Sign in with
Sign up | Sign in
Your question

AM2: FIRST LOOK AT DDR2 PERFORMANCE ON AM2

Last response: in CPUs
Share
April 15, 2006 4:33:25 AM

Just as I said before, a new AM2 revision from AMD IS faster than socket 939 by a margin of 10% running average applications and gives 30% more bandwidth than socket 939 which will be useful for upcoming AM2 revisions. :wink:

Once AM2 and conroe are out, we'll see the REAL benchmarks. :wink:

Quote:
However, the memory bandwidth increases of up to 30% on AM2 and the Latency improvements of 12 to 16% compared to the fastest DDR memory on socket 939 do not yield much in real-world performance. The real-world performance increase for AM2 compared to Socket 939 will likely be very small - in the range of no increase to about 10%, depending on the application.


http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2741&p=1
April 15, 2006 4:35:32 AM

Quote:
Just as I said before, a new AM2 revision from AMD IS faster than socket 939 by a margin of 10% running average applications and gives 30% more bandwidth than socket 939 which will be useful for upcoming AM2 revisions. :wink:

Once AM2 and conroe are out, we'll see the REAL benchmarks. :wink:

However, the memory bandwidth increases of up to 30% on AM2 and the Latency improvements of 12 to 16% compared to the fastest DDR memory on socket 939 do not yield much in real-world performance. The real-world performance increase for AM2 compared to Socket 939 will likely be very small - in the range of no increase to about 10%, depending on the application.


http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2741&p=1

It clearly says 0 to 10% not a overall 10% increase in performance.
April 15, 2006 4:51:47 AM

You realize that Conroe is doing 32 million decimal calcs on Pi in 21min 51seconds which takes a A64 dual core running at speeds near 3.3GHz or greater to equal.

That is fast and priced way less then what AMD will be able to market if and when it reaches that clock speed with AM2.
Related resources
April 15, 2006 4:55:02 AM

I really expect AM2 to deliver 10-13% more performance than s939 once it sees the light of day.
April 15, 2006 4:57:50 AM

The conroe is using SSE extensions, in 128 bit . Not like it's doing it long hand. Too bad it cant do all calcs that way.
There are advantages to Intel's new marchitecture, but we dont know at what cost yet.
April 15, 2006 4:59:26 AM

Quote:
You realize that Conroe is doing 32 million decimal calcs on Pi in 21min 51seconds which takes a A64 dual core running at speeds near 3.3GHz or greater to equal.

That is fast and priced way less then what AMD will be able to market if and when it reaches that clock speed with AM2.

Well, if all you're going to do is run super pi on your comp, then, conroe is better than A64... :roll:

...oops, I forgot this:
conroe performance being busted.
8)
April 15, 2006 5:02:47 AM

Most of the improvement seems theoretical from the memory tests which is obviously while the real world improvements seem much more conservative.

Quote:
Hopefully, this last series of overclocked performance comparisons will finally put into perspective what is and is not possible with improved DDR2 memory performance. Clearly AM2 will launch a bit faster than current Socket 939 performance when comparing the same processor speed. However it is not likely that further increases in DDR2 bandwidth or latency will translate into further improvements in performance with the current AMD/Athlon64 architecture. Further improvements in AM2 performance will have to come with revisions to the core, more cores, die-shrinks and higher speeds, and increases in cache. The move to DDR2 will bring small improvements in performance, but DDR2 alone is not likely to bring the large performance boosts many hope for.

I think that pretty much sums it up. It's still nice to see numbers improving over S939 though.
April 15, 2006 5:11:31 AM

Quote:
Most of the improvement seems theoretical from the memory tests which is obviously while the real world improvements seem much more conservative.

Quote:
Hopefully, this last series of overclocked performance comparisons will finally put into perspective what is and is not possible with improved DDR2 memory performance. Clearly AM2 will launch a bit faster than current Socket 939 performance when comparing the same processor speed. However it is not likely that further increases in DDR2 bandwidth or latency will translate into further improvements in performance with the current AMD/Athlon64 architecture. Further improvements in AM2 performance will have to come with revisions to the core, more cores, die-shrinks and higher speeds, and increases in cache. The move to DDR2 will bring small improvements in performance, but DDR2 alone is not likely to bring the large performance boosts many hope for.

