Sign in with
Sign up | Sign in
Your question

AM2: AMD Reinvents Itself

Last response: in Memory
Share
May 23, 2006 11:50:17 PM

The AMD generation grows up. To introduce you to the new AM2 socket, we bring you a comprehensive analysis of the platform, processors and memory, along with benchmarks and purchase recommendations.

More about : am2 amd reinvents

May 24, 2006 12:49:37 AM

Yay first to a thread!! :D  :D 

Anyways, it looks like AMD has finally delived their AM2 platform, now its the wait for intel. This is going to be a "interesting" summer.
May 24, 2006 1:02:30 AM

Fascinating, not a single mention of Intel's Conroe. I guess that would seriously rain on AMD's AM2 launch parade wouldn't it? Also, the mention of AMD's nonexistent price/performance advantage when compared to Netburst. Let's see how long it takes for the AMD fanboys (Tom's Hardware included.....obviously) to jump the sinking ship when Conroe hits the streets.
Related resources
May 24, 2006 1:07:02 AM

women and children first of course... the men will go down with the ship :D 
May 24, 2006 1:12:33 AM

I'm frankly quite disappointed that AMD hasn't lowered the prices on current CPU's, or even the new CPU's yet. If Intel goes through with the pricing that they have stated, and the performance is even 10-15% better than AMD....crud; Intel will make some serious inroads in AMD's market share.

Sure, AMD may have something fantastic to pull out, Intel may be a flop...but indicators seem to point that Intel looks really good now.

My question is WHY would AMD risk their marketshare? They'd better either be ready with something REALLY great, or ready to slash prices dramatically. Anything else seems to be stupidity.

Can't wait for benches on the Conroe/Core 2 chips, I'm eager to see how they really do. It will influence my next build AND I want to see what AMD has up their sleeves (if anything).
May 24, 2006 2:10:19 AM

Quote:
Fascinating, not a single mention of Intel's Conroe.

Maybe it wasn't obvious, but they are comparing cpu's that are currently available. Core 2 Duo hasn't been released yet. I have no doubt that Tom's will do plenty of benchmarks comparing AM2's cpu's and the Core 2 Duo stuff when it is released.
May 24, 2006 2:32:10 AM

The new DDR2's been reported for months that it doesnt not provide the A64s any boost in performance.....what I dont get is, why are the prices so high??
Intel's next generation chip has certainly been proving itself worthy of the performance crown though its not yet released.

As a 939 CPU owner, I am not looking at an upgrade anytime soon....if I had to, it's definitely not going to be AM2 nor a Conroe!!!

AM2 provides little to no improvement, Conroe will require a new board, CPU and memory but will provide some improvement on gaming , conversion (lets face it, gaming at 50fps, its smooth as already, running it at 100fps wont make a difference; an extra minute is not the end of the world), etc. The Intel setup wont come cheap either and I will not be running SuperPI 24/7!!!!!!!!!

I'd wait for Quad-core CPUs and skip the dual core generation just so it justifies the money spent!!!
May 24, 2006 3:09:41 AM

An Impressively complete article, considering the time between release and this publishing.

As to kitchenshark's question about why AMD would risk their marketshare, the answer is relatively simple: servers. AMD has been showing more and more lately that they're primarily in the server business and their upgrade schedules have been built around that. They are still the best there, and if things go on as they've been predicted to, they'll be the best there for a long while.

Notice that no one in the desktop market really cares about quad cores, but that's the major milestone they have planned next. Who cares about quad cores? servers. 65nm is hardly a milestone from what I've read about what they've got planned. As far as I can tell, its just a die shrink and a few tweaks. It allows higher clocks, sure, but what it really does is a) lower the cost per chip and b) lower TDP. 65nm means its cheaper to make a quad core processor and it runs cooler, which will make the server makers VERY happy.

