AM2 review

[zir_blazer]

Socket M2 Preview
Tom's Hardware (http://www.tomshardware.com/2006/02/21/a_look_at_amds_socket_am2_platform/)

I don't like the fact that the Preview was done by Tom's Hardware because I still don't forget the bad reputation that they earned in old times, it could have been done by Tech Report, X-Bit Labs or Anandtech.
My early speculations where entirely right: 200 MHz Base Clock, supporting higher Frecuency DDR-II Memory Modules using Divisors (They're Quad Pumped), and no performance increase compared to Socket 939 (Even it performs slighty slower though I was expecting slighty higher. We should still wait for DDR-II 800 MHz Bus support to see the real final numbers). Everyone was waiting for Socket M2? Now, "Enjoy" it.

Source: Tom's Hardware (http://www.tomshardware.com/)

[krille]

Hey zir_blazer, come join us here (http://www.xtremesystems.org/forums/showthread.php?t=89853&page=2)for an ongoing discussion in the News section (it's been there for a while)
http://www.xtremesystems.org/forums/showthread.php?t=89853&page=2

We're still not sure whether this is one of the M2 chips with the malfunctioning MC or not. But if it has the fixed MC, this is really depressing results. Otherwise, who knows?

[zir_blazer]

Oh damn, got defeated in the News Thread posting! :( Anyways, most people would miss it if you don't say it in big letters than there is already a Preview of it wandering around :P

[GMX]

"the results show the basic performance difference between current Socket 939 systems at DDR400 speed and ideal CL2.0-2-2-5 1T timings, versus DDR2-667 at average CL4-4-4-12 timings."

Is this fair?
I don't know about DDR2, but it sounds like they are comparing something not average(ie 2-2-2-5 ddr400) to something average (4-4-4-12 is average/JEDEC reference, right?)

[ozzimark]

woooooooowwwwwwwwwwww :stick: :stick: :stick: :stick:

look at this. they really expect us to believe that this is truely the case and the rig isn't misconfigured? :slapass:
http://images.tomshardware.com/2006/02/21/a_look_at_amds_socket_am2_platform/image023.gif

[GMX]

woooooooowwwwwwwwwwww :stick: :stick: :stick: :stick:

look at this. they really expect us to believe that this is truely the case and the rig isn't misconfigured? :slapass:
http://images.tomshardware.com/2006/02/21/a_look_at_amds_socket_am2_platform/image023.gif


I get a server internal error.

[Serge84]

Who cares, our current systems will be better for along time so I won't give up my stuff for the newest when I can just out perform it with so called out of date tech. :rolleyes: Apperently the last stuff made for this gen will be more advanced then the next gen for a good wile.

Until AMD goes to 65nm and 4mb of L2 well then the stuff on M2 is no better then what we have in current tech and it appears our current systems are just better so thats a good thing. DDR2 is still not as good as DDR1. Because DDR can still be made as fast timming wise as the fastest DDR2 at 900mhz that I know of. DDR1 at 600mhz or 700mhz with low timmings is alot better then the horrible timmings that DDR2 800 and 900 use. DDR1 isn't out of date DDR2 is a failing tech until they can make the timmings better. DDR3 will be even worse to start off with in 2 years from now.

The only advantage DDR2 and DDR3 should stay in is video cards. You could get 2000mhz from DDR3 but what advantage is that when the timmings look like this. 10-12-15-26 and so on. Its worse sure it has insane bandwidth but only is useful graphicaly. DDR2 is just starting to maybe get good within the next 3 more years. Don't expect to see any mericals with M2. its been proven time and time again that DDR and amd platforms will kill intels anyday with ddr2. As long as DDR is made it will rule and continue to get better and rival over DDR2. because guess what, DDR is almost as fast as the fastest DDR2 and it has much lower timmings in the 2.5 range. Witch kills DDR2 right there no matter how fast DDR2 is. 3 and 4 timmings will not cut it. Its sad DDR a so called out of date tech is matching and wooping DDR2 across the road. DDR700 and so on will come if not aready. Since ppl can OC there DDR600 to DDR700 and get 10,000mb/s 12,000mb/s dual channeled or more well then DDR2 is no better then DDR1 in the end. lol

May even make M2 fail because not very many ppl will go to it at 1st if not at all. It would be silly to give up your 939 or 754 for a tech thats only going to give you the same performance if not alittle less. Atleast at the molment. I'll be intrested when I see dirt cheap dual cores on 939's. But until I see M2 get 4 cores and 4mb L2 and DDR2 with decent timmings and speeds over 1000mhz and 65nm tech I don't even think I'll give up my current system for the next few years. :slap:

[GMX]

Who cares, our current systems will be better for along time so I won't give up my stuff for the newest when I can just out perform it with so called out of date tech. :rolleyes: Apperently the last stuff made for this gen will be more advanced then the next gen for a good wile.

Until AMD goes to 65nm and 4mb of L2 well then the stuff on M2 is no better then what we have in current tech and it appears our current systems are just better so thats a good thing. DDR2 is still not as good as DDR1. Because DDR can still be made as fast timming wise as the fastest DDR2 at 900mhz that I know of. DDR1 at 600mhz or 700mhz with low timmings is alot better then the horrible timmings that DDR2 800 and 900 use. DDR1 isn't out of date DDR2 is a failing tech until they can make the timmings better. DDR3 will be even worse to start off with in 2 years from now.