I think that pretty much sums it up. It's still nice to see numbers improving over S939


Still, AM2 delivers 10% better performance over socket 939. Once Anand and many others have the real chip (not engineering samples), then we can judge real performance and as sexbomb says, I do believe AM2 will deliver 10-15% real world performance.

I'm really glad to know that the processor is not bandwidth starved as Conroe and as Anand says, future revisions will take full advantage of the extra bandwidth and low latency:
Quote:
Further improvements in AM2 performance will have to come with revisions to the core, more cores, die-shrinks and higher speeds, and increases in cache
April 15, 2006 8:06:20 AM

I have been thinking about AM2 and after some thought I think its a good move for AMD and will allow the A64 to breath better :)  that 2Ghz FSB is useless when tied to slow ram ! how much performance I cant say.... Another good thing to remember is that soon DDR2 will be cheaper then DDR1 and its price will go higher (have you seen the price for SDR133 ???)
April 15, 2006 8:16:57 AM

Quote:
Just as I said before, a new AM2 revision from AMD IS faster than socket 939 by a margin of 10% running average applications and gives 30% more bandwidth than socket 939 which will be useful for upcoming AM2 revisions. :wink:

Once AM2 and conroe are out, we'll see the REAL benchmarks. :wink:

However, the memory bandwidth increases of up to 30% on AM2 and the Latency improvements of 12 to 16% compared to the fastest DDR memory on socket 939 do not yield much in real-world performance. The real-world performance increase for AM2 compared to Socket 939 will likely be very small - in the range of no increase to about 10%, depending on the application.


http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2741&p=1

It clearly says 0 to 10% not a overall 10% increase in performance.
Give up he won't understand.

Word.
April 15, 2006 9:49:27 AM

10% was the best performance gain achieved, while average performance gain was 2-3%
Only Sandra shows greater performance gain, bandwidth gain that is, but we already knew that.

If you had read amdtech.com´s article properly, you would know better now.
April 15, 2006 10:20:21 AM

I will say this loud:
1. MEMORY BANDWIDTH PERFORMANCE IS NOT CPU OR SYSTEM PERFORMANCE!
2. MEMORY BANDWIDTH BENCHMAKRING IS SYNTEHTIC BENCHMARK!
3. SYNTHETIC BENCHMARK HAS NOTHING WITH REAL-LIFE PERFORMANCE!
4. DDR2-800 CL3 IS DAMN F*CKING TOO MUCH EXPENCIVE COMPARED TO DDR-400 CL2 TO ACHIEVE 1%-3% SYSTEM PERFORMANCE!
5. K8 + AM2 + DDR2 = WASTING TIME & MONEY
a b à CPUs
April 15, 2006 12:55:08 PM

"I really expect AM2 to deliver 10-13% more performance than s939 once it sees the light of day."

In purely synthetic mem bandwidth tests, perhaps...

I expect the gain in gaming framerates to be in line with the 1-2% gains shown...and only 18% shy of catching up with Conroe...! :-)
April 15, 2006 1:39:55 PM

I'm curious. What's with the current fad of posting keyboards? What does it mean?
April 15, 2006 2:00:32 PM

Quote:
I'm curious. What's with the current fad of posting keyboards? What does it mean?

Its an ongoing jab at people who post in all caps, claiming their keyboard is broke. Using all caps, can be considered shouting. But even the posting of the keyboard, is just like the topic, both are just shouting at the wind.
April 15, 2006 2:46:43 PM

nope, i think it was because of action man, im not sure though
April 15, 2006 3:31:56 PM

I see. That does make sense. :p 
April 15, 2006 4:22:17 PM

AM2 is indeed faster but is that enough to make it worthy of a good succesor. No. Because 2-3% in most apps is insignificant for most.
Why don't AMD fanboys accept that:
-DDR2 was a step AMD was gonna sooner or later
-while DDR2 is not great now,it'll get better later and it already has some obvious advantages over DDR

DDR2 was a time eater for both companies. AMD is just slower in the implentation process and also,remember, AMD also has to come up with a mem. controller. That's why people waited so much for so little. But it may seem little now,but by the begining of 2007 when real DDR2 will appear the performance gains might increase.
Still, until then Conroe will probably lead. I'm saving up right now for my future Conroe system. 2.4 Ghz chip seems very tasty for the price.
April 16, 2006 5:47:05 AM

Quote:
AM2 is indeed faster but is that enough to make it worthy of a good succesor. No. Because 2-3% in most apps is insignificant for most.