I'm fairly convinced at this point that AMD is fine with their processor line "as-is" and doesn't plan on making major changes for a few years.
a b } Memory
May 24, 2006 3:17:57 AM

Hopefully AMD will impress us with their upcoming roadmap on technology day.
May 24, 2006 3:22:44 AM

well i dont know about kids but AM2 sucks ass.....yay 4 frames....

like i said all AM2 is is the same chip just with DD2 Support...if conroe benchmarks are true then AMD is about to get its ass handed to it this summer untill they relase there new chips :L
May 24, 2006 3:24:42 AM

I agree with you that this is an amazing detail article on new AMD processors!!

Isnt Quad-core CPUs on the 2007 roadmap?

I too, think AMD is OK for now with that processor line as prices are the more or less then same except for a few odd CPUs. When competition comes with a 65nm processor that can easily out do your high end CPU....you gotta be concerned!!!

I just believe that upgrades should always skip a generation just to see a big enough improvement thats all!!!
May 24, 2006 3:32:40 AM

Quote:
like i said all AM2 is is the same chip just with DD2 Support...

No Shit. AMD didn't claim it to be anything more then that.
Edit: Forgot to add, AM2 isn't a chip, it's a socket.
May 24, 2006 3:49:52 AM

Quote:
The AMD generation grows up. To introduce you to the new AM2 socket, we bring you a comprehensive analysis of the platform, processors and memory, along with benchmarks and purchase recommendations.

Quite absurd.
May 24, 2006 4:53:45 AM

I would recommend anyone looking to upgrade, wait at least 4-6 months after it is released. I say this to allow for a leeway for bug fixes, as well as the price will have dropped a little. You will also be able to know what Intels response to this is, and its capabilities.
May 24, 2006 11:46:56 AM

Quote:
I would recommend anyone looking to upgrade, wait at least 4-6 months after it is released. I say this to allow for a leeway for bug fixes, as well as the price will have dropped a little. You will also be able to know what Intels response to this is, and its capabilities.


Hey?!?

Where have you been? This is AMD's response to Intel’s 775 architecture. Long over due if you ask me. DDR2 has proven its worth for long enough now and AMD has never taken advantage of that until now.

As for Intel’s next step, the Conroe / Core 2. Being based on the Centrino / Yonah (We all know what the yonah did to the FX55 with lower clocks and even low power and heat levels is just frightening) I don’t see how AMD's 65nm Chips competing with it. They might be able to push their clocks to 3.2ghz but I personally am not going to go out and spend $1000 on a CPU.

I say this AM2 move should have happened a year ago already. Intel supporters already have the DDR2 and some of them already have the motherboard for the Conroe step becuase intel's "new" architecture is now "old" and well into the market.
May 24, 2006 1:29:41 PM

The reason the prices are still so high is that as of today they are still the best you can buy. Come July/August prices could change.
May 24, 2006 1:53:13 PM

Quote:

I say this AM2 move should have happened a year ago already. Intel supporters already have the DDR2 and some of them already have the motherboard for the Conroe step becuase intel's "new" architecture is now "old" and well into the market.


Given the issues with DDR2 clocking I think its better they waited this long. If we are talking about high priced RAM chips giving the performance increase NOW then if they had released it a year ago the $ of the RAM would be outrageous
May 24, 2006 2:20:47 PM

Heh. Yes, you're right. The logic and pointers are inescapable now that you've directed my attention that way. And there is definitely more money to be made in high-volume server processor sales. Still...as a consumer it would irk me to be ignored for long in the areas where I need good price/performance.

I still don't think it is wise of AMD to ignore the market segment that raised them into the position they are today. My personal opinion only.
May 24, 2006 2:28:08 PM

It's is a very good article... very comperhansive and a very good in detail about the differences of the DDR and DDR2.

But personally, i don't care about the M@ socket, just bought few motnhs ago the 939 socket system and i am very happy with it. I prefer to see new CPU X2 for the 939 with a considerable lower price. Those X2 thingies are to expensive, although i would like tp upgrade in the future from my Venice3000+ to a X2 939 cpu, if they still exist.
May 24, 2006 3:14:28 PM

What AMD does is no more than changing the integrated memory controller to match DDR2 RAM, so the result is nothing to surprise us.