The only advantage DDR2 and DDR3 should stay in is video cards. You could get 2000mhz from DDR3 but what advantage is that when the timmings look like this. 10-12-15-26 and so on. Its worse sure it has insane bandwidth but only is useful graphicaly. DDR2 is just starting to maybe get good within the next 3 more years. Don't expect to see any mericals with M2. its been proven time and time again that DDR and amd platforms will kill intels anyday with ddr2. As long as DDR is made it will rule and continue to get better and rival over DDR2. because guess what, DDR is almost as fast as the fastest DDR2 and it has much lower timmings in the 2.5 range. Witch kills DDR2 right there no matter how fast DDR2 is. 3 and 4 timmings will not cut it. Its sad DDR a so called out of date tech is matching and wooping DDR2 across the road. DDR700 and so on will come if not aready. Since ppl can OC there DDR600 to DDR700 and get 10,000mb/s 12,000mb/s dual channeled or more well then DDR2 is no better then DDR1 in the end. lol

May even make M2 fail because not very many ppl will go to it at 1st if not at all. It would be silly to give up your 939 or 754 for a tech thats only going to give you the same performance if not alittle less. Atleast at the molment. I'll be intrested when I see dirt cheap dual cores on 939's. But until I see M2 get 4 cores and 4mb L2 and DDR2 with decent timmings and speeds over 1000mhz and 65nm tech I don't even think I'll give up my current system for the next few years. :slap:

Whats with your AMD fanboyism??? Nearly every thread you are making negative comments of Intel.

Maybe the best kit of ddr in the world will get 10K+ bandwidth, Most people get around 6gb to 8.5gb. Can you show me 12K?

Did you read the preview? They tested DDR2 667 at 4-4-4-12 and managed slightly slower resultsthan ddr1 400 2-2-2-5. Now just think what ddr-800 will do.

Do you work for AMD? Or all this rubbish information is pulled out of your arse?

[[XC] leviathan18]

http://www.tomshardware.com/2006/02/21/a_look_at_amds_socket_am2_platform/


now i want to know its going to use DDR2 or DDR3 (this is for the DDR3 supporters)

[turtle]

Interesting, although i'm not suprised by the benchmarks. I agree with the writer on several key issues.

*DDR2-800 is key to show improvements over the current platform.

*It boils down to how this performs with the 65nm shrink (vs Conroe).

*Most importantly, it suffers from the loose timings of DDR2, where Socket A/754/939 all benefitted heavily from the tight timings of DDR1 (especially cas2). When high-speed Cas3 DDR2 becomes available, it will most-likely help the platform immensely...Probably much more so than Intel's product.

I know i've said it before, but its good to see someone (Even if it's Tom's) agree with me. :)

[Syn.]

http://www.tomshardware.com/2006/02/21/a_look_at_amds_socket_am2_platform/


now i want to know its going to use DDR2 or DDR3 (this is for the DDR3 supporters)

AMD is trying to make a move to DDR2 to at the moment thinking about DDR3 is crazy.

Interesting, although i'm not suprised by the benchmarks. I agree with the writer on several key issues.

*DDR2-800 is key to show improvements over the current platform.

*It boils down to how this performs with the 65nm shrink (vs Conroe).

*Most importantly, it suffers from the loose timings of DDR2, where Socket A/754/939 all benefitted heavily from the tight timings of DDR1 (especially cas2). When high-speed Cas3 DDR2 becomes available, it will most-likely help the platform immensely.

I know i've said it before, but its good to see someone (Even if it's Tom's) agree with me. :)

Considering that AMD can match the TDP that Intel is suppose to get with Conroe at 65nm with thier 90nm process i dont think it really matters.

As Tom keeps saying that review is of an enginering sample and the only good thing they can deduct is that its 100% stable. One thing i am happy to hear is that DDR2 2GB kits will be cheap which means that i will be able to get my target 4GB for my Vista pc for alot less, i just hope that memory timings can be sorted out on DDR2-800.

[Thorry]

As they state on page 1: Nothing new....

All of this has been known for many months, and since (as expected) performance isn't different between 939 and AM2 (except for offcourse the extra memory bandwidth which may help or may not depending on your use) benchmarks aren't that interesting.

[largon]

DDR1 at 600mhz or 700mhz with low timmings is alot better then the horrible timmings that DDR2 800 and 900 use. DDR1 isn't out of date DDR2 is a failing tech until they can make the timmings better. DDR3 will be even worse to start off with in 2 years from now.
- -
You could get 2000mhz from DDR3 but what advantage is that when the timmings look like this. 10-12-15-26 and so on.:rolleyes:
CAS 10 @ DDR3 2000 = CAS4 @ DDR2 800 = CAS2 DDR1 400 = 10ns

It seems people just stare at the numeric value of the latency and don't think what it means...

Absolute latency (time) of CAS4 @ DDR2 800 is exactly the same as CAS2 DDR1 400. Both are 10ns. All the CPU cares is the time it has to wait for the requested data. Amount of clock cycles (numeric latency) don't mean anything - as long as frequency is higher, ie. clock cycle is shorter.

