A Look At AMDs Socket AM2 Platform

Although the DDR2-powered Athlon 64 platform is now up and running, AMD will not release it until June 6th. While this is mainly a business decision, it also has the benefit of leaving some time for performance optimizations.

im actually impressed - performance is about the same, yet the thermal specs are really impresive (35w for the x2 3800+ etc) - i wasnt expecting a performance boost so much.

When conroe comes out it might still be hotter but if its "20% faster" as they claim it will be valid for the heat, compared to the prescotts which now took like a total joke - a 3800+ X2 - 35w is not even 1/3 the heat of a prescott or smithfield and the prescott doesnt even have an IMC (comparing 90nm chips).

What I found to be ridiculous is their SiSoftware Sandra Memory bandwidth results, they put DDR2 667 at 5.8GB/s, but their past Benchmarks put DDR2 667 in an Intel platform at 6.6GB/s, hmm... 800MB/s loss, that aint due to different RAM, unless this RAM was special RAM...

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

What's up with the cpu-z shot showin memory frequency to be 100mhz?

Just not in the first gen

mike you gotta remember that ddr2 has higher cas latency and that amd procs love low cas ram mine does the poblem could be that the ram is at fualt for the low bencheis i can wait for ddr2 cas 2 or 2.5 ram then ill jump on the m2 bandwagon

Yup, i knew that DDR2 with high latancies will actually decrease performance in most things. The DDR2 800 should at least match the AM2 socket with the 939 if not produce a performance increase of 3-4 %. Though i dunno the latencies for that kind of memory. If they end up high it will be wasted money. The price is another concern. Not all people are willing to pay +2xx$ on 1 Gb of memory.
AM2 may not be a step forward in performance but it is leap AMD were forced to take in order to ensure the chip evolution. Socket F will decide if AMD was right or wrong. Untill then,i say that Conroe will kick AMD-ish arse.

socket 754? The cpu z is gone mad...
http://www.tomshardware.com/2006/0 [...] page6.html


And, just remember the initial problems with ddr 400 who arrived with CAS3.0+.....In this days is only 2.0...that will probably be the same thing with ddr2, it will arrive at a lower latency very soon...

Well, guess that 10% inproovment.....is coming. No reason to upgrade yet(or downgrade?) Looked like nice DDR2 to me, just isn't like my OZC VX DDR500 2-2-2-5. 65nm is needed, fast. AMD/IBM say they will have 40% better for eather power consumtion or speed at 65nm.

Review quotes:
Whether the current engineering sample processors actually suffer from a memory controller bug is hard to say. This could also be due to information that was released intentionally to prevent people from doing early benchmarking.

We do not want to speculate over the existence of a bug in current silicon as reported on the web, since the processor we used is too far away from mass-production anyway.

The processor we used for our tests was an Athlon 64 X2 4800+ for Socket AM2, which made its way to motherboard companies in mid-December 2005.

As already mentioned, the memory is suspected to suffer from a performance bug, which may be true or simple a matter of the early product stage

What it the point of benching this known bad chip? It may be the first, but it's still incorrect and misleading.

What has been known for a long time, has now come to reality: There are no performance benefits...

This is funny to see how AMD is sitting around praying for DDR2 800 to come and save them. I couldn't beleive the whole AM2 thing was to switch from DDR to DDR2. Disappointing.


,,

ok, a few things,
1.)if the new socket is pin compatable with 940/939 but slightly different, what is stoping a 3rd party from making an upgrade path to AM2

2.)i will not be interested in AM2 until the market matures, this will be fall or christmass. For a mature market, i would expect 65nm chips 2+ cores and support for better memory timings than intel.

3.)AMD integrated memory controller is going to be a huge advantage on AM2, with 939/940 Intel had already gone to ddr2 and so the integrated memory controller was able to give DDR1-400 the speed of DDR2-667. With AMD now moving directly to DDR2-800 I would expect 20-33% better memory performance on an AMD solution, these improvements will be driverless, chipsetless, and virtually invisible to the user as they are able to release new steppings of the AM2 chips. You will be able to upgrade from 4-4-4-12 to 2-2-2-5 with only a new cpu. My guess is most will upgrade from a AM2 x2 core to a AM2 x4 core in another year or so.

4.) still not a fan of Vista, I dont like the DRM, but since Microsoft has sugested that users should go ECC, I am wondering if AM2 chips will support memory with or without ECC, or will you have to pair an ECC AM2 chip with ECC memory?

Anybody , who had alook at the picture on page 6 ?
How about the author, for example.... ?

Why ? Well because CPU-Z said, that the memory frequency is 100.5 MHz. In other words, the system ran at DDR2-400 speeds.

Hence the "bad" Sandra scores, I "really" wonder why these are actually in the DDR2-400 range , hence the whole conclusion is for the sewer...

Sorry, but from a decent articel, I demand, that the author is watching his own screenshots.

