A64 to be launched with its 'crippled' brother

[mr_gobbledegook]

From reading this <A HREF="http://www.theinquirer.net/?article=10353" target="_new">artical</A> from the tabloid Inquirer site it looks like AMD will be releasing a crippled version the A64. So here's the summary...

AMD will be releasing two versions of it's Athlon 64.

1) Athlon 64 - 940 pin socket with 1MB L2 cache with Dual Channel Memory controller (for some reason switching to 939 pin later on ??)

2) Athlon 64 Value - 764 pin Socket with 513KB L2 cache (lack of pins suggest single channel memory controller).

Looks like A64 might be a stomper as I see no obvious differences between its larger Opteron brother except that it won't run in multi processor configurations.

Which begs the question whats the difference between a 1xx Series Opteron and an A64?

I suspect this is why AMD are going to switch the A64 to 939 pins so you can't subsitute an 'expensive' 1xx Series Opteron with an equivalent 'less expensive' A64 for people with NForce 3 motherboards (a 940pin socket m/b).

Do I detect AMD taking a similar strategy to Intel with thier P4's and over priced Xeon DP's ?? Hmmmm....

<font color=purple>Ladies and Gentlemen, its...Hammer Time !</font color=purple>

 

[Mephistopheles]

The fact that A64 has 940 does indeed suggest that it has dual-channel memory; however, the Inquirer's article doesn't say a thing about that.

Looks like A64 might be a stomper as I see no obvious differences between its larger Opteron brother except that it won't run in multi processor configurations.

Remember, even Opteron can't compete with a 3.0Ghz P4 at typical 32-bit tasks. I've tried to make a preview in <A HREF="http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&p=141516#141516" target="_new">this thread</A> using linear scaling for an Opteron 2.0Ghz and 2.5Ghz and considered several typical 32-bit software (the ones at THG!) Neither Opteron can beat a 3.0Ghz P4 at all. That's not good news, especially if you consider that only very few softwares will actually get optimised for Opteron this year (by the end of the year, predictions say it's 20% at the most) In the other cases, A64 should perform as seen in the preview...

1xx Series Opteron and an A64

There is none. That's the trick, they both use the same core.

 

[eden]

Dual Channel should not be the thing that takes more pins on the core. I believe the Hyper-Transport does.
Why on this one it is, when it has less HT links, might be the tactic Intel uses, indeed just so people don't use the Xeon on the same socket, knowing Xeon has little difference, unless you start adding cache.

--
<A HREF="http://www.lochel.com/THGC/html/news.html" target="_new">Join the Tom's Hardware Guide Community forum album, send us your pics!</A>

 

[Kelledin]

It makes sense. Keep in mind, it's rather more expensive to implement a second production line to produce CPUs based on identical or nearly-identical cores.

Don't know why one pin gets chopped off later in the cycle. I doubt it would be to introduce incompatibility--if that were the case, I would expect them to <i>add</i> a pin later, and make the two physically incompatible as well as electrically incompatible.

As for whether I might get one, I would be buying it for server apps, although I might use it for desktop apps as well. The "desktop" software I use is already AMD64-optimized, so meh.

<i>I can love my fellow man...but I'm damned if I'll love yours.</i>

 

[Mephistopheles]

The "desktop" software I use is already AMD64-optimized, so meh.

Just out of curiosity, where did you get A64-optimized software? Aren't the compilers currently under development? Doesn't make much sense to me... :frown:

 

[Kelledin]

Just out of curiosity, where did you get A64-optimized software? Aren't the compilers currently under development? Doesn't make much sense to me...

It may make more sense once you know I use Linux. The GNU toolchain for AMD64 is currently working in production, thanks in great part to some hard work done by SuSE.

Considering that gcc-3.2.x now keeps pretty good pace with icc 7.0 for everything but the Pentium4, that's a good deal right there.

<i>I can love my fellow man...but I'm damned if I'll love yours.</i>

 

[Mephistopheles]

interesting. As I understand it, the GNU compiler is already (at least partially) optimized?...
Interesting...

ICC is an excellent compiler for Intel platforms, obviously. A quite impressive one at that.

 

[Kelledin]

interesting. As I understand it, the GNU compiler is already (at least partially) optimized?...
Interesting...

Yup. Mostly because scheduling in the Hammer isn't all that much different from IA32 processors like the Athlon. There's a great many interesting new forms of addressing, of course, like RIP-relative etc, but keep in mind that the Hammer is designed to run "legacy" IA32 code very well.

gcc's Pentium4 optimization isn't so great though. gcc currently doesn't do anything for the Pentium4 except avoid a few of the worst pitfalls of the architecture relative to earlier x86 CPUs. Unfortunately, there are quite a lot of such pitfalls on the Pentium4, so gcc has a lot of room for improvement there.

<i>I can love my fellow man...but I'm damned if I'll love yours.</i>

 

[Crashman]

Well, actually if the memory controller is on the CPU with these new processors, you WOULD need more pins for dual channel memory! Think about it, dual channel is a 128-bit bus, single channel is a 64-bit bus. 128-bit requires more signal traces than 64-bit.

<font color=blue>Watts mean squat if you don't have quality!</font color=blue>