A Look At AMDs Socket AM2 Platform

[kenofstephen]

low latency is a must for K8 performance,

I don't think the result is fake, based on such a high latency, K8 CPU can't do anything

AMD sure will lose this time if the high latency problem can't be solved.

better save some money for Conroe.

P.S. I am a Opty165 user (o/c 2.8GHz).

Then explain to me how a Pentium 4 with DDR2-533 can score better in Sandra than this Athlon64 supposedly running DDR2-667. Or are you saying that Intel's off-chip memory controller is somehow more "suited" to DDR2?

So far, I'm seeing so much BS in this thread that it isn't even funny.

OMG, you dunno Sandra is Intel optimized??

 

[apache_lives]

I'm an older member than you, Joined here aug. 2005, you joined jan 06.
Also, I'm a veteran poster at NVIDIA's Forum: http://forums.nvidia.com/index.php?showuser=10333

well done your a member of a nvidia forum, this here is a general hardware forum (no offence) - dont care who/what you are, aslong as you can supply some good information and advice

 

[Brunnis]

OMG, you dunno Sandra is Intel optimized??

I'm talking about the memory tests, not the computational ones. The memory test usually approaches the theoretical limit of any platform, be it P4, AthlonXP, Athlon64, or whatever. Even my old AthlonXP reached 90%+ memory efficiency in the Sandra bench, no matter if 2-2-2 or 2,5-3-3 timings were used. This Athlon64 shows 55% efficiency and you actually believe that this is because of higher latency memory or some Intel optimization (that hasn't materialized until now?

Allow me to say "OMG".

 

[lewis71980]

To be honest. I'm frustrated that I never seem to see reviews that use for example Kingston Value ram in their comparisons.

It's all well and good benchmarking with low latency memory. But if I am looking at all these reviews using special memory.

A majority of systems you will buy off the shelf won't use it.

I would like to see a benchmark of the systems running at stock speeds , stock memory timings , with standard 'cheap' ram.

Still waiting on the antivirus realtime protection, gaming / office app benchmarks. See which one is the most resource hungry.

 

[slvr_phoenix]

I am sorry Patrick, but I think you have been had.
...
Do you really believe that Sandra would score that low, if you were actually using a 333mhz memory bus?

I don't know if I can believe in the conspiracy theory part, but man oh man, something seems seriously wrong here.

There's no way Sandra's memory bandwidth scores could be so low unless the frequency really was what was reported. Latency just can't account for that much of a discrepancy.

 

[The_nVIDiA_Guy]

why did you start posting here did they kick you from the nvidia forums?

No, and your going off topic. I wanted to post is a more active forum, where their are 30 members online, rather than 2.

 

[TechnologyCoordinator]

(My first post here, hope every thread doesn't get ugly like this one. . .)

This is clearly a step in the right direction for AMD. The benchmarks by TomsHardware will only improve as timings get better, DDR2 800 is used, and with the fix to the onboard memory controller.

I've been an AMD fanboy since the Athlon XP series of processors and recently bought my second AMD processor, an Athlon 64 3200+.

My opinion of AMD is that they offer superior products at lower price points.

My opinion of Intel, as of lately, is that they offer marketing, weird naming schemes, and weird gimmicks like Viiv.

 

[gerhardb]

I haven't had time to read all of the tread here... but based on the first page it's probably a good idea to point a few things out:

AMD created the integrated memory controller to properly feed the Opteron.

All of the current AMD processors are Opterons with all of the features turn on (in the case of a marked "Opteron") and certain other features turned off (in the case of the Athlon 64, for example).

My understanding in speaking with folks from AMD is that the current DDR memory controller based Opterons support PC3200 DDR RAM for only one reason - MARKETING.

They've gone on to tell me that the performance differences between the original memory controllers using PC2100 and the newer ones using PC3200 are more or less non-existant.

They've included "faster" RAM support so that their vendors can show the Intel based world that they're comparing apples to apples. (Who says all of us IT folks are smart enough to understand that an Operton with PC2100 RAM is as fast [or faster, depending] than any Intel P4 based system?)

