Lab Update - DDR2 vs DDR3 performance with the AMD AM3 Processors

[onethreehill]

Talk about a labor of love. Our planned " quick update" on DDR2 versus DDR3 performance with the new AM3 processors has turned into a week of headaches, anxiety, and frustration for the most part. That said, it has actually been enjoyable for the most part as I have been off playing memory detective. Trying to find optimal and stable settings on our AM3 boards has been like playing a good game of hide and seek. Of course this hide and seek game has our AM3 platform hiding in New York and I started seeking for it and the answers to our problems in Alaska.

After several misfires with early boards, processors, and BIOS releases we are on the right track now. New retail boards and processors just arrived along with a slew of BIOS releases this past weekend with more to come tonight. I think we are at the point of starting the "official" benchmark process now and hopefully we will have our first results up this week.........................
Anandtech ( http://anandtech.com/weblog/showpost.aspx?i=559)

[570091D]

Speaking of the 720BE processor, I have to say this little jewel is currently my favorite bang for the buck processor. I have had an absolute blast testing this processor and when overclocked, it truly takes on a "devil with a blue dress" persona. I highly suggest you take a serious look at this processor if you are in the market for an upgrade right now.

wow, well i'm glad to see amd picking up the slack, and i'll be glad to give one a try when they get the ddr3 issues FULLY sorted.

[soundood]

sadly it looks as though the differances between the two, are just too marginal to justify the cost if you ask me.

that doubled with the infancy of the tech with AMD boards and chips, makes this a no go for me.

i sort of expected more, even a little more, maybe that is still to come :up:

[naokaji]

sadly it looks as though the differances between the two, are just too marginal to justify the cost if you ask me.

that doubled with the infancy of the tech with AMD boards and chips, makes this a no go for me.

i sort of expected more, even a little more, maybe that is still to come :up:

Agree, unless you are in the situation of having "left over" ddr3 from intels dual channel high voltage days it is not really worth it yet, might as well go with a AM2+ board that supports am3 cpu's.

[zanzabar]

i hate ddr2, just not having it is a good enough reason to change, and comparing non jdec ddr2 to jdec ddr3 u wont get good performance. but non jdec to non jdec the difference looks way more appealing tword ddr3, and ddr3 isnt that much u can get 4GB of ok ddr3 for $75 or so, and thats only $30 more than ok ddr2

[Hornet331]

sadly it looks as though the differances between the two, are just too marginal to justify the cost if you ask me.

that doubled with the infancy of the tech with AMD boards and chips, makes this a no go for me.

i sort of expected more, even a little more, maybe that is still to come :up:

No surpirse there, actually its right what I expected. There was never a huge gain in switching memory (for desktop apps).

Just compare the switch from DDR-400 cl2.5, people need at least ddr2-677 cl4 to match that.

Its the same now with ddr2, to match a good ddr2 kit (ddr2-1200 cl4) you would need ddr3-1333 cl5 (but theres no such ram, so you need to go for ddr3-1600 cl7)

DDR3 will beat DDR2, cause of its raw speeds that increases over time (just look at the upcoming ddr3-2xxx sticks), just like ddr2 beat ddr2 due its raw speed advantage.

[gOJDO]

So there is no performance boost at all? One more time AMD was BS-ing us with their charts.
http://www.theinquirer.net/img/11093/Pic2-Theoreticalperformanceincrease.jpg

[LexDiamonds]

So there is no performance boost at all? One more time AMD was BS-ing us with their charts.
http://www.theinquirer.net/img/11093/Pic2-Theoreticalperformanceincrease.jpg

I wouldnt say they were BSing us.... That graph shows 3-4% improvement for the move to ddr3. As bioses mature and some higher speed quads come out to better saturate the available bandwidth, I'm sure this is in the realm of possibility.

[Shadowmage]

So there is no performance boost at all? One more time AMD was BS-ing us with their charts.
http://www.theinquirer.net/img/11093/Pic2-Theoreticalperformanceincrease.jpg

The chart predicts a +4% increase in performance for DDR2->DDR3, you know. I suspect that it's because they compare standard DDR2 (JDEC) with standard DDR3.