Take your big finger out of your @ss and don't make your self a jackass with that post. AM2 IS better than socket 939 since it delivers 10% performance over s939. AM2 is not and it won't be a failure like prescott since it couldn't deliver better performance over its previous socket which was based on the Northwood core.
April 16, 2006 5:54:29 AM

You want it after it has been up his ass? :roll:
April 16, 2006 6:03:31 AM

Do you not think sometimes that 9-inch just pretends to be so thick just to see how we will respond to his ill informed posts? I think he does it on purpose purely for entertainment. Sad really.
April 16, 2006 6:04:48 AM

nods, but he has been doing it so long he has bought into it inadvertantly.
April 16, 2006 6:05:30 AM

PhD = Piled higher and Deeper
April 16, 2006 6:08:08 AM

There are 7 basic pathological gaits that should be recognized by their characteristic pattern. These pathological gaits are:

* Hemiplegic
* Spastic diplegic
* Neuropathic
* Myopathic
* Parkinsonian
* Chorea
* Ataxic

These gaits have localizing value because they can indicate levels of neurological disease or systems abnormalities (such as cerebellar or basal ganglia).
April 16, 2006 6:14:21 AM

Quote:
PhD = Piled higher and Deeper


Hey.... I hold a PhD, actually two :)  ..... I guess I can pile it twice as high :)  :lol:  :lol: 

Wow! I had no idea. Hope I didn't offend you. I could tell you were very educated by the consistently high quality of your posts. But really I guess I figured somebody that educated would be too wrapped up in their career to spend a couple hours a day on a forum such as this.
April 16, 2006 12:59:39 PM

9-inch; the length of his.....
:oops:  .....
:p  .....
:roll: .....
:cry:  .....
8O .....
Hair.

Sorry if I disappointed any of you...
*loos significantly at sharikou*
DIE, EVIL NINJA!
a b à CPUs
April 16, 2006 1:03:03 PM

"Still, AM2 delivers 10% better performance over socket 939."

Then it's rather a shame that 10% "better" really only translates to a 1.5% improvement in actual gaming performance...
April 16, 2006 8:31:36 PM

Quote:
PhD = Piled higher and Deeper


Hey.... I hold a PhD, actually two :)  ..... I guess I can pile it twice as high :)  :lol:  :lol: 

Wow! I had no idea. Hope I didn't offend you. I could tell you were very educated by the consistently high quality of your posts. But really I guess I figured somebody that educated would be too wrapped up in their career to spend a couple hours a day on a forum such as this.

No, noooo, no offense at all, I hear all the jokes from family and friends... my dad bought me a post hole digger for graduation (Get it, Post hole Digger).

My degrees are in chemistry and physics, I got a minor in math as an undergraduate. My disseration was on solid state and interfacial phenomena both from a chemistry and physics perspective, specifically on Si(100) and Si(111) surfaces/substrates. In terms of this industry, I understand very accutely the steps in the manufacturing as I follow it very closely, it is related to my research.

Absolutely, no offense, it is hard to offend me overall, if I ever get any emotion over posting to this forum it is more "frustration" with the ignorance of some, and the fact that they want to remain ignorant. Add on top of that the complete and utter inability to objectively analyze data, and, well you see what i mean just look at this thread :) 

When I teach, I tell my students "There is no shame in ignorance, only shame in remaining ingorant"..... there is a lot of shame the flies around on the internet forums :)  .

On the topic of PhD, it is really nothing special... it is just remaining poor for 4-5 years (or in my case 10 years), dedicating yourself to slave labor for the joy of discovery. I have met some people who have a PhD and I wonder, how on earth did they get it? :)  ...

In terms of career, yeah, I put in my time, no doubt.... when the IDF news broke, though I posted a few messages, then found, hey this is a fun way to relax....and get a good laugh, Spud is absolutely cracking me up and down and sideways.