However it shows some weak points on the current AMD CPU core design:

1. It can't use full bandwidth of Dual-channel DDR2 RAM, even FX62 clocked at 2.8GHz can use only 8 of 12.8 GB/s bandwidth, and AMD seems to have no way to rock CPU clock up above 3GHz. The K8 core had done a good job considering it is just a slightly improved core of K7 plus an integraded memory controller.

2. The performance hampers due to higher latency of DDR2 RAM, which AMD has to fight against.

The short term solution is to make the core with bigger caches(seems to be L3 cache) so as to hinder the higher latency of DDR2 RAM.

In a longer term, of course AMD has to release a new core, which utilise higher memory bandwidth at the same clock speed of current ones.

==============================================
The current DDR2 K8 core may have short life, but the AM2 socket motherboard should have a long lifetime... as long as DDR2 RAM survives.
May 24, 2006 3:37:57 PM

The AM2 chips are on pricewatch now. The prices aren't that bad on a performance basis though, compared to the current crop of intels. Looking at the X2 4000 vs. the D 840 (approximately the same performance in the game test) the X2 is ~$325 and the D 840 is ~$440. The 3800 non-X2 is a solid game performer for ~$290 and it performs as well as the D 950 which is ~$345. And I expect the prices to drop when intel releases its new architecture.

They're going to have trouble selling any of the processors who's clock speed isn't divisible by 400 though, at least in the performance market. And fast DDR2-800 is still very pricey.
May 24, 2006 4:13:59 PM

Quote:
Fascinating, not a single mention of Intel's Conroe. I guess that would seriously rain on AMD's AM2 launch parade wouldn't it? Also, the mention of AMD's nonexistent price/performance advantage when compared to Netburst. Let's see how long it takes for the AMD fanboys (Tom's Hardware included.....obviously) to jump the sinking ship when Conroe hits the streets.


why mention Conroe when its not even out yet?

Your fanboism is getting into your head..

AMD always had price/performance when compare to Netburst...its been constant, even for Pentium D, X2 has better price/performance ratio.
May 24, 2006 4:15:27 PM

Actually, from another site ...all these sites were asked not to mention conroe by AMD as part of the agreement on the coverage.

But, yes all of the hardware sites agree that it is not good to compare a future chip to one that is actually released.

Though hexus just released some preliminary test results on the two conroe's they have:

http://www.hexus.net/content/item.php?item=5692

(In short, it says conroe is all that Inetl claimed)
---
I have not upgraded for 5 years! I planned to last January. But then the new tech over the horizon kept me waiting (something i agree you never should do)

here i am 1 year and 1/2 later..still deciding --and now conroe in July

(dx10/vista in january).

But now i am holding off until at least july...and that is pretty close to xmas....

Close your eyes and just pick!! (i have 8 games, and growing, collecting dust while i stall:) 

---
At the time i picked the cutting edge k7t266 pro-r solution ...and got burned
To this day i have usb/raid problems. Had i waited another month, the problems would have been fixed in the A revision.

Yet, i am fairly sure i will buy whatever first comes out for the conroe.
Its been so long!!
May 24, 2006 4:28:47 PM

Since the AM2 is a down grade when using DDR-2 what is the point of upgrading right now. I guess since WATTs consumed on load is so great but still the price is way to high for a good processor. gone are the good old AMD days of great price great processor. Meet the new AMD starting to look a lot like Intel to me. :roll:
May 24, 2006 4:33:26 PM

Quote:
Since the AM2 is a down grade when using DDR-2 what is the point of upgrading right now. I guess since WATTs consumed on load is so great but still the price is way to high for a good processor. gone are the good old AMD days of great price great processor. Meet the new AMD starting to look a lot like Intel to me. :roll:


Agreed. No point to upgrade for now.