[informal]

The chart predicts a +4% increase in performance for DDR2->DDR3, you know. I suspect that it's because they compare standard DDR2 (JDEC) with standard DDR3.

Yep and add on that the fact that many reviews mentioned AMD warned them about the current state of BIOS support for AM3.There are some issue on firmware level that will be ironed out soon(as AT mentions in that "lab update").

Now we need the 950BE to launch in April and make use of the better BIOS support and higher NB clocks(when compared to 940).It will be a great enthusiast chip for sure,only pricing needs to be aggressive IMO(all the chances are it will be).

[gOJDO]

It isn't mentioned the speed of the DDR2, but officialy Phenom I/II are supporting DDR2-1066, so it is DDR2-1066! The chart implies at 4% improvement of DDR3-1333 over DDR2, which is not the case so far. If AMD had any kind of a problems with the BIOS or something, why release a product with such problems?
Until AMD prove otherwise, they are BS-ing us (one more time) with their charts.

[Shintai]

GDDR5 is also slower than GDDR3 at the same speed if people remember. The chart is just for synthetics I bet. DDR2-1066 matches DDR3-1333 approx. 1066 aint JEDEC tho if I recall right. The point is ofcourse that DDR3 scales higher to 1600, 2000 etc (While using less power). Its the old DDR vs DDR2 again as a renewed rerun with DDR2 vs DDR3. 4%? Then it have to be DDR3-1600.

Remember DDR2 is at its peak now, DDR3 just started its first babysteps.

[v_rr]

Until AMD prove otherwise, they are BS-ing us (one more time) with their charts.

Is it?

Nevertheless, we tested the new processor in a Socket AM3 mainboard and saw that it was absolutely viable. DDR3 SDRAM with Phenom II processors provides a noticeable effect: about 3% performance improvement even compared with DDR2-1067 SDRAM.
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_17.html#sect0

[Shintai]

Is it?

http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_17.html#sect0

Untill you rewind...

http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_9.html
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_10.html
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_11.html
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_8.html
http://techreport.com/articles.x/16382/4
http://www.anandtech.com/weblog/showpost.aspx?i=559

The only place that draws up is here:
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_7.html

So ye, like any marketing slide from any company. They are BSing. Remember they also used a CPU to test with less cache. Hence bigger effect.

[msimax]

am3 is doing good with ddr3 hopefully it will get better

http://img27.imageshack.us/img27/6285/screenshot020da4.png (http://imageshack.us)
By deeodee (http://profile.imageshack.us/user/deeodee)

[Shadowmage]

Untill you rewind...

http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_9.html
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_10.html
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_11.html
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_8.html
http://techreport.com/articles.x/16382/4
http://www.anandtech.com/weblog/showpost.aspx?i=559

The only place that draws up is here:
http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_7.html

So ye, like any marketing slide from any company. They are BSing. Remember they also used a CPU to test with less cache. Hence bigger effect.

AMD always runs their DDR2 tests with 667MHz DDR2. I couldn't find the slide pack shown above, but here's a similar one that supports my point. Look at slide 22.

This makes sense, anyways. Isn't 1066MHz the slowest version of DDR3?

http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Shanghai_launch_deck_CONFIDENTIAL_Final_11_11_08.p df

[Hornet331]

This makes sense, anyways. Isn't 1066MHz the slowest version of DDR3?



nope lowest jedec spec ddr3 is ddr3 800.

[Shadowmage]

nope lowest jedec spec ddr3 is ddr3 800.

Lemme guess: it's slower than DDR2 800, right? That must be why no one is selling that on Newegg :p

It would be more fair if they compared the slowest popular DDR2 (667) to the slowest popular DDR3 (1066), or the fastest DDR2 (800) to the current fastest DDR3 (1600). Comparing DDR2 800 to DDR3 1066 simply will show very little speedups, especially due to the much higher latencies with the DDR3.

EDIT: Holy crap DDR3 goes up to 1000MHz...

[Hornet331]

2000mhz ddr3 is not the end, there will be even higher parts.

