 29, 2003 5:08:18 PM

I think we are arguing on terminology here. When do you consider 2 cpu's to be identical ? When you can not see a difference when looking at the die under a "microscope", or when it exhibits an identical featureset ? With the first definition, I agree opteron 1xx== Athlon FX (minus a lasering/microcoding job).

But then you'd also have to say Athlon FX==Opteron 8xx and P4EE==Xeon MP and P4==Celeron. Difference between those are no more than a lasering/microcode job (and repackaging in the case of the EE).

If you use the second definition, then I can't see how you could still claim the FX==opteron. I mean, if anyone knows, wouldnt it be AMD ? There is nothing shoddy about their information, its plain simple: the FX only has 1 HT link, and the opteron has 3. Why would i need to prove that ? Can you prove me the Celeron doesnt have 512 Kb cache ? Doesnt support 533 or 800 MHz FSBs ? Doesnt have hyperthreading circuitry onboard ? Or do you honestly believe the opteron 1xx does not sport 3 HT links, and AMD is just lying about it ? I'm confused.

>If you can actually prove that all of the the A64FXs
>released have one

AMD says they only have one. 2 could be disabled, or maybe just not tested, but AMD delivers you a product with ONE working HT link. What is so ambigious about that ?

>and that all Opterons released have three non-coherent HT
>links that all actually work

Well, assuming you mean opteron 1xx here, I kinda trust AMD to release a product that lives up to its basic featureset as outlined in their datasheets. You don't really believe they are lying about this, are you ?

>In so far as there is no hardware available to even prove
>any of this yet though, that point is pretty meaningless.

There are hardly any dual opteron boards that I know off, that use more than 1 HT link for I/O. Does that prove the 2xx may also only have 1 coherent and 1 incohorent HT link ?

>My guess would be that a lot of those coherent HT links are
>disabled using lasering because they failed validation at
>some point.

I doubt it; HT links are pretty small; how small would the odds be of a die sporting a defect just in one of them ? Even worse, how small would the odds be that a HT link would not work as a coherent link, but would work as a non coherent link ? pretty close to zero IMHO. I'm pretty sure 99.9% of the opteron 2xx's could just as well have been sold as a 8xx and at least 99% of the 1xx could have been a 2xx or 8xx; pretty much like I assume hardly any P4 die is salvaged as a Celeron just because the hyperthreading part didnt work. Its a possibility, but extremely unlikely.

>1) Doesn't it bug you of the terminology that AMD is using?
>Coherent?

(cache)Coherent is just the name. AMD didnt invent it, its a standard term that has been used for decades.

>2) Doesn't it bug you even more how little information is
>actually available on the exact details of the A64FX and
>the Opteron and what their differences really are

No, not really. What else should we know ? Same die, different number of HT links, different clock speed; what more is there to know ? What bugs me though, is that we still havent seen any precise power consumption numbers.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
October 29, 2003 6:54:00 PM

Quote:
When do you consider 2 cpu's to be identical ?

When they have the exact same features straight out of the box.

Quote:
If you use the second definition, then I can't see how you could still claim the FX==opteron. I mean, if anyone knows, wouldnt it be AMD ?

And <i>that's</i> my point, that it's AMD that is purposefully clouding the facts. I don't trust AMD on this. <i>That</i> is my point.

Quote:
Well, assuming you mean opteron 1xx here, I kinda trust AMD to release a product that lives up to its basic featureset as outlined in their datasheets. You don't really believe they are lying about this, are you ?

Lying? Gee, it's not like AMD has <i>never</i> lied before. :\ Come on. What kind of a question is that? Only a fool believes the marketing of a product. That's what independent research and testing is done for, to clear up the truth about products that the companies have been lying about.

And in this case AMD isn't even giving enough clear information to determine if they are or not. It's all been rather sad as far as technical information goes. And that's the point. The exact details of the A64FX are being obfuscated.

Further all that we have to go by is whatever AMD puts down on paper, which in this case may or may not be the same as what comes off of the production line. I have to wonder why no one is providing that detailed information that proves if the hardware implementation is or isn't the exact same.