Yes, AMD used to be cheaper while faster than Intel. Now its only faster, not cheaper...If core duo 2 comes out to beat AMD, then AMD loses on both.
May 24, 2006 4:34:49 PM

It seems to me that AM2 is more of a socket update rather than a new range of CPUs to compete with Conroe. In the end of the day, the CPUs are the same, just with an updated IMC, and for the mobos a new chipset to boot. However, I see this as AMD's attempt at buying more time as they prepare their 65nm processors, along with developing a decent platform that should last them for some time. I think its better than nothing, but I'm not impressed with the current lineout of CPUs, since they could have been tweaked more than what AMD has done. The prices will drop eventually, just give it some time, since Conroe will force AMD to drop their prices in order to compete (That's if Conroe lives up to the hype). Overall, it seems AMD is trying to shed the underdog image, but unless they do something about the current lineout of CPUs, they will be keeping the underdog image for a while longer.
May 24, 2006 4:43:42 PM

Quote:
I'm frankly quite disappointed that AMD hasn't lowered the prices on current CPU's, or even the new CPU's yet. If Intel goes through with the pricing that they have stated, and the performance is even 10-15% better than AMD....crud; Intel will make some serious inroads in AMD's market share.

Sure, AMD may have something fantastic to pull out, Intel may be a flop...but indicators seem to point that Intel looks really good now.

My question is WHY would AMD risk their marketshare? They'd better either be ready with something REALLY great, or ready to slash prices dramatically. Anything else seems to be stupidity.

Can't wait for benches on the Conroe/Core 2 chips, I'm eager to see how they really do. It will influence my next build AND I want to see what AMD has up their sleeves (if anything).



AMD is smart. Intel Core 2 has to compete with NetBurst too. Unless you think there aren't anymore.

And the perf advantage over NetBurst is not nonexistent at least until NetBurst is clocked 1GHz higher. Why can't AMD survive for a few months being a little slower. Intel got away with being A LOT SLOWER for 2 years with the same prices.
May 24, 2006 5:02:58 PM

I hear you. The AM2 is the same K8 chips that were on Socket 939 but with a new DDR2 memory interface and it seems that there are still a few bugs to get worked out, especially with the memory multipliers.

AMD did AM2 because of the upcoming quad-core chips needing the bandwidth and the price of DDR2 dropping to DDR levels. I think that they released AM2 now because you generally get better reliability by making only one major change at a time to your products. The K8 architecture is proven and AMD simply swapped out IMCs between Socket 939 and AM2 chips. This allows them to get the bugs out of the IMC and hone it to as fine of performance that their DDR controllers had before shrinking dies to 65nm and then introducing a new microarchitecture in K8. Intel got burned with the Prescotts when they did a die shrink from 130nm -> 90nm, reworked the uarch, and added a new Socket 775 socket with DDR2 support. They now have the Prescott relatively reasonable now that it is shrunk to 65nm and called the Cedar Mill, but it took a little while to work out all the bugs from so many new things.
May 24, 2006 5:16:11 PM

The lower wattage of the new AM2 "A" level X2 chips are the same as my Socket 939 X2: 89W. The new 65W models and 35W models are an improvement.
May 25, 2006 12:03:21 AM

I know that I am going to wait for a while before doing an upgrade or building a new system. My current Athlon64 3200+ is still hanging with the pack and I agree with some of the previous posts that if you are a gamer who cares if you get 100 fps or 50 fps both give you a good gaming experience. And what is all this hype about Conroe do you really think it is going to give you all the power you will ever need? There will continue to be bottle necks in computers no matter how fast your system gets. Hell if Conroe is really what it's all cracked up to be great! Then wait six months and Conroe will be old news. I can tell you the future of computers right now. We will have new cpu's with new socketts that are faster and faster with 10 or more cpu cores, then you will need DDR3, then DDR4 then Hyper drive DDR, the PCI express 64bit and PCI hyperactive, and 50,000 watt power supplies. You see where I am going with this ;-) and the driving force? MONEY!
May 25, 2006 2:16:27 AM

Holy christ 62 fking pages. Please put more than 1 paragraph on a page plz
May 25, 2006 7:53:40 PM

I think AMD missed the proverbial "boat" this time.