[CandymanCan]

:/ well looks like Ill be keeping my s754 system for along time to come

[Sentential]

Hmm this is strange... why is the HTT showing 200? :stick: Ive been told multiple times that it would be 333. Plus if you compare to conroe's 266 having a chip with a base HTT of 200 makes no sense at all.

CPUz is probably reading it incorrectally like half hte programs out there.

[Repoman]

Why not? Base HTT makes no difference in performance.. they can just use larger mem dividers to get the higher speeds.

I think they did something wrong, look at the Sandra mem benchmarks.. DDR2-667 gets beat by DDR400

[agenda2005]

CPU-Z gave nonesense readings in almost all the specs. It might look like THG was running at DDR2-400 and very high latency of 4-4-6. But, time will tell if the extra bandwith from a correct DDR2-800 can make any significant boost.

[Vapor]

If it's going to help anywhere, we'll see it on dual cores.

[HARDCORECLOCKER]

:D Didn't somebody say THG is with the stupid..........???

:toast:

[dickeywang]

Who cares, our current systems will be better for along time so I won't give up my stuff for the newest when I can just out perform it with so called out of date tech. :rolleyes: Apperently the last stuff made for this gen will be more advanced then the next gen for a good wile.

Until AMD goes to 65nm and 4mb of L2 well then the stuff on M2 is no better then what we have in current tech and it appears our current systems are just better so thats a good thing. DDR2 is still not as good as DDR1. Because DDR can still be made as fast timming wise as the fastest DDR2 at 900mhz that I know of. DDR1 at 600mhz or 700mhz with low timmings is alot better then the horrible timmings that DDR2 800 and 900 use. DDR1 isn't out of date DDR2 is a failing tech until they can make the timmings better. DDR3 will be even worse to start off with in 2 years from now.

The only advantage DDR2 and DDR3 should stay in is video cards. You could get 2000mhz from DDR3 but what advantage is that when the timmings look like this. 10-12-15-26 and so on. Its worse sure it has insane bandwidth but only is useful graphicaly. DDR2 is just starting to maybe get good within the next 3 more years. Don't expect to see any mericals with M2. its been proven time and time again that DDR and amd platforms will kill intels anyday with ddr2. As long as DDR is made it will rule and continue to get better and rival over DDR2. because guess what, DDR is almost as fast as the fastest DDR2 and it has much lower timmings in the 2.5 range. Witch kills DDR2 right there no matter how fast DDR2 is. 3 and 4 timmings will not cut it. Its sad DDR a so called out of date tech is matching and wooping DDR2 across the road. DDR700 and so on will come if not aready. Since ppl can OC there DDR600 to DDR700 and get 10,000mb/s 12,000mb/s dual channeled or more well then DDR2 is no better then DDR1 in the end. lol

May even make M2 fail because not very many ppl will go to it at 1st if not at all. It would be silly to give up your 939 or 754 for a tech thats only going to give you the same performance if not alittle less. Atleast at the molment. I'll be intrested when I see dirt cheap dual cores on 939's. But until I see M2 get 4 cores and 4mb L2 and DDR2 with decent timmings and speeds over 1000mhz and 65nm tech I don't even think I'll give up my current system for the next few years. :slap:

I agree with you that, at least for now, DDR performance better than DDR2. The sad thing is, most RAM manufactures are going to switch to DDR2. It's nothing about the performance but $$$:mad: . Another thing mentioned in the preview is that the new DDR2 will have higher capacity (2GB or more per stick), I know 2GB RAM sounds more than enough right now, but with the M$ Vista coming in the next 12months or so, things may change a lot. I had a rig with 256MB RAM running Win98SE smoothly back to 2001, but had to get 512MB RAM when I switch to WinXP, and now we all think 2GB is standard.

[nn_step]

AMD is trying to make a move to DDR2 to at the moment thinking about DDR3 is crazy.
You forget Intel is Dropping DDR2 completely for DDR3 Q4(06)/Q1(07) :rolleyes:

[Cooper]

If it's going to help anywhere, we'll see it on dual cores.

THG reviewed Dual-core ;)
Well 200 HTT does look strange. Maybe CPU-Z can`t tell a single thing right about AM2 except it`s been ES :D
I`m not pleased with performance - expected to see 5-10% increase in some multythreaded tasks :(

[OmegaMerc]

Good news, this means I don't have to upgrade at all for at least another year.:banana:

[[XC] leviathan18]

this means a ES performs the same as one production sample 939... we have to see production samples of the AM2

[Cooper]

With your current setup it`s a total watse of money even if AM2 had (or it really will) 5-10% performance gain. Just get yourself better cooling and OC your Opty :)

[Vapor]

THG reviewed Dual-core ;)Yes, sorry...I was talking about DDR2-800 vs. what was just reviewed (DDR2-400? or was it actually DDR2-667? duumbasses didn't do a bandwidth test).

[zir_blazer]

Don't expect 800 MHz Bus DDR-II do wonders either. Everyone seems to forget that the K8 Architecture isn't Memory Bandwidth dependant (Or Cache for that matter). Going from Socket 754 in Single Channel (3.2 GB p/s) to Socket 939 in Dual Channel (6.4 GB p/s) is worth in average 200 MHz in performance, though is fairly dependant on the application likeness for Memory Bandwidth and is more common to see applications that can do more with a 200 MHz boost than that extra Memory Bandwidth.
So, who is expecting actually than the extra Memory Bandwidth that DDR-II gives improves performance, when it seems that 3.2 GB p/s is enough to feed decently a K8 Core? Anandtech already did a Review using the extra Divisors from K8s Rev. E to support 533 MHz Bus DDR Memory Modules and the extra performance was around less than 3%, and maybe 5% for Dual Core (Meaning that slighty more that 3.2 GB p/s is what a high end K8 Core needs today). Not wonderful.