Quote:
AMD says they only have one. 2 could be disabled, or maybe just not tested, but AMD delivers you a product with ONE working HT link. What is so ambigious about that ?

That depends entirely on if the other two also work or not. If they haven't even been disabled and do work then how would that be any different from an Opteron then? It's easy to say on paper that it has only one when there's no hardware out there to prove otherwise.

Quote:
I doubt it; HT links are pretty small; how small would the odds be of a die sporting a defect just in one of them ?

I've no idea. Just because they're small doesn't mean that they're unlikely to perform up to spec.

Quote:
Even worse, how small would the odds be that a HT link would not work as a coherent link, but would work as a non coherent link ? pretty close to zero IMHO.

Actually I'd say that the odds are pretty high. The bandwidth difference between the two is rather large. So a coherent link that couldn't perform up to spec could still easily perfom up to non coherent specs.

But that however isn't even the point. The point is that if it didn't perform up to spec it could be disabled entirely. Zero to three coherent <i>and</i> always three non coherent. If any of the coherent fail, disable it and downbin it's multiprocessor level.

Quote:
P4 die is salvaged as a Celeron just because the hyperthreading part didnt work.

No, but a lot of them are salvaged because of cache failures.

Quote:
(cache)Coherent is just the name. AMD didnt invent it, its a standard term that has been used for decades.

Used for decades or not, it's still a funny term to use in my opinion.

Quote:
No, not really. What else should we know ? Same die, different number of HT links, different clock speed; what more is there to know ?

Good data sheets would be nice. Clear and consise details would be nice. And of course verification by independent third parties would be invaluable.

Quote:
What bugs me though, is that we still havent seen any precise power consumption numbers.

If that's <i>all</i> that bugs you about AMD's technical documentaion of both the A64FX and the Opteron then you're a much more forgiving person than I.

<pre><b><font color=red>I've always wondered why people liken the taste of blood to copper.
It tastes much more like iron to me.
<- :evil:  - :evil:  - :evil:  - :evil:  - :evil:  -></font color=red></b></pre><p>
Anonymous
a b à CPUs
October 29, 2003 7:22:30 PM

>When they have the exact same features straight out of the
>box.

Well, since its kinda hard to determine if the FX has 1 or 3 incoherent HT links, its pretty much like the MP versus XP. Maybe the FX is capable of more than what AMD promises, and indeed all 3 HT links are functional in some (all ?) cases. so ?

>Lying? Gee, it's not like AMD has never lied before. :\

*cough*.. they may deliver more than they promise (like 3 HT links, when they only promise 1), but seriously, you don't expect them to deliver less ?? no f*cking way. It would be as bad as intel selling a P4 with 512 kb cache, while only 348 Kb works; no way in hell.

>Actually I'd say that the odds are pretty high. The
>bandwidth difference between the two is rather large.

I thought bandwith had nothing to do with it, just capability (of synchronising caches). AFAIK, coherent inter cpu links are just as fast/slow as I/O links. They both work at 800 MHz.

>No, but a lot of them are salvaged because of cache
>failures.

Right. but then, the cache of a P4 is what ? 30-50% of a P4's die ? HT links are maybe 5%. Celeron versus P4 unit sales are probably roughly 50%, wheres as 8xx opteron are maybe 5% of 1/2xx sales. See the difference ?

>Good data sheets would be nice. Clear and consise details
>would be nice. And of course verification by independent >third parties would be invaluable.

Verification of what ? I don't need verification of the fact that opteron lives up to what AMD claims. They may be stupid, but not THAT stupid. Do you want verification that some chips deliver *more* than what is promised ? Why would that matter ? Its not like we have independant verification that some Celerons have a fully functional L2 cache, that HTT works, and these chips can run faster than what they are specced at.

>If that's all that bugs you about AMD's technical
>documentaion of both the A64FX and the Opteron then you're
>a much more forgiving person than I.