The last couple years they have really been putting out an excellent product. AMD has/had won over tomshardware in price and performance.

I am extremely dissappointed in this release. They would have done better to wait to either deliver similar performance to Intel that was lower priced or to deliver a superior product at a slightly lower price.

In the past the price/performance ratio even in the extreme chips was really good and far surpassed Intel.

I will not be buying the new chips anytime soon and now I have to re-think why I left Intel chips to begin with. *sigh*
May 26, 2006 3:23:45 AM

I am planning to build a brand new system in 1 month time.

It will be used for both: latest games and applications

It will have SATAII RAID0 with NCQ enabled on both harddrives
It will be compatible with SLI, but will run only one video card, another card will be added in the future.

I plan to get a cheaper CPU.

Do you think I should go with Nforse 4 AMD 939 and DDR RAM or Nforse 5 AMD AM2 with DDRII RAM ?

Or maybe I go with intel, it's 2x cores CPUs seems to be a lot cheaper ?
May 26, 2006 4:29:54 AM

I think after reading the reviews and hearing the comments maybe I will just go with socket 939 when I upgrade.

Damn, all my excitement just faded when I read the benchmark scores.
May 26, 2006 4:44:55 AM

In one month from today, the best thing you can build is an AM2 on nforce 5. Intel's dual core cpus are cheaper, yes, but they are also quite a bit slower. Clock-for-clock AM2 is faster than 939, even though it is just a few frames, and you get upgradability for later. Just remember to buy your CPU with a frequency that is divisible by 400 MHz to get the full benefit of DDR2-800.

That being said, I don't think AMD did as bad as I thought they would have. Clock-for-clock the performance is slightly better, but if you compare model numbers directly, the difference is interesting. I was looking at the 939 and AM2 X2 3800's and noticed that the AM2 one was clocked 400MHz slower yet only lost ~3 fps in fear at max details with the same amount of cache. That's a 16% frequency reduction and a performance hit of only about 1.5%.
May 26, 2006 5:06:22 AM

So what was the deal with AMD supporting DDR2 up to ONLY DDR2-800? If they want to support DDR2-1066 or faster, will they need to release a new socket?
May 26, 2006 7:22:30 AM

This article on the AM2 platform was excellent -- very detailed.

I don't get where anyone could say that Toms is biased toward AMD since they show in detail how there is not much in the way of performance boost and the cost goes up?

AMD has just been a better company with better chips.

Intel sucks with their generic crap, and is politically bad to boot.

Having said that, what is with the AM2? I read the article and I don't get it -- why switch from 939? What would be the main reason? Anyone?
May 26, 2006 10:36:02 AM

I think, point being, Netburst is over, AMD's price/power/performance advantage is also over. Hope you enjoyed it while it lasted.
Let me guess, you think all the preliminary Conroe benchmarks are faked or exaggerated. That's good. It allows you to remain in residence in fantasy land until the official Conroe release.
May 26, 2006 12:54:19 PM

I know I'm new to this forum but why is it there is always an Intel supporter making belittling comments about AMD supporters.

It's a very odd position to take.
May 26, 2006 1:36:50 PM

Quote:
I think, point being, Netburst is over, AMD's price/power/performance advantage is also over. Hope you enjoyed it while it lasted.
Let me guess, you think all the preliminary Conroe benchmarks are faked or exaggerated. That's good. It allows you to remain in residence in fantasy land until the official Conroe release.