[nn_step]

Yes, sorry...I was talking about DDR2-800 vs. what was just reviewed (DDR2-400? or was it actually DDR2-667? duumbasses didn't do a bandwidth test).
They reviewed effectively DDR2-400.. which is completely retarded..
it is identical to a DDR-400 setup with :banana::banana::banana::banana:ty timings...
and just a reminder
DDR is from 100mhz-200mhz (according to the standard)
DDR2 is from 200mhz-400mhz (according to the standard)
DDR3 is from 400mhz-800mhz (according to the standard)
and a part of the lawsuit that is going on between Intel and AMD is over if DDR3 is going to be pin compatable with DDR2:fact:

[ozzimark]

So, who is expecting actually than the extra Memory Bandwidth that DDR-II gives improves performance
there's the problem. look at the last graph on page 11. the memory bandwidth is the same when we would expect a healthy increase due to the 133mhz increase in memory clock. this is where i'm calling shens, tom's did the entire review at ddr2-400 while saying it's ddr2-667, whether they knew it or not.

[Vapor]

Don't expect 800 MHz Bus DDR-II do wonders either. Everyone seems to forget that the K8 Architecture isn't Memory Bandwidth dependant (Or Cache for that matter). Going from Socket 754 in Single Channel (3.2 GB p/s) to Socket 939 in Dual Channel (6.4 GB p/s) is worth in average 200 MHz in performance, though is fairly dependant on the application likeness for Memory Bandwidth and is more common to see applications that can do more with a 200 MHz boost than that extra Memory Bandwidth.
So, who is expecting actually than the extra Memory Bandwidth that DDR-II gives improves performance, when it seems that 3.2 GB p/s is enough to feed decently a K8 Core? Anandtech already did a Review using the extra Divisors from K8s Rev. E to support 533 MHz Bus DDR Memory Modules and the extra performance was around less than 3%, and maybe 5% for Dual Core (Meaning that slighty more that 3.2 GB p/s is what a high end K8 Core needs today). Not wonderful.It may not be bandwidth dependent, but why bottleneck it? Give it what it wants...DDR2 can provide most of the bandwidth it'd ever need while also clock high enough with modest timings to not lose performance to latency (in terms of nanoseconds). In Rosetta we see efficiency of credit per core declining on dual cores with even 300MHz 2.5-3-3. There are CERTAINLY gains to be had, especially when you DOUBLE the bandwidth (from DDR2-400 4-4-4) and keep the timings the same (so essentially halving them in regards to nanoseconds).

I'm surprised that the AM2 chips did as well as they did.

p.s., that AT review you allude to, where they increase memory clocks 25%, and provides a (if you round down) 5% boost for dual cores, is quite substantial. Most people here would love a 5% boost if it were available ;)

[Sentential]

The numbers they put up seem wrong across the board. Just to clarify the NDA has not expired nor will it for quite some time :slapass:

Somehow I think intel is involved with this....

[nn_step]

The numbers they put up seem wrong across the board. Just to clarify the NDA has not expired nor will it for quite some time :slapass:

Somehow I think intel is involved with this....
Agreed heck The last TIme I check the M2's NDA doesn't expire till 2/16/2007 :fact:
So I definately question this review:mad:

[Sentential]

Agreed heck The last TIme I check the M2's NDA doesn't expire till 2/16/2007 :fact:
So I definately question this review:mad:
The divider isssues they are seeing are a problem but the production level chips look nothing like this at all. I really think there is some alternative motive behind them posting this.

I'm hoping AMD will force them to take this rubbish down.

[nn_step]

The divider isssues they are seeing are a problem but the production level chips look nothing like this at all. I really think there is some alternative motive behind them posting this.

I'm hoping AMD will force them to take this rubbish down.
They should leave it up so that they are burned horribly by this...http://img.myyearbook.com/zenhex/images/smilies/burn4.gif

[Vapor]

While I don't usually buy into conspiracy theories of any sort....this one is plausible at worst. That said, it's still not bad

I was satisfied by the results. :shrug:

[turtle]

I'm not. I won't be either...Until I see how this or better performs:

http://www.pcconnection.com/ProductDetail?sku=5777450&SourceID=k15506

[nn_step]

I'm not. I won't be either...Until I see how this or better performs:

http://www.pcconnection.com/ProductDetail?sku=5777450&SourceID=k15506

Got a good point.. put it with some D9 fatbody chips and see what it can do.. if it is really DDR2 they should beable to run it at 400mhz with half descent timings...

[turtle]

Yep. I overlooked the whole fatbody topic as DDR2 wasn't especially important to me until recently, thanks for the semi-enlightenment on the matter. The whole point was, which thankfully you understood, was to test an AM2 at 3-2/3-2/3-8 at a speed comparible to it at 4-4-4-10. Wasn't there a huge amount of hooplah about the support of Cas3 on AM2?