DDR2 support is likely the same exact requirement being met by AMD to prevent a slump in their sales as the vendor market begins to ask "When are you going to support DDR2?"

I would not have expected huge improvements in memory performance while the core of the processor remains basically the same.

The move to DDR2 is more of a quality of life issue, as you can get the RAM without paying huge amounts of money for it... which is really important if you are HP, Dell, Gateway, Sony, Toshiba, etc...

 

[MJuric]

Question from the ignorant.....

Just looking at the benchmarks etc it would appear that the two systems are exactly the same save the faster memory. It woudl also appear that the results are nealry identical unless I'm missing something.

I guess my question is why should there be any difference? How much overall improvement should one see with faster memory?

Again maybe an ignoratn question but wouldn't the only reason the system with faster memory woudl be faster is iof the CPU was waiting for infomation from memory? Is this a typical situation?

~Matt

 

[Rob423]

i agree with the second poster..

how can they bench something that doesn't even beat an older technology.

Somethings 2 weird in these marks/results.

 

[IVAces]

Has it occured to anyone that they might have put the results out in order to be the first to do so?

That would explain the release of the results even though they were as mediocre as they were.

What wasn't explained in the test what the quarky things everyone is arguing about. I think the followup article should include such things.

 

[CompGeek]

Memory? Do you think you'll get tha' boost from memory? Maybe but only once they'll have good DDR2 . Current DDR 2 is by no means better than DDR. It has high bandwitch but at the expense of latencies. However, DDR has already reached a dead end while DDR2 can be improved and overall is a better base for creating better memory.
DDR 2 800 mhz might bring minor improvement in certain areas but still do not expect anything special.
http://www23.tomshardware.com/cpu.html?modelx=33&model1=206&model2=218&chart=67
Look at how a EE 3.4 is performing with DDR and DDR2. I doubt the DDR had low latencies so DDR actually performs better.
AMDs memory controller mostly helps in games. It didn't transform a DDR in a 667 mhz DDR2. 667 mhz DDR2 is roughly equal to DDR performance wise,spare me the bandwitch crap. In fact i would still believe that current DDR low latency is better than the best DDR2 though i might be wrong(couldn't find benchmarks testing this).

So let me rephrase all that mambo jambo in one sentence:
AM2 will be aproximately equal to 939 at clock per clock. The performance will mostly come from higher clocked CPUs. I expect AMD to break the 3 GHZ barier only after they implemented 65 nm into their chips. I suppose the new mem. controller will improve gaming!?

 

[luminaris]

Well, I agree with MPJesse and ltcommander. We all will just have to wait until they release the final production chips.

The whole point of the review is just to give us a taste of what to expect, that's all.

Everybody calm down, relax and drink LESS coffee

 

[mpjesse]

In theory, it should be faster because DDR2-667 has an overall higher throughput than DDR400.

DDR400: 3.2GB/s per channel (6.4GB/s w/ dual channel)
DDR2-667: 5.333GB/s per channel (10.6GB/s w/ dual channel)

However, DDR2 memory, in general, has higher latencies than plain 'ole DDR. Meaning, it takes longer for the processor to access the DIMMs. As you can see from the benchmarks, the overall high bandwidth of DDR2-667 somewhat cancels out the high latencies associated with it.

And yes you're right, the systems perform nearly identically. But like someone said earlier, the move to DDR2 has much, much more to do with volume pricing for OEM's than it has to do w/ technology. It's very, very likely Dell will start selling AMD based systems soon. Dell can get better pricing if they're buying nothing but DDR2. If they have to buy DDR and DDR2- their pricing won't be as good. Same goes w/ HP, Gateway, etc.

Make sense?

 

[luminaris]

Makes perfect sense. Dell is in a position to buy products at volume and get the best prices. That's why they can sell a product so cheap.

Not only that, they want diversity and being able to sell both processors should make them compete better and drive prices lower.

 

[slvr_phoenix]

the move to DDR2 has much, much more to do with volume pricing for OEM's than it has to do w/ technology.