So where is Conroe??????????????????????? I guess we will have to wait and see how it does. Even if Conroe is all its cracked up to be it will still only give you a few more FPS in games than a lot of your budget chips. With memory and video being equal my wife's old sockett A 2500+ Barton chip is only a few FPS slower and loads games a little slower than my A64 3200+ Hmmm let me think 50 fps and 60 fps? Look at the charts and actually think about the score. After any new technology is released I wait at least 6 months or a year before thinking about an upgrade. Yes it leaves you a little behind but in the long run it is cheaper to wait to upgrade with last years stuff and there is not a huge performance hit. I think it is ashame that AMD did add support for DDR2 with the 939 but all the benchmarks I have looked at show that there is very little difference in DDR2 and DDR. Maybe once they get the DDR memory speeds up higher it will make a difference but right now not so much. I am sure DDR3 and DDR4 are right around the corner ;-) Just keep spending your money people! AMD and Intell will love ya forever.
May 26, 2006 1:55:40 PM

First off, this site (tomshardware) seems Intel biased in the first place, secondly, the intel fanboi's seem to have more time on there hands and/or there are more of them here than amd fanboi's. It's almost like Honda fanboi's making fun of domestic car owners, really. Kinda sad that people want to make it an issue.

Anywho, I think people are really missing the boat FAR more than AMD did. Case in point: Is EVERYONE hearing running there 939 boards at 5x200? No. So, the memory bandwidth issues are not really a problem. Run these new chips at normal ENTHUSIAST settings, and the bandwidth is there regardless if your clockspeed is divisible by 400. Use 1066 memory with a lower multiplier, it still works. Jeez people.
May 26, 2006 2:29:20 PM

Quote:
First off, this site (tomshardware) seems Intel biased in the first place, secondly, the intel fanboi's seem to have more time on there hands and/or there are more of them here than amd fanboi's. It's almost like Honda fanboi's making fun of domestic car owners, really. Kinda sad that people want to make it an issue.


I see both kinds of fanboys here. However, I have noticed that you generally see the fanboys for Intel posting junk in the AMD product threads, and AMD fanboys posting crap in Intel product threads. AMD has recently introduced a lot of new stuff (Turion X2 and the AM2 platform) so you will see more Intel fanboys than AMD ones. Wait until Conroe and Merom ship and then you'll see the AMD ones come out more.

And BTW, the classic "fanboy" argument is Ford versus Chevy. That was going on for ~60 years before the first Honda was shipped to the U.S in the early 1970s.

Quote:
Anywho, I think people are really missing the boat FAR more than AMD did. Case in point: Is EVERYONE hearing running there 939 boards at 5x200? No. So, the memory bandwidth issues are not really a problem. Run these new chips at normal ENTHUSIAST settings, and the bandwidth is there regardless if your clockspeed is divisible by 400. Use 1066 memory with a lower multiplier, it still works. Jeez people.


I bet a bunch of people here really don't do much with their systems to OC them. A lot of the HTPC and SFF guys undervolt and underclock their systems for reduced heat or put mobile CPUs in them that have low FSBs (Pentium M) or single-channel memory (Turion.) I don't understand the "5x200" but I assume you meant to say that people run their memory clock over the stock DDR-400 200 MHz. When I see "5x200," I think that the 5x is a CPU multiplier rather than a memory divider as I've never seen memory close to DDR-1000.
Athlon 64s and X2s with Cool 'n Quiet *do* have a 5x CPU multiplier, but that's the bottom 1GHz idling one.

And as far as the sub-optimal DDR2 memory frequencies with chips whose frequencies do not cleanly divide by 400 MHz, I think that AMD will end up putting half-multipliers in the next stepping of the AM2 chips so that the 1.8, 2.2, and 2.6 GHz chips will be able to run DDR2-800 at DDR2-800 speeds. I bet they'd also be able to run DDR2-1066 at that point too. AMD did it in the past as the E stepping X2s (maybe A64s as well) can use DDR-433, -466, and -500 without overclocking and the D stepping and before could not.
May 27, 2006 2:25:19 AM

Hi All,

talk about long review.
Isn't the point which everyone seems to have missed so far is that you can now support up to 8GB of memory? Finally a reason to change some of your software to 64-bit, also with Vista fast approaching unfortunately without a new kernel / HAL but I hear likes 2GB+ of memory.

