"AM2: First Look at DDR2 Performance on AM2" at AnandTech

[shimq1]

" With AM2 just six weeks away will AM2 memory performance be all we expect? Find out in this first in-depth look at DDR2.."

Link: http://www.anandtech.com/cpuchipsets...oc.aspx?i=2741

[vapb400]

http://www.xtremesystems.org/forums/...ad.php?t=95648 you're a couple days late

[Shadowmage]

Look at the article. It's a new article

[shimq1]

Vapb400, the article you're pointing to came out April 10th. The article I linked came out on April 14th.

[gundamit]

Not much for the gamers to be excited about. Everest fans might get a little excited.

[Bar81]

This is a more interesting article as it features the most current spin of the AM2 CPU as well as more recent mobo revisions with updated BIOSes. Likely all of these are what will actually be the release hardware or very close thereto.

What I found most interesting was that DDR500 provides almost identical performance to DDR2-833 showing that anyone with a system right now that is running DDR500+ has no reason to upgrade to AM2. However, those AMD users still running DDR400 are now seeing that there is still a definite improvement that may justify the improvement. If the Conroe numbers from Intel were using DDR400 for AMD (I believe they were) then that huge gap between AMD and Conroe just got a lot smaller. It looks like Conroe should still be 10-15% ahead of AM2 but it's not going to be the absolute domination we expected.

It seems that those with top of the line heavily overclocked AMD DDR systems would probably be best served to wait until next year and see what K8L and Intel have to offer.

[Cobalt]

I think I'd prefer just to grab some nice g.skills to upgrade my system to ddr500 than shelling out for a new mobo, cpu and ram.

[Bar81]

Apparently this is the final spin of the CPU as AM2 has already shown up in France in final form per the inquirer.

Guess that puts a final nail in the coffin of another nn_step delusion (the magical conroe killing AM2 revision before release)

[Bar81]

I think I'd prefer just to grab some nice g.skills to upgrade my system to ddr500 than shelling out for a new mobo, cpu and ram.

I think a lot of people will agree with you. While I was gonna get Conroe, I'm considering waiting now until Q2 07 to see what hits unless some game comes out that's both incredible and needs a new CPU. I'm also not fond of the whole CPU/mobo/RAM upgrade to get maybe a 10-15% boost and we're still waiting for "real" reviews of conroe to see if that is the case.

[nn_step]

Apparently this is the final spin of the CPU as AM2 has already shown up in France in final form per the inquirer.

Guess that puts a final nail in the coffin of another nn_step delusion (the magical conroe killing AM2 revision before release)

Learn not to Double post and not to confuse what I am taking about and AM2

[Bar81]

Learn not to Double post and not to confuse what I am taking about and AM2

Learn what a double post actually is and, more importantly, learn how to join the rest of us in this crazy thing called "reality" :thumbsup:

[nn_step]

Learn what a double post actually is and, more importantly, learn how to join the rest of us in this crazy thing called "reality" :thumbsup:

Done
http://www.urbandictionary.com/defin...rm=double+post

and I don't know about you but my reality is accurate enough.

[HKPolice]

These CPUs should overclock better as well, I don't see why 3Ghz on air would be hard. Combine that with a 5% boost from 4-2-2 D9 chips @ 800Mhz or higher and it could be a tempting upgrade if you don't want to wait for Conroe.

[Vapor]

...there won't be a 5% boost from DDR2 800 4-2-2 D9 chips. That's what this article is showing (among other things). >20% boost in bandwidth and a ~15% reduction in latency nets a very small gain. Increasing bandwidth and decreasing latency on a few percentage points won't net you 5%, and very likely won't net even 2%.

In fact, when you OC the CPU past stock 1.8GHz, and set S939 system to DDR500 3-3-3, the gap actually NARROWS in actual performance increases.

EDIT: this shows that AM2 (and to an overwhelming degree, S939) has given dual cores all the memory performace that it can handle. AM2 is forward thinking though

[Bar81]

Done
http://www.urbandictionary.com/defin...rm=double+post

and I don't know about you but my reality is accurate enough.

Oh, how could I overlook the authority that is the "urban" dictionary Save it for someone who cares. And you couldn't find reality if it smacked you upside the head (which could only do you some good).

What is that now nn, 0 for 3 on your delusional ramblings. I'm starting to think that just by sheer dumb luck you should be getting something right by now.

[Bar81]

These CPUs should overclock better as well, I don't see why 3Ghz on air would be hard. Combine that with a 5% boost from 4-2-2 D9 chips @ 800Mhz or higher and it could be a tempting upgrade if you don't want to wait for Conroe.

