AMD’s M2 socket, lowering performance?

[savantu]

That's not looking good at all. Did AMD just give up on M2? Seems like they plan to release a mediocre chipset just to appeal to the loyal [read: mindless] AMD followers who will never touch Intel. Meanwhile, Intel is moving forward with rather effective technology, with Conroe, a processor that can beat AMD's best offering in the performance per mhz market, a place AMD once was the proud owner of.

AM2 is the result of 3 factors :

-DRAM makers pressure
-OEM pressure , DDR is more expensive
-DDR2 better suited for notebooks and server market

Fortunately, Conroe's gains over the FX-60 aren't as amazing as AMD's Athlon series was against the Pentium 4. So maybe AMD still can get it together, if they don't suddenly shift their focus on K10.

Sorry , but that is pure BS.

A64 was better in games ( 5-10%) while Prescott everywhere else , especially in encoding/rendering/multitasking.

X2 was better overall than PD 8xx by 5-10% while Presler 9xx put it on par.

BTW , Conroe was 20+% better not compared to FX60, but to FX62 .Hope you can see the difference...

 

[savantu]

....

Update:
Just an additional bit of info... doesn't look like Intel is going to be having these "wonder" chips available for mass-production until 2007... which is definately long enough for AMD to get some of their AM2 chips out, and revise the process somewhat... then by mid-2007 they'll probably go through a process switch, down to 60nm or whatever is next on the list.

http://www.overclockers.com/tips00932/

10% Intel production is 2x entire A64/X2 production while in Q4 20% of Intel production will be higher than the entire AMD cpu production ( of which 50% are Semprons )

 

[GilmourD]

10% Intel production is 2x entire A64/X2 production while in Q4 20% of Intel production will be higher than the entire AMD cpu production ( of which 50% are Semprons )

His point is that none of that is going to be Conroe, so what's the point?

 

[mICKEL]

i have a feeling 250mhz at 2-2-2 will be pwned by 400mhz at 4-4-4 in bandwidth and latency, check this out:

with the cpu held at 1.8ghz...
200mhz, 2-2-2 = 51.5ns
300mhz, 3-3-3 = 43.8ns

i double checked and made sure i didn't do something odd. interesting eh?

CAS is in clock cycles.

 

[GilmourD]

OCZ Gold VX is 250mhz, 2-2-2, with over 3.3 volts. I dont think it gets better than that. And BTW, Intel's 865/875 hate this UTT. Like I said, some performance DDR2 will show up for AM2.

Everybody seems to question what you say, Mark. No wonder why you're always experimenting CONSTANTLY! (whatever happened with those weird timings in the SPD?)

How many boards can supply 3.3Vdimm? Not a whole lot. And is that in a 2x1GB configuration? If not, 4 DIMMs for 2gigs will cause most AMD systems to drop to 2T command rate, thus worsening latency. 250 @ 2-2-2 will probably be around 300 @ 3-3-3 in latency and overall bandwidth. However, which would be less expensive to implement (thus make more sense)?

 

[(cf)Eclipse]

How many boards can supply 3.3Vdimm? Not a whole lot. And is that in a 2x1GB configuration? If not, 4 DIMMs for 2gigs will cause most AMD systems to drop to 2T command rate, thus worsening latency. 250 @ 2-2-2 will probably be around 300 @ 3-3-3 in latency and overall bandwidth. However, which would be less expensive to implement (thus make more sense)?

i have the ram to test it, and i fully intend to find out at what mhz 2-2-2 becomes the same latency as 300mhz, 3-3-3.

trust me mickel, 400mhz at 4-4-4 will own your utt-bh.

 

[E4g1e]

Sounds like AMD has not yet implemented dual-channel memory support in the first version of the AM2 memory controller. In that initial prototype, all DDR2 memory runs only in single-channel mode.

 

[(cf)Eclipse]

Sounds like AMD has not yet implemented dual-channel memory support in the first version of the AM2 memory controller. In that initial prototype, all DDR2 memory runs only in single-channel mode.

i was thinking that the ratios were not correctly implemented yet, and nothing over 1:1 worked right. the bandwidth in tom's test was right on with a dual channel ddr2-400 setup...

 

[mICKEL]

I saw AMD's latest Roadmap today, and I'm not allowed to disclose information (w00t w00t Info Restrictions) but AMD does have something up their sleeve, and they have taken conroe into mind.

This is the internet. If you can't link to something, it isn't real. Got a link? To anything?

 

[GilmourD]

This is the internet. If you can't link to something, it isn't real. Got a link? To anything?

I believe there's a school of philosphy that believes that if you can't see something, it doesn't exist. Two of my friends have been dating a long time, but before my buddy Keith met Danielle, I was friends with her, and he would yell "YOU DON'T EXIST!" while I was on the phone with her since they hadn't met. LOL

 

[mICKEL]

This is what amd has, http://www.dailytech.com/article.aspx?newsid=418
"We'll work our buns off to beat Intel." - Hecto Ruiz.
In juanuary he was out trying to get money for 65nm, and made that mistake with his words. He did not need to beat Intel, there were(and are.) He knew Conroe was a winner, before this stunt at the IDF. Also, 65nm coming soon? Next year is not "soon." I do hope AMD comes up with a surprise. I like suprises!