Not really. I just want to know what I'm sure my product will achieve. If its more, than so much the better, but I don't expect the manufacturer to promise me that, or a third party to confirm such a suspicion.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
October 29, 2003 7:52:08 PM

Quote:
Well, since its kinda hard to determine if the FX has 1 or 3 incoherent HT links, its pretty much like the MP versus XP. Maybe the FX is capable of more than what AMD promises, and indeed all 3 HT links are functional in some (all ?) cases. so ?

So? That's the entire point of this discussion. Is or is not the A64FX exactly the same as the Opteron1xx? If the A64FX could do more than what AMD promises then they are the exact same chip, just sold under different names. Does A64FX == Opteron1xx? That's what inquiring minds want to know.

Quote:
*cough*.. they may deliver more than they promise (like 3 HT links, when they only promise 1), but seriously, you don't expect them to deliver less ?? no f*cking way. It would be as bad as intel selling a P4 with 512 kb cache, while only 348 Kb works; no way in hell.

First of all I more lean towards the idea that the A64FX delivers more than it is meant to on paper. Second of all how would anyone even know if an Opteron only delivered one? How could anyone miss a feature that isn't even used?

Quote:
I thought bandwith had nothing to do with it, just capability (of synchronising caches). AFAIK, coherent inter cpu links are just as fast/slow as I/O links. They both work at 800 MHz.

That's the confusing part. Earlier at Opteron's debut what I'd read indicated both a clock speed <i>and</i> a data path size difference. Yet all that I can find from AMD <i>now</i> is specs for only one path type. Obfuscation? At the very least they're causing unintentional confusion. It's only a matter of intent now, not of resut.

Quote:
Right. but then, the cache of a P4 is what ? 30-50% of a P4's die ? HT links are maybe 5%. Celeron versus P4 unit sales are probably roughly 50%, wheres as 8xx opteron are maybe 5% of 1/2xx sales. See the difference ?

When speed and path noise are involved it isn't just a matter of flaws according to space but also according to performance and stability.

Quote:
Do you want verification that some chips deliver *more* than what is promised ? Why would that matter ?

That's exactly what I want and I already said why it would matter. That's the whole point. It would prove if A64FX == Opteron1xx or not.

Quote:
Its not like we have independant verification that some Celerons have a fully functional L2 cache, that HTT works, and these chips can run faster than what they are specced at.

Funny, it's not like CPUID or any other software exists to display the size of the L2 cache. And certainly no article has ever been written on if chips can run faster than their specs. I mean that'd be what, overclocking? Gosh. Nothing like that exists in the whole world.

Quote:
Not really. I just want to know what I'm sure my product will achieve. If its more, than so much the better, but I don't expect the manufacturer to promise me that, or a third party to confirm such a suspicion.

I'm not arguing on if it's more, so much the better. But if it is more then one has to question how the A64 is in any way <i>not</i> a worstation/server CPU. Further I do expect a third party to confirm such suspicions. That's what they do. If they don't then they're just marketing parots and/or wasted space.

<pre><b><font color=red>I've always wondered why people liken the taste of blood to copper.
It tastes much more like iron to me.
<- :evil:  - :evil:  - :evil:  - :evil:  - :evil:  -></font color=red></b></pre><p>
Anonymous
a b à CPUs
October 29, 2003 8:00:31 PM

basically what you are wondering about is wether or not AMD disabled two HT link or not on the FX. I mean who cares ? If they disabled them, would that validate the FX as a deskop chip, and if they did not bother, would it invalidate the FX as a desktop chip ? I really don't see what you are trying to prove here. Yes, any FX could most likely be sold as an opteron 1xx as well, without lasering (or not ) the 2 HT link, so ?? Some rumours seem to indicate that intel has not taken he necessary steps to disable the SMP capabilities of the P4EE. So what ? does that make the EE a server chip? Just like in the FX case, there are no motherboads to prove this (ie dual cpu P4 EE motherboards), so f*cking what ??