I bet the differance is huge since it's based off of A64's...On top of that, Cas3 should scale better as DDR2 matures, as opposed to cas2 on DDR1, shouldn't it? I wonder how high it will go...

[Vapor]

Compatibility is an issue so far....I think there's a reason Mushkin likes their new sticks so much ;)

[nn_step]

Yep. I overlooked the whole fatbody topic as DDR2 wasn't especially important to me until recently, thanks for the semi-enlightenment on the matter. The whole point was, which thankfully you understood, was to test an AM2 at 3-2/3-2/3-8 at a speed comparible to it at 4-4-4-10. Wasn't there a huge amount of hooplah about the support of Cas3 on AM2?

I bet the differance is huge since it's based off of A64's...On top of that, Cas3 should scale better as DDR2 matures, as opposed to cas2 on DDR1, shouldn't it? I wonder how high it will go...
Actually DDR2 is spec'd for 200-533 MHz right now and with all do respect DDR3 is spec'd for 400-800 MHz..
and if the Intel/AMD turns out right.. DDR2 and DDR3 will pin compatable...

[[XC] leviathan18]

if they are pin compatible this mean the mem ctrl of the AM2 is going to support both formats?

[krille]

if they are pin compatible this mean the mem ctrl of the AM2 is going to support both formats?lol mass cookie exchange :slobber:

[Vapor]

No, just pin compatible.

Eventually there might be an A64 on AM2 that can support both (but not at the same time I imagine). That's still if the courts push them to be pin compatible.

[[XC] leviathan18]

IF... so for now im winning the cookie

[Kunaak]

it's THG for gods sake...

the same people that tried to rip off JCViggen, with thier "Iceman Cometh" article, where they tried to claim to be the first at 4 ghz... when JC was already doing it, for like 6 months, and way ahead on the ORB with 4+ ghz benchmarks.

THG really should only be taken for a grain of salt...
don't put much faith in what you see there.

wait for anandtech, they always do good articles. :)

this article was half ass, and fairly misleading.
wait till overclockers get AM2, and some good Cas 3-2-2-4-1T ram, and start pushing 400+ mhz at 1:1 with it.

not this BS, where they compare DDR2 running at 300 mhz with a 4:5 or 2:3 divider... which is useless
comparing cas 2-2-2 ram against cas 4-4-4 ram, what are they thinking?
the ram running faster then the CPU's FSB is pointless, it does nothing but look cute on a screen...

the real test, will come when we see a good AM2 board from DFI come out, and someone push the ram at 1:1.

thats when things will get interesting :D

[HARDCORECLOCKER]

it's THG for gods sake...

:D

:D Yep - see post #19........;)

:toast:

[Cybercat]

Didn't have the patience to read the whole thread, but couldn't this conceivably be one of the first AM2s with DDR2 performance issues? They are supposed to be working on, or just finished a revision that fixes that, thus the delays.

[nn_step]

IF... so for now im winning the cookie
if The lawsuit goes AMD's way.. looks like we will be exchanging Cookies :D

[FUGGER]

Ive been telling you AMD guys all along about the BS 4-4-4-15 and 5-5-5-x timing on Intel vs AMD benches.

There is no reason not to run tight timing at very high speeds.

[Kunaak]

Ive been telling you AMD guys all along about the BS 4-4-4-15 and 5-5-5-x timing on Intel vs AMD benches.

There is no reason not to run tight timing at very high speeds.

then why do you ;)

+ isn't it pretty much a well known fact, that P4's just aren't that sensitive to latency, while on the other hand, A64's are extremely sensitive to latency? :D

[nn_step]

Honestly there are going to be a few telling things coming out shortly...

[FUGGER]

Why do I run 4 gigs or DDR2 at 400FSB 1:1? because AMD cant...

The biggest gain for AMD will be the ability to go far beyond 300FSB at 1:1, heel if you can reach 500FSB, my new Micron does that at 3-3-3-4 with all slots filled.

large denominations of DDR is a mess on AMD

[nn_step]

I think that has more to do with the Flaws in DDR and Windows than it has to do with the Flaws in AMD..

[Kunaak]

Why do I run 4 gigs or DDR2 at 400FSB 1:1? because AMD cant...

how do you know?

I can hit HTT speeds of over 455 just booting into windows... and thats with the 5 and 6 multipliers, and thats the max my Board will do from bios. with clockgen I can get up to about 485, before my OS quits on me.
with the 5 multiplier obviously...

obviously, with DDR, I am held back signifigantly.

I do wonder, if that would be the case, with DDR2 and a A64 :)

if AM2's OC as well as my current board will go, then 400 at 1:1 shouldnt be a big deal for a A64.
afterall, whats the record for HTT speed right now?
530? or something like that?

imagine that with DDR2 at 1:1... after all, there is 1 Ghz DDR2 out there ;)

so how do you know it can't do it?