I was thinking intel 775 a while back but thought I'd wait for AMD's DDR2 platform because memory would be cheaper by then. I'd like to see a motherboard test with 8GB loaded to see if it works. Also the network / SATA other useability changes look really useful and probably translate into more meaningful performance compared to what Intel is offering.

Processor speed really does not make that much difference. Buy an X-Box 360 / O.C. the 4000. I'm still doing any normal task (no games) dvdshrink, p.shop with a bx motherboard and an evergreen celeron 1.3 and it's just as quick as any of the 4 or 5 machines I use at work for 99.999% of the things I do. Although my 3.2ghz work machine does seem to be doing o.k. on the BBC Climate Change Experiment.
I find re-imaging XP fixes any performance problems with my home machine and installing only the 9 or 10 apps that actually get used works a treat (if you use more than that think about alternatives)

Now where do I get cheap 2GB 800mhz overclockable (ballistix/ocz/teamextreme) memory from in Australia.
May 27, 2006 3:03:06 AM

Nice first post. I think it's the most appropriate I've seen. Good job, welcome to tomshardware and good luck.
May 27, 2006 3:15:38 AM

I am a Chevy guy myself. I realize not everyone here is OC'ing their CPU, but I would expect people to have some idea of what they are talking about. Unfortunately, it seems most people just want to read someone else's review and use it as fact. No one seems to care about real world performance, either, just synthetic benchmarks. It's great that AMD added the new memory multipliers to revision "e" chips (I have one myself). But if they really were thinking, they would have realized that the multipliers that would allow more tinkering between 166 & 200 are FAR more important than those between 200 & 250. When I was talking 5x200, I was talking about the Hypertransport multiplier that can be used to change the memory bus. Just as tweaking a CPU's frequency consists of the multiplier and bus, so does the memory, so to run DDR500 you can use the divider for 5X200 (stock only on revision E), or you can change multipliers and go 4X250 (doable on any K8 with proper chipset support). When jumping to 939 from 754, the dual channel memory controller gained 100X theoretical bandwidth for the cpu to memory. DDR2 at double the speeds sure does bring twice that bandwidth, but so does a quad channel DDR1 system. There are usually many ways to get the same results, but the fact of the matter is the K8 is not starved for bandwidth anyway. I wish I could see how great Conroe would do with EVEN more bandwidth. And I can't wait for DDR3. All this competition is great. Now if only Ford could get their act together and make an engine with dual overhead cams that produced as much power as the old school tech - pushrod chevy does (I'm speaking strictly no power added/ NA engine). Similiar analogy there are AMD/DDR and Intel/DDR2, don't you think?
May 27, 2006 4:21:19 AM

Aha. Okay, the HT bus. One thing that you are forgetting is that the HT bus doesn't do squat with affecting overclocking. There is a base clock that is usually set at 200 MHz and both the HTT bus and the CPU core multiply it by their own independent multipliers to get the actual HTT bus and CPU core speed. When you raise the base clock, both the CPU core and HTT bus get overclocked too- you mentioned that.

A memory divider is something different- it is the ratio of base clock speed to memory frequency. DDR 400 has a frequency of 200 MHz and a stock K8 has a base clock of 200 MHz, so the memory multiplier is 1:1. The E stepping K8s simply allow the memory divider to be changed to 200/216.7 (DDR 433), 200/233 (DDR 466) and 200/250 (DDR 500) so that the base clock can still be 200 MHz. Previous K8s could not lower the base clock/memory ratio below one, meaning that running anything over DDR 400 at its native speed would overclock the CPU.

Memory modules since the PC66 era (K5/Pentium) send data in 8-byte (64 bit) chunks. Single data-rate memory (PC66, PC100, PC133) send 8 bytes/clock, so a single PC100 memory module will transmit 800 MB/sec (8 bytes/clock and 100 MHz.) DDR memory sends two 8-byte chunks per clock, so a single DDR 400 module is capable of 3200 MB/sec in data transmission. Single-channel DDR memory controllers can only access one memory module per half-clock-cycle, so you get two pulses of 8 bits per clock. Dual channel memory controllers can access two modules at the same time, transmitting two pulses of 128 bits (16 bytes) per clock cycle. So dual channel = doubling the bandwidth.