= The views stated herein are my personal views, and not necessarily the views of my wife. =
October 29, 2003 10:53:34 PM

Whether it is marketing or a technical difference. The difference is there and something to be aware of when making purchasing decisions.
October 30, 2003 12:52:35 PM

Quote:
basically what you are wondering about is wether or not AMD disabled two HT link or not on the FX.

That's one of a few things, yes. I'm also wondering if there are any electrical/thermal differences as well. Is or isn't there <i>any</i> differentiating factor between the A64FX and the Opteron 1xx? That's what I want to know. So far all of the physical evidence says no. But if there was even one single provable difference then it would in fact give even a tiny bit of credence to A64FX being desktop instead of workstation/server.

Quote:
I mean who cares ?

Well obviously I do. Duh. ;) 

Quote:
If they disabled them, would that validate the FX as a deskop chip

Validate it? I can't really say. It would certainly make it a <i>much</i> stronger case for the A64FX being a desktop chip debate though.

Quote:
I really don't see what you are trying to prove here.

I've made it more than clear. So if you really don't see it then just stop debating against it already. :p 

Quote:
Some rumours seem to indicate that intel has not taken he necessary steps to disable the SMP capabilities of the P4EE.

They don't have to. Without the pins required for it (the only difference between a P4's pin array and a Xeons pin array) then there's no reason to disable it on chip. It's made impossible to use by its packaging. So what's your point?

Quote:
Just like in the FX case, there are no motherboads to prove this (ie dual cpu P4 EE motherboards), so f*cking what ??

There are no dual P4EE motherboards because such a thing is physically impossible thanks to the difference in packaging between a P4 and a Xeon. It is however entirely possible to design a motherboard which uses the non-coherent links of an Opteron because the pins <i>are</i> there. The fact that you even stick to such incredibly faulty logic in order to try and prove your points means that you have absolutely no ability to debate this anymore. Until you can prove otherwise I'm not even going to bother replying to another of your posts because it'll just be a waste of my time.

<pre><b><font color=red>I've always wondered why people liken the taste of blood to copper.
It tastes much more like iron to me.
<- :evil:  - :evil:  - :evil:  - :evil:  - :evil:  -></font color=red></b></pre><p>
October 30, 2003 12:56:31 PM

Quote:
Whether it is marketing or a technical difference. The difference is there and something to be aware of when making purchasing decisions.

That's entirely my point though. If the difference is not technical at all and is in fact just pure marketing then it can make for a very different perspective on purchasing decisions than if it is a technical difference.

Until we know the whole picture how can we make an educated purchasing decision? And why is AMD making that educated purchasing decision so difficult to make?

<pre><b><font color=red>I've always wondered why people liken the taste of blood to copper.
It tastes much more like iron to me.
<- :evil:  - :evil:  - :evil:  - :evil:  - :evil:  -></font color=red></b></pre><p>
Anonymous
a b à CPUs
October 30, 2003 1:26:57 PM

>here are no dual P4EE motherboards because such a thing is
>physically impossible thanks to the difference in packaging

B*llshit. Its perfectly possible to have a dual socket 478 board. There is nothing in the design of the socket that prohibits this, just like there was nothing that stopped socket 370 from going dual cpu. However, intel claims the p4 cpu is incapable of SMP operation. This could be true or not, no way of verifying it, but there is nothing in the socket design that would make this somehow "physically" impossible. The socket just provides power and a connection to the chipset, and with an SMP capable chipset and SMP capable socket 478 cpu, there is nothing to stop you from making a dual 478 board, if not probably non available cpu's.

>fact that you even stick to such incredibly faulty logic in
>order to try and prove your points means that you have
>absolutely no ability to debate this anymore.

I've been accused by you of faulty logic times and times again; and I don't think so far I have ever been wrong or proven wrong, while mostly proving my logic and statements where correct, so I suggest you tone it down a bit.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
October 30, 2003 2:54:43 PM

Quote:
B*llshit. Its perfectly possible to have a dual socket 478 board. There is nothing in the design of the socket that prohibits this

Oh, I suppose then that the information between two CPUs just happens to travel along the pins that aren't there by magic then. Tell me exactly how this data gets shared without the pins being present and I'll grant you a point. Come on. Prove me wrong if you can.