[HARDCORECLOCKER]

:D Problem here is when all discussions about DDR-2 have endet DDR-3 is already born..........;)

:toast:

[ozzimark]

not this BS, where they compare DDR2 running at 300 mhz with a 4:5 or 2:3 divider... which is useless
comparing cas 2-2-2 ram against cas 4-4-4 ram, what are they thinking?
the ram running faster then the CPU's FSB is pointless, it does nothing but look cute on a screen...

the real test, will come when we see a good AM2 board from DFI come out, and someone push the ram at 1:1.

thats when things will get interesting :D
are you implying that running faster than 1:1 decreases memory effeciency with an integrated memory controller? ;)

[Kunaak]

are you implying that running faster than 1:1 decreases memory effeciency with an integrated memory controller? ;)

not even close.

what I was saying is, when I use to run a Asus 955X board, I was always trying different mem ratios, like 2:3 and 4:5 and I kept getting what I thought were defective results.

while I could get the ram as high as 350 mhz at cas 3-2-2-4 and such... I was getting zero benifit from it.
I even got to the point where I made threads about it in the Intel area, where I was asking what the problem could be, cause I figured... faster ram, must have some sort of benifit.
the general answer I got, was, ram, running faster then the FSB, does nothing.

like if your ram was running at 500 mhz, but your FSB is only 400 mhz, then you really won't see any benifit from that extra 100 mhz on your ram.

afte reading that from people with alot of intel experience, then alot of my own testing later to see if I was wrong, I came to the conclusion that ratios higher then your FSB are pretty much useless.

if you have a A64 revision E you can test this yourself.
DFI board come with higher ratios then 1:1 and if you have some DDR 500 or so, you can try it yourself.
you will see exactly what I mean.
running your ram higher then your HTT gives you no extra bandwidth, and while it looks nice, it does nothing.

[FUGGER]

AM2 is not out yet, thats why I say AMD cant do it now.

Maybe when it is 400+ FSB 1:1 will be possible. But as it stands right now you are forced to ratio down the memory speed.

Whats the current highest 1:1 on AMD, 330ish?

I do know 3-3-3-x @ 400+ 1:1 on AM2 will beat any current AMD bandwidth bench posted to date. This will greatly help dual cores

[nn_step]

Yeah its 334 but that by large has to do with the limitations of DDR not AMD

[HARDCORECLOCKER]

AM2 is not out yet,

:D Does that mean You'll switch to AMD in June???......:ROTF: :cheer: :nuts:

:toast:

[largon]

There's no sense in talking about K8 and 1:1.
It's just silly because there is no 1:1 on A64, there's always a divider involved because memory frequency is derived directly from cpu frequency, mem clock in K8 system has nothing to do with HTT (calling it FSB is just wrong).

What you call "1:1" means absolutely nothing for K8 IMC.


You all know this, but still you keep talking about 1:1 and FSB. :confused:

[FUGGER]

No, Conroe XE for me.

[Kunaak]

There's no sense in talking about K8 and 1:1.
It's just silly because there is no 1:1 on A64, there's always a divider involved because memory frequency is derived directly from cpu frequency, mem clock in K8 system has nothing to do with HTT (calling it FSB is just wrong).

What you call "1:1" means absolutely nothing for K8 IMC.


You all know this, but still you keep talking about 1:1 and FSB. :confused:

if you feel like nit picking, then your kinda in the wrong place.
this isn't a place where this kinda lingo goes over peoples heads.

FSB is the same as HTT.
its just a habit in my case.
I can almost garauntee, that anyone reading my posts, knows what I mean.
its not like anyones reading my posts going "oh I didn't know A64's had FSB's". its just a part of the language of overclocking.
nitpicking little details like that, is just detracting away from what was actually said.

[HARDCORECLOCKER]

No, Conroe XE for me.

:cool: Hm - I expected somethin' like that......, but sometimes a lite comes down to the blindet....... :rofl:

:toast:

[ozzimark]

if you have a A64 revision E you can test this yourself.
DFI board come with higher ratios then 1:1 and if you have some DDR 500 or so, you can try it yourself.
you will see exactly what I mean.
running your ram higher then your HTT gives you no extra bandwidth, and while it looks nice, it does nothing.
sure thing. watch this... timings are held constant throughout testing, except MAL and RP, which are set to auto to allow consistant booting ;)

200x10, ram at 1:1, for cpu/10 to get 200mhz:

http://eclipseoc.com/phpwcms_filestorage/1/1_179.png

nothing unexpected here. the low cpu mhz is preventing a high efficiency, so i'm a bit under the 6400mb/s mark.


now, to bump the memory up without touching fsb or cpu clock, i run 200x10 with the ratio set to 5:4, which gives cpu/8 for 250mhz (actually just a touch over, damn 200.9mhz htt :slapass:)

http://eclipseoc.com/phpwcms_filestorage/1/1_180.png



hmm, not a whole lot better. perhaps your words have some weight.. but just to make sure... 250x8, to keep cpu speed constant, and the ram at 1:1 to give cpu/8 for 250mhz ram...

http://eclipseoc.com/phpwcms_filestorage/1/1_181.png

hrm. not a whole lot better.. it's actually ~20mb/s slower, though part of that is due to the slightly slower ram clock speed. guess the ratio doesn't have a whole lot to do with it, but instead, the CPU SPEED!

300x10 with ratio at 5:6 to give cpu/12, 250mhz ;)

http://eclipseoc.com/phpwcms_filestorage/1/1_182.png

looks like the cpu.. no, MEMORY CONTROLLER SPEED is the limiting factor, not the ratio itself.


sorry for going a touch off-topic, but this is a concept i expect everyone to fully understand. ratios mean NOTHING, it's all about getting the cpu clock up.