I think with the new low voltage parts we will see 2200mhz, maybe even more.

But all that bandwidth is still useless for desktop apps. :p

I notice no difference on my Ci7 regardless if i run the memory at 1200 or 1600 or 1800Mhz. They only thing i see a difference is in synthetic benchmarks.

[Bo_Fox]

Lemme guess: it's slower than DDR2 800, right? That must be why no one is selling that on Newegg :p

It would be more fair if they compared the slowest popular DDR2 (667) to the slowest popular DDR3 (1066), or the fastest DDR2 (800) to the current fastest DDR3 (1600). Comparing DDR2 800 to DDR3 1066 simply will show very little speedups, especially due to the much higher latencies with the DDR3.

EDIT: Holy crap DDR3 goes up to 1000MHz...

Nope, the slowest popular DDR2 is like 533. :p:

[gOJDO]

Is it?

http://xbitlabs.com/articles/cpu/display/phenom-ii-x4-810_17.html#sect0
Well, the gaming scenario is the best. The performance difference between DDR3-1333 7-7-7-20 and DDR2-1067 5-5-5-15 is:
CW 0,22%
FC2 1,82%
UT3 1,78%
WiC 1,49%
L4D 1,92%
-------------------
average 1,44%

If you take into account the rest of the benchmarks (excluding the synthetic) the difference in average is less than 1%.

These benchmarks are made with the 4MB L3 version. With the 6MB version the difference should be smaller.


AMD always runs their DDR2 tests with 667MHz DDR2. I couldn't find the slide pack shown above, but here's a similar one that supports my point. Look at slide 22.

This makes sense, anyways. Isn't 1066MHz the slowest version of DDR3?

http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Shanghai_launch_deck_CONFIDENTIAL_Final_11_11_08.p dfThat is for server CPUs. These officialy are supporting rDDR2-667, so it is the fastest RAM that should be used by specs in the dual/quad socket servers.


Nope, the slowest popular DDR2 is like 533. :p:it is DDR2-400.

[Shintai]

I´m sure we can all agree that DDR3 only gets better in the future. But its a no brainer that DDR3 struggles against highend DDR2 today. However DDR2 aint gonna go any further. DDR3 will "leap ahead". :p:

[v_rr]

Well, the gaming scenario is the best. The performance difference between DDR3-1333 7-7-7-20 and DDR2-1067 5-5-5-15 is:
CW 0,22%
FC2 1,82%
UT3 1,78%
WiC 1,49%
L4D 1,92%
-------------------
average 1,44%

If you take into account the rest of the benchmarks (excluding the synthetic) the difference in average is less than 1%.

These benchmarks are made with the 4MB L3 version. With the 6MB version the difference should be smaller.

That is for server CPUs. These officialy are supporting rDDR2-667, so it is the fastest RAM that should be used by specs in the dual/quad socket servers.

it is DDR2-400.

You are comparing DDR3-1333 7-7-7-20 vs DDR2-1067 5-5-5-15.
As far as I konw you have another Bar with DDR3-1600Mhz and you are excluding it from your math. Right... so if you exclude DDR3-1600Mhz you also exclude DDR2-1067 and compare DDR3-1333 vs DDR2-800.

Also you are excluding synthetic and that is wrong. You are manipulating review with your own personal settings and valid benchmarks.

And as far as I know the CPU with 4Mb L3 is sold on market, so the score is valid.

You have to take also into account that the AMD slide should explore the high average scenario using server benchmarks where the extra wide-band make a big difference.

Review with lots of benchmarks to a wide range of applications (server, non-server, professional, non-professional):
http://img4.imageshack.us/img4/295/38744082gu0.jpg
http://ixbtlabs.com/articles3/cpu/amd-phenom-2-720-810-920-p4.html

[ToTTenTranz]

I´m sure we can all agree that DDR3 only gets better in the future. But its a no brainer that DDR3 struggles against highend DDR2 today. However DDR2 aint gonna go any further. DDR3 will "leap ahead". :p:

Yes, the same way the last generation DDR had better price-performance ratio than the first generation DDR2, mainly because of lower latencies. The past has shown us that we change memory architecture because of the newcomer's potential, not because of its out-of-the-box advantage.

