AM2 Reviews and Benchmarks

[tehdigit]

<3 is a heart

sideways

 

[septoid2]

I could careless, whatever gives me best priceerformance ratio, I'll buy.

 

[idiotec]

A bit off topic, but AMD has announced 65nm Rev. G will be out by the end of this year. http://www.dailytech.com/article.aspx?newsid=2489

I will bet that with this core change they add some IPC improvements as well so it will close the gap a bit. It could lead to higher scaling as well.

 

[Green Radience]

65nm is nice, but without significant architectural changes (basically a whole new architecture) nothing else is going to matter. Conroe seems to hold a 1:1.33 Conroe:A64 ratio. That might not seem so bad; until you consider that CPUs don't operate on the 1Ghz scale. A 2.0Ghz Conroe would be like a 2.66Ghz Dual Core A64. And that places the $244 2.13Ghz Conroe close to the $1010 2.8Ghz Athlon 64 FX-62.

Even with 65nm, AMD would need to clock a A64 at 3.55Ghz to compete with the 2.66Ghz Conroe E6700. And then, they would need to cut the price in half to compete with the $530 pricetag of the E6700. I'm not trying to be a Intel/Conroe fanboy, I've provided you with clear proof.

K8L isn't an architectural change. As AMD said, "evolutionary, not revolutionary." It simply adds Quad Cores, faster HT, FB-DIMM/DDR3 support, etc etc. All things that are nice to have, but nothing to catch up to Conroe. Yea, neither 65nm, nor K8L will do anything without actual architectural changes. And all revision G seems to be is a move to 65nm. Sure, that might lead to some performance improvements, but only in the single-digit %s.

 

[Nitestick]

You don't seriously think that it was the Pentium 4 that kept Intel on top? Intel is about 6 or 7 times as big as AMD just in manufacturing capacity, AMD had no chance. And have you seen Intel's advertising? Don't even bother to hold up this particular argument.

AMD's K8L won't do anything even close to that for AMD.

well the fact that it needs to be explained to 80% of the population that clock speed isn't the sole factor in determining performance says that they had the market whipped. do you think if everyone understood the difference they still would have bought P4's? i understand AMD can not challenge Intel's marketing. but Intel's mareketing crutched on Netburst.

 

[gaara]

My guess is anandtech is referring to a tweak in the memory controller to improve memory read/write times with larger memory modules, this will of course only apply to socket F opterons and won't really make any difference to most people

What I'm hoping is that they know something about AMD sampling DDR3 with Socket F and future memory controller revisions as DDR2 does not like working with ECC enabled for some reason and they are taking early steps to avoid this problem...it is interesting to note that the desktop revisions were released before the workstation revisions which is usually not the case so there's definetely something up, if it was anything big though there would certainly be more information about it though

 

[Green Radience]

But do you really think any of those will make up for the Performance-per-Clock and Performance/Price lead that the Conroe has?

Oh, and look what I found about pre-release processors;

You have to bear in mind that Engineering Sample are effectively worse performers than the final core - anyone remember early Clawhammer performance.

Linky. Turns out that the release Conroes may infact be even better performers.

 

[gaara]

Believe I mentioned a while ago that alpha and beta cores are not accurate measurements as there are still bugs to be worked out, although this more applies to LDT compliant cores IMO as the memory controller is much more difficult to tweak since as you can see the A64 memcontroller wasn't perfected until almost two years after it was mainstream

And no, I don't think it will neccesarily save AMD or put them on the save plateau as Intel currently stands, however if they play their cards right the potential is certainly there. I believe they've already binned DDR3 as high as about 1.5GHz effective speeds which basically means RAM is getting pretty **** close to being much larger off-die L2 cache extensions. Think about it, the Athlon 3000+ effectively executes it's L2 cache at 1.8GHz + whatever latencies are present. If it can execute addresses from the memory at around ~1.5GHz with the added latencies of an off-die cache which the HTT bus will probably compensate, you're starting to get pretty close

Furthermore, rumour has it DDR3 modules are supposedly going to be a complete byte per pincount so you looking at DDR3 running at 800MHz running as 3.2GHz effective speeds...plus some people are also saying Samsung has started sampling DDR3 with 256bitrates rather than 128bitrates although that seems unlikely IMO

Now if AMD had gone the road I wanted and used XDR instead of DDR3 they'd be getting 8GHz effective speeds and possibly higher as RAMBUS has been sampling XDR2 now I believe

This all said though, I believe Intel is going to incoporate DDR3 support before AMD will, but the HTT link will play a vital role IMO as the memory frequencies are starting to get close to effective core frequencies, and considering the HTT already operates at 2GHz and can go a lot higher welllllllll

 

[Green Radience]

Concerning that last paragraph...you mean HT? I'm not aware of a 2.0Ghz HTT, but 2.0Ghz HyperTransport Links, yea.

 

[gaara]

HTT = HyperTransporT

yes I realize most people call it the HT but I prefer HTT and don't wanna get confused with hyperthreading