I believe there's a school of philosphy that believes that if you can't see something, it doesn't exist. Two of my friends have been dating a long time, but before my buddy Keith met Danielle, I was friends with her, and he would yell "YOU DON'T EXIST!" while I was on the phone with her since they hadn't met. LOL

I would accept a phone call from amd. But a post in hard fourm, with no link?

 

[savantu]

His point is that none of that is going to be Conroe, so what's the point?

The point Sir is that in Q3 10% of Intel cpu production will be Merom based and that will rise to 20% in Q4.

20% in Q4 is over 30 million CPUs....

 

[GilmourD]

The point Sir is that in Q3 10% of Intel cpu production will be Merom based and that will rise to 20% in Q4.

20% in Q4 is over 30 million CPUs....

And it's still Q1. So what's your point? AM2 was rumored to be out in December/January. PS3 was supposed to be out already, too. StarCraft missed its release date by 2 years. Duke Nukem Forever was almost done in the '90s.

If Q3 rolls around and 10% of Intel CPU production is Merom and it's 20% by Q4, then all fine and good. However, my crystal ball is nearsighted. It's Q1 and I never hedge any bets on product that's not supposed to be out for more than half a year. Remember Tejas? That was going to be Intel's HUGE chip. I think it ran off to Mexico.

 

[FreiDOg]

i have a feeling 250mhz at 2-2-2 will be pwned by 400mhz at 4-4-4 in bandwidth and latency, check this out:

with the cpu held at 1.8ghz...
200mhz, 2-2-2 = 51.5ns
300mhz, 3-3-3 = 43.8ns

i double checked and made sure i didn't do something odd. interesting eh?

You're absolutely correct
Accessing the memory core itself takes the same amount of time,
everything else happens 50% faster.
DRAM operations are very deeply pipelined, all that goes into decoding the address, putting the infromation on the bus, scheduling the read et al happens in the same number of clock cycles no matter what the memory bus clock speed is. A faster memory clock means all those operatoins surrounding the read take less time. I'm a bit surprised it was almost 8ns faster, but not surprised it was faster.

 

[Logan321]

The point Sir is that in Q3 10% of Intel cpu production will be Merom based and that will rise to 20% in Q4.

20% in Q4 is over 30 million CPUs....

I have 2 words for you: Tier one.

Just because Intel is making chips that run cooler and faster, doesn't mean they will be available for retail sale. Intel is most likely going to bring their new cpus to the server/workstation market first, because that's where they're losing ground, and reputation.

Intel doesn't give two sh1ts about the enthusiast market, I guarantee it. We're peanuts to their bottom line, and they know it. The problem is, many enthusiasts don't.

 

[savantu]

I have 2 words for you: Tier one.

Just because Intel is making chips that run cooler and faster, doesn't mean they will be available for retail sale. Intel is most likely going to bring their new cpus to the server/workstation market first, because that's where they're losing ground, and reputation.

Intel doesn't give two sh1ts about the enthusiast market, I guarantee it. We're peanuts to their bottom line, and they know it. The problem is, many enthusiasts don't.

Intel's chips were always available on retail days after the product was launched.

 

[dave_graham]

Intel's chips were always available on retail days after the product was launched.

not true at all.

intel has paperlaunched with the BEST of them.

dave

 

[savantu]

not true at all.

intel has paperlaunched with the BEST of them.

dave

Huh?! Which one ?

 

[dave_graham]

Huh?! Which one ?

take your pick outta history. i don't have the time or the energy to list all the things that they've done in that regard. suffice it to say, they HAVE done it and i as an Intel reseller (amongst other things) have suffered for it.

end of story.

Dave

 

[empoy]

let's forget about history. just focus on what's going to happen in July when Conroe comes out in the retail channel and beat AMD's offering in price and performance

 

[savantu]

take your pick outta history. i don't have the time or the energy to list all the things that they've done in that regard. suffice it to say, they HAVE done it and i as an Intel reseller (amongst other things) have suffered for it.

end of story.

Dave

I'm speaking from a consumer POV.

I could buy notebooks with Yonah in late January/early Feb and I'm in Europe , Romania.

 

[mICKEL]

Shush, dont't tell dave that.

Intel may not use processes to output the best product possible, but use processes to have the highest yeilds. Don't under estimate their ability to manufacture chips.

 

[FreiDOg]

Intel, like AMD, has a pretty good track record of actually selling parts they release.
The 1.13ghz Coppermine and first 3.8ghz P4s are the only ones that really come to mind for Intel. Though Coppermine was more a failure of quality control than it was of anything else.
There were some chips, like the P4Fs, there intentionally only made availible to certain parters (IBM and Dell), But since they weren't ever really sold - or intended to be sold - to consumers I'm not sure that's really a paper launch.

It certainly isn't like the GPU industry where seemingly one always 'releases' their next big chip when the other starts selling theirs (even if you haven't manufactured a single one for production).

 

[(cf)Eclipse]

I'm a bit surprised it was almost 8ns faster

you're not the only one..