Quote:
The socket just provides power and a connection to the chipset

Wrong, wrong, wrong! The socket provides connectors for data paths. If data paths are not present in the socket then those data paths cannot be utilized by the CPU. If the data paths cannot be utilized by the CPU then there is no way to enable that feature of the CPU. This isn't a serial connection where data is just streamed with tags to determine what data goes where. This is a parallel connection where the presence of specific data paths matter because the appropriate data is <i>only</i> sent along the approrpriate paths.

Quote:
I've been accused by you of faulty logic times and times again; and I don't think so far I have ever been wrong or proven wrong, while mostly proving my logic and statements where correct, so I suggest you tone it down a bit.

Get over yourself, bbaeyens. You <i>have</i> been proven wrong. You just <i>were</i> proven wrong yet again. You <i>don't</i> know what you're talking about. You <i>are</i> using faulty logic. If you could for one minute <i>prove</i> that you know what you're saying then I would tone it down. But over and over and over again you only prove otherwise. And I'm sick of it. <i>Someone</i> has to tell you because <i>clearly</i> you refuse to see it. I just happen to care so little about what you think of me to have no reason not to tell you.

<pre><b><font color=red>I've always wondered why people liken the taste of blood to copper.
It tastes much more like iron to me.
<- :evil:  - :evil:  - :evil:  - :evil:  - :evil:  -></font color=red></b></pre><p>
October 30, 2003 7:04:38 PM

Yo guys/gals slow your roll. I don't post much on this site but this discussion is basically moot. It seems as though, to some degree, it was just created to start a flame war.

This discussion is basically going in circles. I'll see if I can at least start a meaningful dialogue regarding this. But to some degree arguing over which single star in the universe is more important than another is futile.

So anyway, here goes....

From what I can gather you guys are arguing whether an AFX and an Opt 1xxx are the same. This is a no brainer. For all intensive purposes they are. The differences come in verification, CPUID, and a couple of other small nuances. However, these differences are a BIG deal.

You have to remember the K8 design from the start was to be a mission-critical server design - supporting 3 HT links. The Opt 8xx support all 3, 2xx support 2 and the 1xx supports essentially 1. Intel does this a little bit differently; the Pentium 4 is the holy grail of features and bus speeds, while the server line is more of a subset of the P4 with less features, and lower bus speeds. The caveat of the server line? - CACHE and lots of it (plus enough testing to make your eyes hurt and your kidneys fail).

In a server/workstation environment the Opteron processors need to work right and for along time until the particular hardware is obsolete. AFX is a gamer CPU sure the architecture is the same, but if you think for a moment that I would bet that an AFX is going to out-last an Opteron class you're crazy. This particular scenario is exactly the decisions that companies have to make. AMD and Intel spend a lot of money to make sure the odds of losing this bet are as close to zero as possible. If I was an IT admin / dealt with network topographies for say Verizon, MCI, or Digex would you really want to put your (100K a year )job on the line by putting an AFX in a 2-way server / a workstation scenario just to test a theory? - I wouldn't!


As I said before, AMD and Intel handle it a little bit differently but the result is the same. Really what you guys are arguing is the different philosophies of marketing different processor lines. AMD goes server line down. Intel goes P4 up to Xeon and down to Celeron. You can tell this by how the processors are rolled out i.e. which comes first.

Since AMD does this of course the differences between AFX and Opt 1xx are little. They are the same architecture and they ride in the same socket. AFX pretty much equals P4EE they have the same purpose market-wise. You didn't think Intel nor AMD didn't know what the other was doing did you?

So to end my point what I am saying is that you are both right and wrong at the same time. For a typical end user the differences between the AFX and Opt 1xx (right now before the socket change) are virtually nil (there are some, but not enough to kill your mother over). However, for a company with mission critical assets and services they couldn't be more different.