Actually, I thought it also had a lot to do with how quirky DDR1 can be when run in dual-channel at high speed, especially with four sticks. Something that DDR2 doesn't suffer from. OEMs like as little quirkiness as possible...

 

[slvr_phoenix]

Has it occured to anyone that they might have put the results out in order to be the first to do so?

Did it occur to me? Sure. Does that make it any better? Hell no. :lol: :lol: :lol:

 

[mpjesse]

Never thought of that. You're right though- dual channel is flaky w/ DDR. Still, remember that dual channel was popular in all OEM machines before DDR2 came along.

-mpjesse

 

[congo]

So, no one has yet to talk about the HTT change in AM2 yet. Didnt that get a big bump as well?

Also wondering if they fixed the 2T bug with full RAM banks, anyone know this also?

 

[MJuric]

I guess my point is that even with the idea that the RAM is supposed to be 50-60%, maybe less with the latency issue you speak of, what portion of total system performance would that be? Even if it was a 1:1 improvement with the RAM it woudl still only be, what, 5-10% total system performance, assuming everything was running perfectly?

What kind of error was there on the benchmark testing? System setup? etc etc etc. The point is, from a fairly uneducated point of view, the "theoretical" total system performance gains are likely pretty close to the likely error level of the benchmarking tests, not to mention that teh new technology is probably not tweaked as well as the old stuff.

I don't know I find rather amazing that everyone is surprised by the results, maybe I'm naive but everything was the same except the memory. I guess I just didn't expect top see any amazing gains form that change. If anything I think the test merely proves the new technology is viable and DOESN'T create performance loss.

~Matt

 

[mpjesse]

Agreed. I for one didn't expect to see any performance gains. However, I do expect to see something with DDR2-800.

-mpjesse

 

[MadModMike]

I was under the impression that the AM2's were going to be 333MHz Base Clock w/ 3x or 4x HTT Multi (for 1Ghz or 1.33GHz) and they would use DDR2 667 @ 1:1. I believe, even with the higher latencies, that this should bring performance increase significantly, but it appears that either these samples were clocked lower, something was wrong, or they aren't going to raise the FSB to 333MHz. Also w/ the DDR2 800, you think AMD would just implement that into the FX line and give the FX's the 400FSB? I know the 1:1 ratio doesn't improve performance significantly, but just having it is nice .

I really can't wait to try the FX-64 at 3GHz w/ DDR2/800 at sweet timings like 4-4-4-10, that would be some mad performance, at least bandwidth wise. But like I said earlier, I'm not sure if this extra bandwidth (up to 10GB/s or more) is really necessary on desktop PC's, I think the DDR2 800 should be saved for Socket F and try to get the DDR2 667 timings down to 2-3-3-6, then it would be nice performance boost at 333MHz.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

 

[MasterWolverine]

Which begs my question, what is the point of the article? A 'sneak peak' at a screwed up PC that in no way reflects real world results? :roll: That's not exactly useful or informative. :?

The Point of the article is to get us drooling over the pretty little silicon that we are going to buy June 7th to put in our computers June 8th!

 

[mpjesse]

Also w/ the DDR2 800, you think AMD would just implement that into the FX line and give the FX's the 400FSB?

I had the same thought. To me it makes more sense to do that. But I think AMD wants to wait and see how the new Core Duo's perform in the desktop arena. My thinking is they'll only give Athlon 64's, X2's, etc 400mhz FSB's if they have to.

I guess we won't know for sure for at least a couple months. But everyone in the hardware journalism circles is screaming "DDR2-800 for everyone!"

🙂

-mpjesse

 

[MadModMike]

I think it'd be sweet if they could give you the option of 333 or 400 BUS on the FX' s where going to 400 wouldn't effect the CPU hat much or force you to shove more volts in it. Maybe they could design the FX's around a 400 FSB and then drop it to 333 or give the option to go 400, just a thought. But ya, once Conroe comes out, people can see whether or not 400MHz is required, I'll still OC my AM2 I get to 400 w/ DDR2 800, but it'd be nice to be standard, maybe could hit 500 then :lol:.

~~Mad Mod Mike, pimpin' the world 1 rig at a time