DDR2 is a little different animal as each module actually interleaves DDR memory chips to get a doubling of the I/O rate (and latency)...but I digress. Single and dual-channel setups with DDR2 work the same as with DDR SDRAM, with a dual channel = 2x the bandwidth of a single-channel setup. The bit width of a single DDR2 module is the same 64 bits as a DDR or PCxx SDRAM module.

It would be possible to use a quad-channel setup where 256 bits or 32 bytes was transferred per memory access, and that would give 4x the bandwidth of a single-channel setup and double the bandwidth of a dual-channel system. But as we have seen in some pair-of-dual-channel (two pairs of RAM modules) setups, this really taxes the memory controller and generally forces a reduction in the speed that the memory runs at. If you had very good, matched RAM, it could be done. But you would eat up all four slots on your board from the getgo, requiring a purchase of all new RAM of a larger size to upgrade the amount of memory in the system. This would also force a new CPU socket for AMD as now you have even more memory lines leading to the CPU core- if it's DDR2, you have 240 pins per channel. Quad channel would mean 960 pins, and that would be really ugly. as you'd have to have something like 1500 pins on the socket.

A better option would be for 128-bit-wide RAM chips to be developed so you can double the bandwidth of the current arrangements with half the number of modules. I bet that will happen in a little while, maybe even in DDR3. It happened before when the old 386/486 32-bit-wide EDO RAM was being run in dual channel in early Pentium systems and the manufacturers would rather be free to use odd numbers of modules or mismatch sizes of a pair of modules, so 64-bit-wide memory modules were made. Frankly, I thought that DDR2 should have been 128 bits wide as 128-bit-wide dual-channel memory controllers were already common and running dual-channel is always more of a hassle than running single-channel if the bandwidth is the same.
May 27, 2006 4:29:07 AM

I think you are getting most of what I am saying and I didn't need your explanation, but it was a good one, nonetheless. My point about the HTT and Ram and CPU multipliers is that you can use a lower multiplier on the cpu along with different HTT settings to use DDR500 on any K8 cpu without overclocking the CPU itself. I guess I didn't make that clear enough. I wasn't saying that a quad channel memory controller was a good idea at this point, either, just that it is another method to double the bandwidth. My other reason of bringing it up, however, is that I am curious how bandwidth will limit quad-core CPU's when they come out. Also, you never chose a side, Ford or Chevy :) 
May 27, 2006 5:20:24 AM

I choose Ford as their engines have a much better specific output and usually more *total* output than Chevy's engines. This is very apparent in the smaller engines, where GM's 3.1 L V6 makes a whopping 170 hp and Ford's 2.5L twincam V6 makes 175. Ford has its omnipresent 3.0L twincam that usually makes 220 hp. It takes GM 3.9 L to beat that figure (the 3.9L engine makes 240hp.) Ford has a 3.5L version of that 3-liter motor that makes 265 hp and predicts that it will eventually top 300 hp.

Even in the V8s, Ford still does better. GM's 5.3L Gen IV V8 makes 295 hp and 330 ft-lb torque whereas Ford's 5.4L Triton 18-valve SOHC motor makes 300 hp and 365 ft-lb torque. Ford also has three other versions of its modular naturally-aspirated V8s of note- the 292 hp 18-valve SOHC 4.6L, which beats the Chevy 4.8L (275 hp) the old 320 hp twincam 4.6 L Mustang Cobra engine that makes its random reappearance in vehicles such as the Lincoln Aviator and Mercury Marauder, and the 32-valve 5.4L Navigator engine with 300 hp.

GM's biggest and baddest truck motor is the massive 8.1L big block with 345 hp. Ford's 18-valve SOHC V10 makes 365 hp from "only" 6.8L.

Now if you let me use some of Ford's excellent blown V8s, Chevy doesn't stand a chance...
May 27, 2006 7:19:40 AM

Nice metaphor.
!