[Sentential]

Why do I run 4 gigs or DDR2 at 400FSB 1:1? because AMD cant...

large denominations of DDR is a mess on AMD
Cant say for sure what RAM will be an issue as memtest and others are not supported for AM2 yet, but its not looking good. As a warning to the community, be very careful what you buy. Micron is most likely going to be the safest choice (ie BT-3 or B6).

[vapb400]

sentential, you keep bringing the bad news for AM2. :(
I hope AMD gets it sorted out.

[nn_step]

sentential, you keep bringing the bad news for AM2. :(
I hope AMD gets it sorted out.
He keeps bringing BAD news for DDR2.. there is a big difference

[bachus_anonym]

Why do I run 4 gigs or DDR2 at 400FSB 1:1? because AMD cant...

The biggest gain for AMD will be the ability to go far beyond 300FSB at 1:1, heel if you can reach 500FSB, my new Micron does that at 3-3-3-4 with all slots filled.Charles, this is not going to be the case on AM2, as it has not been on A64 altogether. AM2 will not have to run 400HTT to run DDR2-800.

With Intel platform, CPU and NB (memory controller) communicate at Front Side Bus speed (e.g. 800MHz). Performance of memory controller increases (latency get reduced) with FSB increase. That's why for the lowest latency possible you need the highest FSB possible.

On A64 (including AM2) memory controller communicates with CPU at speed of CPU (2GHz CPU = 2GHz memory controller). Whether we run 200HTT (200x10=2GHz) or 400HTT (400x5=2GHZ) it does not matter as memory controller will still work @ 2GHz. On AM2 we will be using (what even S939 processors have now) "upclock" memory ratios. So we will run 200x12=2400MHz, and DDR2-800 ratio from BIOS (CPU/6) resulting with 400MHz on RAM. We could run 400x6=2400MHz (provided mobo/cpu can produce 400HTT) to get "1:1" of 400MHz RAM speed. But performance will be the same, just like it is the case now, with S939 and DDR.

[FUGGER]

I see what you mean about MC running at chip speed. I was under the impression bus speed improved MC performance.

[Revv23]

i just hope that with AM2 high density ram will not kill performance like on s939.

im tired of having to have 2x512 because the ODMC cant handle 2x1gb so well. Not only that, i want it in 4 dimms, thats right, i want 4x1gb, fully compatible, no performance hit for running all slots. Then ill be happy. (very very happy lol)

[largon]

Revv23,
Why would you say S939 can't handle high density ram (64x8)? Do current 1GB modules do better on IMC-less platforms? And what do you mean that S939 kills high density dram performance? What I have seen, it's the exact opposite. 128x64 modules at a fixed speed and timings give better B/W than conventional 64x64 modules.

[gundamit]

So due to to a "performance bug" that will be addressed in revision F, TH couldn't show us DDR800 which is the whole point of AMD making the move to DDR2. So instead of a sight performance bump we saw performance degrade slightly. I wonder if selecting memory that was actually rated for DDR800 would have made any difference.

In any case nothing dramatic here. :yawn2: I would have rather read about how the boutique memory makers in the PC enthusiast market are going to approach the new AM2 market. They should be ramping things up now. Lower latencies and higher speeds and price drops (leveraged with volume) all sounds good to me. :D

[Revv23]

Revv23,
Why would you say S939 can't handle high density ram (64x8)? Do current 1GB modules do better on IMC-less platforms? And what do you mean that S939 kills high density dram performance? What I have seen, it's the exact opposite. 128x64 modules at a fixed speed and timings give better B/W than conventional 64x64 modules.


You misunderstand me, i have no problem with bandwidth. What i am saying is that i wish that Rev.f ODMC is more compatible. Certainly with 128x64 B/w is better. but try running 4 DS Dimms, or 4 gb of memory, and then trying to overclock it. Fact is, more memory you have the slower you have to clock it. Personnally id like to take the limit off of the MC and onto the IC.

So due to to a "performance bug" that will be addressed in revision F, TH couldn't show us DDR800 which is the whole point of AMD making the move to DDR2. So instead of a sight performance bump we saw performance degrade slightly. I wonder if selecting memory that was actually rated for DDR800 would have made any difference.

In any case nothing dramatic here. :yawn2: I would have rather read about how the boutique memory makers in the PC enthusiast market are going to approach the new AM2 market. They should be ramping things up now. Lower latencies and higher speeds and price drops (leveraged with volume) all sounds good to me. :D

According to what sentinal was saying, the bug was more then that. The bug made it so that when you set the DDR2-667 divider, the cpu actually ran at DDR2-3200. making the tests that toms ram basically low latency DDR vs. High latency DDR2.

[ozzimark]

According to what sentinal was saying, the bug was more then that. The bug made it so that when you set the DDR2-667 divider, the cpu actually ran at DDR2-3200. making the tests that toms ram basically low latency DDR vs. High latency DDR2.
ah, i see. that's more of a simple math bug, with the clock generator on die picking the wrong cpu/x for the ram speed than an actual performance bug. thanks for the clarification :toast:

[guess2098]

a pic from my friend at CBB

http://coolaler.kj.idv.tw/AM2/AM2_5.JPG

[vapb400]

nice pic.
I take it that is an Epox board?

[nn_step]

Probably..