Word.
April 16, 2006 8:46:30 PM

Quote:
Not really, but I read the article this afternoon and wanted him to re-post it because I noticed that 1/2 his calculations were off. The 10% that 9-Inch is quoting is not correct. I noticed that the reviewer actually mis-calculated and has since been updated with the correct numbers. The max boost is just over 6% at the best timings availab.


Once more, here's the link you've posted:

Quote:
However, the memory bandwidth increases of up to 30% on AM2 and the Latency improvements of 12 to 16% compared to the fastest DDR memory on socket 939 do not yield much in real-world performance. The real-world performance increase for AM2 compared to Socket 939 will likely be very small - in the range of no increase to about 7%, depending on the application


7% overall performance is still a good figure against s939. I still blieve once the real thing come out and every reviewer has one on their hands, those results will be something like 10-13% at best.
It's really weird Anand didn't have a conroe chip to test it at this time. This makes me believe that something is not going too good with intel's upcoming processor. This is even more proof that intel had a "hand picked" processor for their idf debut. 8)
April 16, 2006 8:54:06 PM

Quote:
7% overall performance is still a good figure against s939. I still blieve once the real thing come out and every reviewer has one on their hands, those results will be something like 10-13% at best.
It's really weird Anand didn't have a conroe chip to test it at this time. This makes me believe that something is not going too good with intel's upcoming processor. This is even more proof that intel had a "hand picked" processor for their idf debut. 8)


since OEMs are expected to get AM2 mid-may, don't expect much more performance, especially since AMD has already had months to squeeze 7% out of it.

and why are you so optimistic about AM2 and so pessimistic about conroe (this question might as well be rhetorical)? if review sites have conroe, they're surely under strict NDA's.
April 16, 2006 9:11:47 PM

Quote:
:)  This is funny, because everyone who wants to "Bust the Conroe Performance" by quoting the victor wang benches on a subset of 2 anomolous results in Sciencemark 2.0. That CPU is one stepping younger, and it still is scaling correct based on other synthetics. PCMark05 on victors bench was 6100 or so, in the IDF benchs it was about 20% higher, Victors CPU was a 2.4 Ghz, IDF's was ~20% higher clocked at 2.6. So the numbers are jiving overall with a good scaling.

Next, on the memory bandwidth.... how many times have we heard "synthetics are meaningless".... actually, one could conclude that AMD has more work to do because the % increase in the bench bandwidth comes not where close to the theoretical increase in bandwidth in DDR to DDR2. In other words, if this is the best they can do, then it is not really worth it.... their older implementation reached closer to the theoretical bandwidth than the new AM2. Furthermore, the impressive increase in bandwith by % is not translating into real performance observations in real applications, therefore you hear and read time and time again that it is not a bandwidth problem.... the K8 is now at it's pinnacle and will go no further in improving IRC.

Now, going back to the post you quote.... I am critical of what you post because most of what you post is either unsubstantiated or misquoted or incorrectly calculated. That's ok because it is easy to confute. On top of that, I understand the Conroe criticisms, a few bench marks a conclusion does not make. However, if you are going to be critical of the Conroe Sciencemark 2.0 results, then you must include the entire breadth of the data and not just selectively pull what makes your case better. In the end it destroy's your credibility and you loose the debate.

Overall, the trend in the data is not on your side.

Jack


this is an excellent post. if you're going to be critical about certain information coming from intel, you have to be critical about the same type of information coming from AMD.

aside from that, it's pretty safe to say 9-inch has 0 credibitilty, as he's constantly doing exactly what you've pointed out. selectively choosing data that makes AMD look very good, and disregarding similar data from intel as "bunk" so intel cannot look good.
April 16, 2006 9:14:21 PM

Quote:
Overall, the trend in the data is not on your side.

I've used your own quote about Anand 6% figure and still I refute you with your own link.
On the other hand, Sharikou's blog acknowledged that conroe would perform to its days but once you use datasets that are larger than conroe's shared L2 cache, things start to look nasty for the chip. It's that hard for you to understand?

Quote:
and why are you so optimistic about AM2 and so pessimistic about conroe (this question might as well be rhetorical)? if review sites have conroe, they're surely under strict NDA's.