 

[dave_graham]

I'm speaking from a consumer POV.

I could buy notebooks with Yonah in late January/early Feb and I'm in Europe , Romania.

good for you. however, it takes an OEM to get the parts (and the resellers along with them). if said company (AMD or Intel) can't get them in the channel, then there's the requisite waiting times.

dave

 

[dave_graham]

Shush, dont't tell dave that.

Intel may not use processes to output the best product possible, but use processes to have the highest yeilds. Don't under estimate their ability to manufacture chips.

therein lies a problem. quantity, without quality, is meaningless. Let's take a quick look at why:

1.) Recalls. More costly than PRODUCING chips is recalling them. remember the i820?
2.) Defects. Intel has CONSISTENTLY produced more errata per retail chip than AMD. If you're not careful, this turns into point #1.
3.) Marketing. you get slagged in the marketing department, you psychologically allow your competitor "the edge" as far as product delivery is concerned. The egg on your face phenomenon is especially damaging since it causes relationships to become unbalanced between those who vend, distribute, or manufacture parts based on your core logic. you throw your OEMs off, you're killing their revenue along with yours.

Moral of the story: everything is connected together, sometimes with very thin threads. the slightest hint in delivery problems can have wide-spread affects on valuation, sales, etc.

dave

 

[dave_graham]

Intel, like AMD, has a pretty good track record of actually selling parts they release.
The 1.13ghz Coppermine and first 3.8ghz P4s are the only ones that really come to mind for Intel. Though Coppermine was more a failure of quality control than it was of anything else.
There were some chips, like the P4Fs, there intentionally only made availible to certain parters (IBM and Dell), But since they weren't ever really sold - or intended to be sold - to consumers I'm not sure that's really a paper launch.

It certainly isn't like the GPU industry where seemingly one always 'releases' their next big chip when the other starts selling theirs (even if you haven't manufactured a single one for production).

not doubting that they can deliver (duh!) but it's the appearance of delivery that's a bit of a problem.

for the record, there were delays in P4 production (initially) and have been hiccups in subsequent production starts since then. the 3.8ghz P4 was an idiotic move and any of the P3 speed grades above 1.0ghz were delayed.

dave

 

[mICKEL]

Dave, I mean yields, not defective products but junk from the wafers. The number are secret enough that very few even know. Some rumours say AMD uses two processes, one for yields and another for the "cherrys." Do you remember Winchester, I wish they recalled mine.

AMD said something about SiGe last december, but Intel said alot about that years ago. Defects were to high.

 

[MrGuvernment]

hehe.. I will never switch back to Intel, never.. I dont care how much better they say or it they really are.. Intel ruined their reputation with me.. sorry Intel

Your the one who should be sorry, sorry your not intellegent enough to be a smart consumer and "possibly" buy a better / faster technology for a possible equal / lesser price then a counter part.....


How did intel ruin their reputation? i would love to hear this...

 

[LstBrunnenG]

http://www.hardwareanalysis.com/content/topic/56534/

A bite from Anandtech
"Needless to say my excitement vanished after I ran the first performance numbers and it offered about half the memory bandwidth of an average Socket-939 platform. Remember that the major change with AM2 is the migration to DDR2-667 (and DDR2-800) over DDR-400, so memory bandwidth should go up - not be cut in half."
------------------------------------------------------------

Wow, I am glad that I didn't hold off for this new platform and happily purchased my beloved Opteron 146 + DFI SLI eXpert! Finally I can put my system together! (I finally bought a eVGA 7900GT CO SuperClocked!)

To be honest, I skipped most of the rest of this thread, but I wanted to weigh in on the original topic.

I remember back when the A64 was just starting to pick up steam, and I and all the kiddies in the Intel forum were saying "Wait for Prescott! Wait for Prescott!" I mean, come on! The Northwood-Cs were great chips, and back at the very start AMD didn't have that much of a lead over Intel.

Then leaked reviews started coming in from Chinese sites which put Prescott's performance as slightly less than Northwood's. What you say??? Why would Intel release a chip that was slower than what was already out? Many didn't believe it, and many others declared victory for AMD.

Then Prescott came out. And it was slower. Not only that, but it was much, much hotter too. Meanwhile AMD was just getting started with K8, and slowly worked to eliminate any trace of Intel's performance lead, until we have the situation we have before us today. True, Intel did win back some performance because of the 600-800 MHz that Prescott allowed them to scale, but it wasn't enough.

So now let's consider the current situation. On one side we have a superior new architecture being released by Intel. On the other, we have a simple socket change. Furthermore we have leaked results saying that the new socket will present a performance decrease rather than an increease.

I'm going to conclude that these results are right in line. Most likely, AMD will win back some performance by allowing DDR2 to scale an additional 200 to 400 MHz, or perhaps their cores by 200 to 400 MHz, but it most likely won't be enough to make it competitive with Conroe.

 

[lifesboring]

Somebody in HK have posted AM2 (again an engineering sample) benchmarks compared against 939. There is indeed no significant gain with socket switch.

AM2 Benchmarks (in Chinese, so you probably need babelfish translation)