[perkam]

Results are ho hum, as expected....though tom's kept the cas latency at 4 even though the ram can run 3. Little difference from ddr1, which is both good and bad for amd. The Inq calls it a review...not with an AMD ES chip ;)

The Socket AM2 is a generation change that is becoming increasingly necessary for the sake of supporting memory that will be mainstream in mid-2006 and into 2007. It won't change the performance landscape by much, unless AMD decides to make some fundamental changes to the architecture such as upgrading the FX to 4 MB cache. However, it will also be the basis for the upcoming 65 nm processors that are due at the end of this year.

http://www.tomshardware.com/2006/02/21/a_look_at_amds_socket_am2_platform/index.html

Perkam

[en4cer]

you're right.. so so.... but I think these chips will get better with time.

especially with better ram. :)

[bachus_anonym]

Gotcha, Kam :D http://www.xtremesystems.org/forums/showthread.php?t=90090

[DilTech]

Perkam, that was with one of the chips with the bad memory controller..

The ram, at that high latency, was only truly running at DDR2-3200 speeds.

Basically, the fact that it kept up at ALL with the 939 X2 shows that the chip IS going to be a force to be reckoned with.

It's no surprise though that THG couldn't wait to post numbers for a AMD chip that weren't promising. No surprise at all!

Notice the fact that bandwidth was pretty much identical to the chip running DDR1. Now you know why.

[perkam]

Gotcha, Kam :D http://www.xtremesystems.org/forums/showthread.php?t=90090 I was thinking someone would've gotten this already, but decided to post it anyway :p

--Threads Merged--

Perkam

[iboomalot]

How is 4-4-4-12 avg timings on DDR2 ???

most DDR1 will run 2-2-2-5 @ 400mhz

I was looking at OCZ's site and they only showed the following timings

DDR2 3-2-2 @ 533mhz (OCZ DDR2 PC2-4200 Platinum Enhanced Bandwidth Revision 2 Limited Edition size 1024 mb)
DDR2 4-2-2-8 @ 667mhz (OCZ DDR2 PC2-5400 Platinum Enhanced Bandwidth size 1024 mb)
DDR2 4-5-4-15 @ 800mhz (size 512 mb)
DDR2 5-5-5-15 @ 1000 mhz (size 512 mb)

I wonder how much dropping from 4-4-4 to 4-2-2 would have effected the tests??

I hope AMD has more up its sleeve to stomp the Conroe.

[nn_step]

They do.. it just requires them to time Their Launch Perfectly with Intel...

[Rickster_64]

I am assuming the bus speeds will remain 200mhz HTT for now with AM2??? I know they are supporting PC2-6400 (800mhz DDR2) but thats kind of crappy because unless these new boards are ungodly, we won't be hitting these speeds 1:1. Specially not for the top end DDR2 1066 stuff. It just seems to me that there is hardly any increase in speed running ram faster than your bus.

[idiotec]

I hope AMD has more up its sleeve to stomp the Conroe.
The release chip will no doubt be better than these bench's, but 'stomp' Conroe, I wouldn't count on it :)

[[cTx]Philosophy]

This is exactly why I just upgraded to 170 expert and some nice cronus instead of waiting for am2..

Im holding onto the asrock mobo just in case there are a hit and all..

But I think my newest investment will hold me over until they get these bugs and support and all hammered out, its certainly gonna be a good thing to watch as amd and P$% battle it out on the ddr2 battlefield..

FX with 4meg lvl 2 is gonna PWN for sure im assuming..
I wish they would go from 512 to 1m and 1m to 2m etc
give all proc a boost in preformance, but im sure this is way to costly for them to produce except for the most highly esteemed processors..

[ahmad]

Well look how many revisions it took to bring A64s to what they are now.

Newcastle - Winchester - Venice... then look at the opterons and what they can do (Venus...)

Clawhammer - Sandiego

Just a though...

[largon]

FX with 4meg lvl 2 is gonna PWN for sure im assuming..L2 size on AM2 FX's will be the same 1MB/core. Rumour is that FXs - along with Opterons - will get a 4MB L3.

Hmm. A dual core die with 2x1MB L2 + 4MB L3 is a tight fit on AM2 40x40mm package.
Or is it? :p:


Have they not tested their 65nm machines enough already? :stick:

[nn_step]

L2 size on AM2 FX's will be the same 1MB/core. Rumour is that FXs - along with Opterons - will get a 4MB L3.

Hmm. A dual core die with 2x1MB L2 + 4MB L3 is a tight fit on AM2 40x40mm package.
Or is it? :p:


Have they not tested their 65nm machines enough already? :stick:
You do realize that DDR2 @ 1Ghz with 2-2-2-5 timings would have the exact same latency as L3

[largon]

nn_step,
Ok. If you say so. :p:
May I ask how did you actually estimate the latency for DDR2 1GHz 2-2-2-5?

K8 cache latencies
L1 (3 cycles): 1.1 - 1.5 ns
L2 (8 cycles): 2.9 - 4 ns (in case of VB overflow = 6 - 8.5 ns)

I have to guess the L3 latency. :) Umm... Let's say 15 cycles.
L3 (~15 cycles): 5.4 - 7.5 ns

I woulnd't mind a 4MB with a latency of 5.4 ns. :D

[largon]

Willis,
With VB overflow it's 17 clocks. w/o VB: 8 clocks