Sign in with
Sign up | Sign in
Your question

Athlon 64 vs Opteron

Tags:
  • CPUs
  • Opteron
Last response: in CPUs
Share
May 6, 2005 1:56:53 PM

What exactly is the difference between an Athlon 64 and the Opteron. Why can you double (or quadruple) Opterons on a MB but not with the regular 64's?

More about : athlon opteron

May 6, 2005 2:47:41 PM

No difference really. They come in a 940-pin package instead of the 939/754-pin one that A64s come in, but the core is essentially the same - All Opterons have 1Mb Cache (IIRC) though, whereas A64s vary depending on the exact core being used.

By doing it this way you can charge more for the SMP-validated chips like the Opteron, as people buying big multiprocessor servers care less about cost of the CPU than your average joe does.

Intel does the same with P4s Vs Xeons. Same core (more cache though, for the most part).

---
<font color=red>"Life is <i>not</i> like a box of chocolates. It's more like a jar of jalapeńos - what you do today might burn your a<b></b>ss tommorrow."
May 6, 2005 4:03:29 PM

Except that the Opterons have more HT channels specifically for working with multiple processors.

<pre><font color=purple><i>Jesters do oft prove prophets.</i> -Regan in
King Lear (Act V, Scene iii) by William Shakespear</font color=purple></pre><p>@ 187K -> 200,000 miles or bust!
Related resources
May 6, 2005 4:36:33 PM

<pre>that too...</pre><p>I had a little voice inside my head trying to remind me of something like that, but I foolishly ignored it...

---
<font color=red>"Life is <i>not</i> like a box of chocolates. It's more like a jar of jalapeńos - what you do today might burn your a<b></b>ss tommorrow."
May 6, 2005 5:42:48 PM

>Why can you double (or quadruple) Opterons on a MB but not
>with the regular 64's?

Athlon 64s and OPteron 1xx are basically identical twins. One pin difference, and (unless I'm mistaken) opteron requires registered ram, whereas A64 gives you the choice, it will support either registered or unregistered. Neither chip will work in anything but a single CPU config.

Opteron 2xx and 8xx add 1 resp 3 cache coherent HT link to enable 2 way and >2 way SMP. Actually, those links are also present in Opteron 1xx and most likely even in A64, but they are either disabled or the cache coherency has been disabled. This is called market segmentation, its not good business selling $100 Athlon 64's that would work in 8 way servers that cost $60.000.

Intel does pretty much the exact same thing btw, there is precious little difference between Pentium 4s, Xeons (2 way) and Xeon MPs (>2 way) other than packaging, price, and in some cases, cache configuration. But for the exact same reasons, intel disabled SMP functionality in P4, and restricts Xeon to 2 way.

Segmentation is also done downwards, by disabling parts of the cpu or cache, A64s and P4s become cheap Semprons or Celerons. Those budget chips should have some better yeilding than their higher end twins, but the cost of production wouldn't be far apart (unlike their price).

In short, one could say there is hardly a difference between a $60 Celeron (Sempron) and $4000 Xeon (opteron), neither technically, nor production cost wise.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
May 6, 2005 11:54:43 PM

The biggest diference is where on the wafer they come from.
If you are trying to show people that the 8 way opteron is the best, you better be using only the center of your best wafers. Not even all wafers would be good enough for an 8 way. Sort of like beef. A herd of cows may look pretty much the same, but once they are cut up, some pieces will be prime, while others will be hamburger.
If we only made hamburger from all of it, the price of the hamburger would go up.
If we only used the prime cuts, the price of that would go up.
If you could enable all of the "8 way" parts, on a sempron, it would still only be an 8 way sempron, becuse the transistors, and pathways are not up to opteron standards.
May 7, 2005 12:29:51 AM

>If you could enable all of the "8 way" parts, on a sempron,
>it would still only be an 8 way sempron, becuse the
>transistors, and pathways are not up to opteron standards.

I don't buy that.

First, about the HT links, have a look at a die picture, and you'll notice they occupy maybe 5% of the die estate. chances that a die works perfectly well, but has a defect in one of HT links are extremely small, and really not the reason for this segmentation.

Next, in order to turn a Sempron into an opteron, not much needs to be done. Both AMD and intel will claim these server chips are validated much more thoroughly. Could be true, but I doubt it. A 99% functional sempron is just as useless as a 99% working opteron. And there is no software that will run on the server part, but fail on the budget chip.. so if there is an error, neither will get sold, period.

Where I am willing to accept there are differences, is certain tolerances. Therefore, maybe a few potential Semprons would not pass opteron validation at any frequency or voltage, but IMO at the very least 95% would qualify just as well as an opteron, though likely a speedgrade or so lower, or at a different Vcore. And even that is questionable, considering Xeons or Opterons don't overclock significantly better than desktop parts from what I've seen.

I think you're just buying AMDs and Intels talk to warrant the huge price premium, while in fact, the only reason is that the server market is willing to pay such prices for the cpu's, since they aren't that significant in the TCO, where the desktop market isn't. Next will you try and convince me Windows 2003 Data Centre somehow costs more than XP Pro to produce ?