GDDR is a completely different story, though.

[gOJDO]

You are comparing DDR3-1333 7-7-7-20 vs DDR2-1067 5-5-5-15.
As far as I konw you have another Bar with DDR3-1600Mhz and you are excluding it from your math. Right... so if you exclude DDR3-1600Mhz you also exclude DDR2-1067 and compare DDR3-1333 vs DDR2-800.BS. Phenom II officially supports DDR3-1333. Anything above means OC-ing your CPU.

Also take a look at the AMD chart! It clearly says DDR3-1333!


Also you are excluding synthetic and that is wrong.Synthetic are utter BS. And those included in the x-bit article are not measuring CPU performance. Only synthetic CPU bandwidth.


You are manipulating review with your own personal settings and valid benchmarks.I'm not manipulating anything. Read Anand's conclusion, read the charts and the conclusions of the rest of the articles.


And as far as I know the CPU with 4Mb L3 is sold on market, so the score is valid.BS. Take a look at the AMD slide again! It clearly says "additional 4MB L3"


You have to take also into account that the AMD slide should explore the high average scenario using server benchmarks where the extra wide-band make a big difference.Even more BS. The title clearly says "AMD Phenom II Performance Uplift". It's all about DESKTOP!


Review with lots of benchmarks to a wide range of applications (server, non-server, professional, non-professional):http://img4.imageshack.us/img4/295/38744082gu0.jpg:rofl: In that article there is no comparison of Phenom II with DDR2 vs DDR3(which we were talking about, and is the toppic of this thread) You fail to comprehend basic things...I'm done with you.

[v_rr]

BS. Phenom II officially supports DDR3-1333. Anything above means OC-ing your CPU.

Also take a look at the AMD chart! It clearly says DDR3-1333!

Synthetic are utter BS. And those included in the x-bit article are not measuring CPU performance. Only synthetic CPU bandwidth.

I'm not manipulating anything. Read Anand's conclusion, read the charts and the conclusions of the rest of the articles.

BS. Take a look at the AMD slide again! It clearly says "additional 4MB L3"

Even more BS. The title clearly says "AMD Phenom II Performance Uplift". It's all about DESKTOP!
BS :rolleyes:

- Where on the AMD chart says DDR2-1066 for DDR2 comparison?
- PH2 and mobo makers support DDR3-1600Mhz so it´s valid to use it.
- Synthetic are benchmarks as any benchmark you cant remove some benchs and put other at your personal will.
- DESKTOP? What are desktop apps? As far as I know I have all desktop computers and I use Web servers (Apache Tomcat, GlassFish, JBoss), Maple, MySql, Oracle, Java EE-SE-ME, Visual Studio C,C++,C#, etc etc and I can be here all day naming apps that 90% of reviews don´t care.
So the Synthetic is a way to show bandwidh advantage. The review I put here show professional and server apps. Is a good review.
Bandwidh advantage is much more noticed in server apps.

Your way to show DDR3 vs DDR2 is only Games. Who told you that everyone played those games? You told you that Games is the only way to compare DDR2 vs DDR3?

[xVeinx]

So, in the post by the op, it notes that new processors, boards, and bioses are coming in. That processors is mentioned has me curious; is a new stepping needed here? Again, the assumptions at this point have been that it's a bios problem, but is there work that needs to be done on the memory controller for things to become more stable? I may be reading too much into the comment of course, but it does have me curious.

[informal]

So, in the post by the op, it notes that new processors, boards, and bioses are coming in. That processors is mentioned has me curious; is a new stepping needed here? Again, the assumptions at this point have been that it's a bios problem, but is there work that needs to be done on the memory controller for things to become more stable? I may be reading too much into the comment of course, but it does have me curious.

New stepping is indeed underway(C2A),but not due to this "reason".This is a bios/firmware issue like AMD already told madshrimp.be,not a chip/"silicon" problem(issue is with dimm supply voltage and 1333Mhz mode).BIOS update(s) are underway and will address this.