Sorry if I rambled, I just don't like to see bickering needlessly.
October 30, 2003 7:27:32 PM

But kc1905 can you show any proof of a difference other than marketing? Right now I can't even find proof of a validation difference not to mention any actual hardware difference.

The thing is it comes down to questions like these:
1) Why is AMD being so neglectful about providing information on the A64FX?
2) Once Socket939 comes out will any Socket940 A64FX owner be able to upgrade their system with anything other than an Opteron 1xx?
3) Why does A64FX get to be benchmarked against P4, P4EE, AXP, and A64 but the Opteron 1xx does not?

It's not a moot discussion. It's a quest for the answers that AMD isn't giving us.

<pre><b><font color=red>I've always wondered why people liken the taste of blood to copper.
It tastes much more like iron to me.
<- :evil:  - :evil:  - :evil:  - :evil:  - :evil:  -></font color=red></b></pre><p>
October 30, 2003 8:05:03 PM

Ah, try not to confuse technical ability /technical decisions made by engineers with business matters. I've worked in the business for years (testing GPU, CPU, software and more) and some of the decisions I've seen would make your head hurt. Anyway, on to your points.

1) Why is AMD being so neglectful about providing information on the A64FX?

>This is a business decision, I wouldn't say they are being neglectful about it, but marketing-wise what company would shout from the roof tops "Hey guess what Joe-six-pack you can over clock your Opt 1xx and reach AFX speeds with a pretty good return on that bet, let me show you how"? No company that I can think of would do that not Nvidia, ATI, or any other company for that matter. Remember we are enthusiasts, with above average technical knowledge! The other side of this regards competition with Intel. Both of them play games in releasing temp attributes, core steppings, socket placement, etc. It's like showing a deck of cards to the competition and then making them play 52- pickup and have them put them all back in order to find out what the company was doing. As enthusiasts it is just more visible to us than Joe-six-pack.

2) Once Socket939 comes out will any Socket940 A64FX owner be able to upgrade their system with anything other than an Opteron 1xx?

> Yes. For how long? I don't know. Its just rumor at this point but most people are saying AFX followers should see at least two speed bumps after the switch on Socket 940. Is that messed up?? Maybe although who said being the first to the line to buy the fastest hardware would be cheap or completely economical?

3) Why does A64FX get to be benchmarked against P4, P4EE, AXP, and A64 but the Opteron 1xx does not?

>That's almost completely up to the reviewing web site. They have to place all of the processor, GPU, and software releases into some category. They usually follow what the companies tell them. If they feel that what they are hearing is completely bogus they may mention it, they may not. It depends, if I was reviewing a piece of hardware and it was due to be posted by 5AM on the web I wouldn't spend too much time talking about how the product was marketed. I would focus on the performance, some back-story of the design, and maybe hit up some additional information my listeners were looking for, but other than that I would want my article to be ready before my boss walked in the next day.

I hope I answered your questions fully.
Anonymous
a b à CPUs
October 30, 2003 8:34:19 PM

>Oh, I suppose then that the information between two CPUs
>just happens to travel along the pins that aren't there by
>magic then.

Exactly which pins arent there ? Your assumption you'd need more pins to enable SMP is based on nothing, except your own assumption. The P2/3 didnt need more pins, neither did the Athlon, the pentium pro or even the original pentium. Nor does the G5 or any other CPU that i'm aware off. Though my memory might be fading, I think you even had to *disable* a pin on the P3 Celeron to get it working in SMP.

>Wrong, wrong, wrong! The socket provides connectors for
>data paths....

...to the chipset. Xeons don't talk to each other, they talk to the chipset. Again you are assuming certain pins are missing, but have you got anything to back that up ? Are you aware that VIA's PX333 refence board had a jumper to enable SMP on a socket 423 board ? They never released a dual cpu socket 423 board, but that doesnt prove its impossible. I see no evidence whatsoever this is not perfectly possible.


= The views stated herein are my personal views, and not necessarily the views of my wife. =
!