= The views stated herein are my personal views, and not necessarily the views of my wife. =
May 7, 2005 2:05:31 AM

So, buying a dual opteron setup for a desktop (to maximize multi-tasking abilities) would be a waste of money then, especially since the dual core athlon-64's are about to hit the street.
May 7, 2005 3:43:37 AM

So, what you are saying is that all parts of the wafer are equal, and that binning is just a division of chips to thier end use, without reguard to ....
Having been the beneficiary of that binning, when using an xp-m chip, I dissagree.
While the opteron chips might only be .0001% less likely to run into an error, than an A64 chip, that would be a significant advantage.
Please remember that a server chip is more likely to run into heavy loads in adverse conditions, than a desktop A64.
If you are saying that the A64s are the best desktop chip you have ever seen, I would agree. I just believe that the opterons are a half notch higher.
As i said before, the 8 way opterons are the best of the best.
May 7, 2005 1:24:26 PM

>So, what you are saying is that all parts of the wafer are
>equal

No, that is not what I said. Where did you read that ?
Its indeed true that in general, dies from the center of the wafer will be better and/or yield and/or binsplit better. I just don't believe that opterons come from the center and semprons from the edge. In fact, I recall reading entire wafers are predestined to be opterons, as they run on different production lines. So the "center ones" (or rather: the best ones) may end up as HE opterons, or 2.6 GHz parts or whatever, where most of the edge ones might only pass qualification for a full voltage, or lower clocked opterons.

AMD might or might not recover some dies that could not be used as Opterons (because of too many cache defects, non functional HT links or other reasons) as semprons, but even *if* they do that, it will be a very rare occurance. Its simply not true that AMD (or intel) produces a lot of generic dies, and then reserves only the best ones to become server parts, and packages the rest as desktop parts. If there is a difference, its only tighter qualification (lower tolerances) of the server parts.

>Having been the beneficiary of that binning, when using an
>xp-m chip, I dissagree.

That is totally unrelated. Obviously, mobile parts are binned and validated because of exceptional abilities to run at lower Vcore, and or lower power consumption, so these parts are pretty much guaranteed to have a great overclocking potential. But what you bought as a XP-M 2500+ @1.2v (just guessing, numbers don't matter), might just as well have been sold as a desktop XP 3200+ @1.5v for instance. Although even here, I'm wouldn't be surprised if mobile parts come from different production lines, with a process skewed for lower clock binning, but better power characteristics. Just that maybe some of the worse parts of this line could still end up as cheap desktop chips, if they can neither clock high, nor run at low voltages.

But that is for mobile chips, which indeed need to have different characteristics than desktop or server parts. But I don't really see different requirements between (full power) Opteron 1xx, 2xx and 8xx or even A64s, HT links aside.

>While the opteron chips might only be .0001% less likely to
>run into an error, than an A64 chip, that would be >significant advantage.

Fair enough. But the best chances to obtain that 0.001% increase is clock the part lower. Its indeed plausible that on average A64s run slightly closer to their tested limits than Opterons. But the difference will not be bigger than on average half a speed grade or so, and if you look at the pricing, you'll notice a 2.4 GHz opteron 850 doesn't cost anywhere near the same as 2.2/1Mb A64. So binning doesn't explain the price differential, the main reason is opportunistic pricing, and to a lesser degree, the worse economy of scale for the different testing, packaging, marketing, etc of the low volume server parts.

In short, there is basically no real difference (HT links aside again) on the die level between an Opteron 1xx and 8xx, or any opteron and an A64 running, say, 100 or so MHz slower.

>As i said before, the 8 way opterons are the best of the
>best.

I would suspect the Opteron EE versions are the "best of the best", regardless wether they are 1xx, 2xx or 8xx ones, because the EE's *are* binned differently. But they are likely not any better or different than 100 Mhz higher clocked Turions with all HT links enabled .

= The views stated herein are my personal views, and not necessarily the views of my wife. =
May 7, 2005 1:43:45 PM

>So, buying a dual opteron setup for a desktop (to maximize
>multi-tasking abilities) would be a waste of money then,
>especially since the dual core athlon-64's are about to hit
>the street.

Most likely, but not necessarely. Dual core A64s will be sold as high clock, high end, and be priced above single core chips.

I havent compared the prices with opteron yet, but its possible you could buy a pair of lower clocked 2xx opterons cheaper than a single A64 X2. For some apps, 2 opterons at ~1.8 GHz might give better price/performance than 2.2GHz+ dual core A64. And this will give you the additional benefit of being able to upgrade the opterons to dual core, giving you 4 cores :) . Alternatively, a single dualcore Opteron 165 (1.8 GHz) might be considerably cheaper than an A64 X2 (just compare the prices, i don't know), for not that much worse performance.

2 way Opteron motherboards can be found cheaply, the main problem might well be Opterons requirement to run registered ECC RAM (depending how much RAM youd want). This memory is more expensive and slightly slower than unregistered ram, and as a deskop users, you might not find that a worthwhile tradeoff for the the increased stability/data integrity and increased memory capacity (some opteron boards can run up to 8 Dimms per cpu).

A side note, to maximize "multi tasking abilities", learning to set thread priorities might give you a better, and far cheaper solution. Unless you are one of the very few people that care about the execution time of the background thread (like encoding, rendering, virus scan..) while running a cpu intensive foreground thread (like a game). Most people only care about not visibly losing performance in the foreground thread (game), and therefore, you really don't need 2 cpus/cores.

If you want dual cores to run SMT friendly apps like rendering or some encoders however, and you don't need state of the art single threaded performance (games), and you don't mind high power consumption/heat, a Pentium D might be worth a look too. Its considerably cheaper than either dual opterons or A64 X2s, and its drawbacks might not be important to you.

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