Then of course you're assuming no overclock on Conroe. Overclocking one CPU to its max and running the other stock is a silly comparison.

edit: and see Vapor's reply as to your mem theory.

[HKPolice]

What are you talking about? The overclocked benches clearly show a few % benefit from running a DDR2 setup. Those tests were all done with a measly 2.5Ghz clock + single 7800GTX.

If you OC to 3Ghz + a SLI or CF setup where memory demands are higher, then a 5% boost is well within expectations.

...there won't be a 5% boost from DDR2 800 4-2-2 D9 chips. That's what this article is showing (among other things). >20% boost in bandwidth and a ~15% reduction in latency nets a very small gain. Increasing bandwidth and decreasing latency on a few percentage points won't net you 5%, and very likely won't net even 2%.

In fact, when you OC the CPU past stock 1.8GHz, and set S939 system to DDR500 3-3-3, the gap actually NARROWS in actual performance increases.

EDIT: this shows that AM2 (and to an overwhelming degree, S939) has given dual cores all the memory performace that it can handle. AM2 is forward thinking though

[[XC] Lead Head]

Oh, how could I overlook the authority that is the "urban" dictionary Save it for someone who cares. And you couldn't find reality if it smacked you upside the head (which could only do you some good).

What is that now nn, 0 for 3 on your delusional ramblings. I'm starting to think that just by sheer dumb luck you should be getting something right by now.

nn has been saying wait and see, he didn't say AM2 was going to crush the conroe.

[Bar81]

he's been saying a lot of things. All of which are bull.

[RTB]

...there won't be a 5% boost from DDR2 800 4-2-2 D9 chips. That's what this article is showing (among other things). >20% boost in bandwidth and a ~15% reduction in latency nets a very small gain. Increasing bandwidth and decreasing latency on a few percentage points won't net you 5%, and very likely won't net even 2%.

In fact, when you OC the CPU past stock 1.8GHz, and set S939 system to DDR500 3-3-3, the gap actually NARROWS in actual performance increases.

EDIT: this shows that AM2 (and to an overwhelming degree, S939) has given dual cores all the memory performace that it can handle. AM2 is forward thinking though

So basically, the expected increase from going DDR2, and no other improvements yet. It's not that I'm disappointed with AM performance, but it could've had more.

[zakelwe]

This certainly seems to indicate that the required increase neded by AMD to match Conroe is not going to come from the new memory interface. Where is it likely to come from then ? There's no speed bumps due, what else could give it ?

Regards

Andy

[RPGWiZaRD]

About when will AMD stop production on S939 btw? I need to know which time is right for me to yet pick up a DC Opteron or X2 before it's too late that I will have to keep for maybe up to 2 years, can't afford a completely new motherboard and ram and cpu for now. I think AMD dual core prices are still too damn high for both S939 and upcoming AM2.

sux being a student still.

[coldpower27]

This is a more interesting article as it features the most current spin of the AM2 CPU as well as more recent mobo revisions with updated BIOSes. Likely all of these are what will actually be the release hardware or very close thereto.

What I found most interesting was that DDR500 provides almost identical performance to DDR2-833 showing that anyone with a system right now that is running DDR500+ has no reason to upgrade to AM2. However, those AMD users still running DDR400 are now seeing that there is still a definite improvement that may justify the improvement. If the Conroe numbers from Intel were using DDR400 for AMD (I believe they were) then that huge gap between AMD and Conroe just got a lot smaller. It looks like Conroe should still be 10-15% ahead of AM2 but it's not going to be the absolute domination we expected.

It seems that those with top of the line heavily overclocked AMD DDR systems would probably be best served to wait until next year and see what K8L and Intel have to offer.

Not really the Conroe system was running DDR2-667 4-4-4-15, which are decent timings but not crazy good, Intel will also gain some performance on Conroe like AMD will with a switch to DDR2-800.

[AgentVX]

This certainly seems to indicate that the required increase neded by AMD to match Conroe is not going to come from the new memory interface. Where is it likely to come from then ? There's no speed bumps due, what else could give it ?

Regards

Andy

L3 cache maybe? (but how much will gain will L3 alone result in, I wonder...)
Dunno about you guys but I'm waiting for the NDA binding s7e9h3n (and others) to lift so that we can finally find out what he's been hinting at

[Bar81]

Not really the Conroe system was running DDR2-667 4-4-4-15, which are decent timings but not crazy good, Intel will also gain some performance on Conroe like AMD will with a switch to DDR2-800.

That's of course possible but there's a big difference from going from DDR400 to DDR2833 versus DDR2667 to DDR2800. Anyway, what this really tells us is that those early conroe numbers are mostly meaningless until we see a review of one in the wild in direct comparison to an AMD unit.