Just the way you're so optimistic about conroe but pessimistic about AM2. 8)
April 16, 2006 9:19:34 PM

Quote:
Just the way you're so optimistic about conroe but pessimistic about AM2. 8)


me being pessimistic or optimisitc about either is not the point. the point is you can't take data from AMD and praise it, then disregard the same data from intel.

I think AM2 is going to perform exactly how it's show. and frankly, I'm somewhat impressed they pulled 7% out of it considering it's just a plaform change.

as far as conroe, so far it looks to be good, but I'm with a lot of people in terms of waiting until someone like anand or tomshardware has it to do a full batch of benchmarks.
April 16, 2006 9:33:40 PM

Quote:
as far as conroe, so far it looks to be good, but I'm with a lot of people in terms of waiting until someone like anand or tomshardware has it to do a full batch of benchmarks.

The best post I've seen today. :wink:
April 16, 2006 10:43:35 PM

Quote:
nope, i think it was because of action man, im not sure though
No I started the keyboard thing. In one of my past posts I suggested that his keyboard was broken or something, but Action_Man thought it was absolutely comical and went around spreading the joke on every one of this guy's threads. He insists on stretching out a long old joke and now has everyone else doing it. :?
April 16, 2006 11:14:09 PM

Quote:
AM2 is indeed faster but is that enough to make it worthy of a good succesor. No. Because 2-3% in most apps is insignificant for most.
Why don't AMD fanboys accept that:
-DDR2 was a step AMD was gonna sooner or later
-while DDR2 is not great now,it'll get better later and it already has some obvious advantages over DDR

DDR2 was a time eater for both companies. AMD is just slower in the implentation process and also,remember, AMD also has to come up with a mem. controller. That's why people waited so much for so little. But it may seem little now,but by the begining of 2007 when real DDR2 will appear the performance gains might increase.
Still, until then Conroe will probably lead. I'm saving up right now for my future Conroe system. 2.4 Ghz chip seems very tasty for the price.


I think the big problem was that AMD was tight-lipped and most people who have been in PCs for the last 5-6 years have yet to see AMD NOT increase by 20% when they change sockets and/or RAM. I guess AMD just knows that their bread and butter is in the 4Way server space. They can make a lot more money maintaining their current perf @ 2.8GHz and 3GHz until K8L and 65nm. I would hope there would be a few more spins to get DDR2 EFFICIENCY higher but alas, I wasn't getting AM2 or Conroe anyway. I will upgrade to an FX60 later and move my 4400+ to my studio.
I still expect that they are working with mobo manuf to raise the efficiency of DDR2 bandwidth. There also is no talk of the HTT being raised to 333MHz which I thought would provide addtional bandwidth.

I guess they also realize that there are a lot of P4s to get rid of before Conroe can replace it so there is no threat until nearly 2007. They'd be fools to drop their prices at either rate since the only really "high-margin" desktop chips will be the FXs. The X2s will still be competitively priced with the mix of Conroe/PD/P4/CoreSoloDuo. Even just expanding the current architecture with more cache and registers and increasing HTT will put them right back up there.
April 17, 2006 1:27:38 AM

I'm not so worried that AMD is about to lose the initiative. I wouldn't mind some healthy competition from Intel though. Sooner or later AMD has to make a transition to faster memory technologies and during that process they naturally like to keep the older cheaper chips on the market place. That is probably why we see all those "new" chips to be presented on June 6th.

Does that mean that AMD is out of the game? Hardly! When it comes to technology Intel is ahead of AMD. They made the transition to DDR2 before AMD and will be quicker at adopting 65 nm processes. On paper, Intel ought to be ahead in performance as well, but in reality they are not. AMD has been able to strike some golden design mines along the way. In fact, Intel needs 65 nm to beat AMD's 90 nm designs.

Personally, I find it hard to believe that all architects at AMD need to be involved in adapting the memory controller. I don't even think it's the same group of people designing the memory controller and the cores. Do not let the large number of models fool you. What the processor core people are doing, I don't know, but they are not fiddling with memory controllers.
April 17, 2006 1:30:19 AM

X2? priced competitively? 8O

I don't think so.
April 17, 2006 1:30:35 AM

No, I am Spartacus!
!