Yes CPU-Z is displaying a lot of nonsense, but the RAM devider is read directly from the chipset, no speculative guessing needed, like for the CPU name.

So please check the DRAM frequency. Dont rely on the BIOS post messages, these can be wrong, especially on beta, engineering boards...

byebye

FalseInformation

What the hell do you think the switch from Socket 478 to LGA775 was?

Besides PCI-E, it was a change from DDR to DDR2.

Your post makes no sense.

Save them from what? Have you missed all the positive financial analysis' of AMD's future? Did ya miss the part where AMD continues to gain market share? Did ya miss the part where they're continuing to make more and more money every quarter?

Intel fanboy, get lost.

-mpjesse

That would be DDR2-200.

The benchmarks clearly don't reflect that level of performance. CPU-Z is wrong.

It also says DDR, not DDR2? :?:

Save them from what? Have you missed all the positive financial analysis' of AMD's future? Did ya miss the part where AMD continues to gain market share? Did ya miss the part where they're continuing to make more and more money every quarter?

Intel fanboy, get lost.

-mpjesse
Financial analysis is not the same as product analysis and to get it straight: Analysts rarely do have any idea of the products at all. Intel also keeps making more and more money and they do it some sizes bigger than AMD.
Besides that, your comments has like nothing to do with what he said...

Anything else to say?

I don't think that anybody has mentioned here that the test sample supossedly suffers from a memory bug that is causing poor performance, and that is supposed to be rectified in an upcoming revision before release? That could explain the strange memory timings.

It seems a lot of people on here, ESPECIALLY Intel fanboys are treating this as the end all comparison, while the article itself states this is not meant to be an accurate benchmark of the final performance, just a sneak preview. Regardless of other factors, AMD is VERY unlikely to release a new socket with lower performace than the current gen. I can gaurantee that things will change before the official launch. Thats just the way things work.

also in chase this post is going to be deleted, because it´s not pro intel or pro tomshardware:

thanks to rkinet for his ideas:

Sandra 2005 pro: DDR-II 667 // Memory Bandwith < 6000 MByte/s ?
using DDR-I 400 socket 939 makes approx. 6 GByte from theoretical 6,4 Gbyte/s = -0,5 GByte/s, so why does DDR-II 667 make only less than 6 GByte/s at theoretic 10 GByte/s = -4 GByte/s ?

this is equal to DDR2/400!!!

take a look at the DRAM Frequency on page 6: 100Mhz to the RAM?!?!?!?

what do you say??

sorry, but it took some time to get everything straight in english, as it isn´t my mother tongue

By him stating "funny how AMD is sitting around praying for DDR2-800" he is implying that there is impending financial doom for AMD. It's quite the opposite actually.

So explain to me how my comments had nothing to do with his statement?

1. I addressed his insipid statement that the switch to AM2 is somehow stupid because it's a change in memory technologies.

2. I addressed his comment about DDR2-800 "saving" AMD. So what does AMD need saving from? Hmmm?

3. Quite the contrary, I'd argue that most financial analysts (esp. in semiconductor) do know the products. Most financial analysts are really research experts. They spend hours studying products, supply, demand, etc. What evidence to you have to support your claim that analysts "rarely have any idea of the products at all?"

You too need to go away. Ur nothing but a troll around here anyways.

I think it's a pretty good showing for a pre-production system running on slower memory than it will be able support upon release.

An the release of a new platform that's slower than the old is not unprecedented or unheard of. Remember the initial release of the P4? Although it clocked in at 1.5 ghz initially (the highest processor clockspeed at that time), it was stomped in performance by the higher IPC athlon running at 1.2 ghz and even edged out by it's little brother, the P3. However as intel ramped up the clock speed the P4 became the fastest processor on the market, until AMD's hammer architecture that is.

I don't think that will be the case here though. When M2 is released I am confident that performance will be on par or slightly better than equivalent s939 parts. Remember this isn't a major architectural change and noone should expect it to perform like one.

http://en.wikipedia.org/wiki/DDR2

Wikipedia says quite the contrary...

Hardly a definitive source... but still.

-mpjesse

I got something to say... We all know you are an idiot. thank you hergie i was thinking that just now but since you said i voted for your post thx man for saying what i was thinking

First off, jesse, go jump off a cliff because you're a useless fat moron.
Secondly, the fact that the controller doesn't support DDR2-800 and that the CAS is high, proves my original point you fat oaf, that these benches are not credible, go pound down a cheeseburger and shut up fatty.

Brunis: I noted on the memory too, I can relax the timings on my DDR1 to that of the DDR2 (excluding the CAS of 4) and I can still squeeze 7GB/s out of 3-4-4-10 @ 240MHz, so there is no way that those at DDR2667 and same timings are getting less then 6GBs, and dont tell me "that extra cas is a big hit" because that's the newb saying of the year. This further proves the credibility of those benches is not correct.

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