Monday, July 03, 2006

Intel brags about Woodcrest's 83MB/s memory system

According to Intel's featured quote, Woodcrest has an amazing 667 megabit per second (or 80MB/s) memory system which once again raises the bar. Seven days after the Woodcrest launch, finally, someone in NJ ordered two chips - ETA August 2006.

In other news, Woodcrest's unborn brother, the Conroe, was touted as the SuperPi computer in a PC by some Intel investors.

67 Comments:

Blogger netrama said...

Ha ha !
The Woodcrest is more and more becoming a big Joke now !!

3:36 PM, July 03, 2006  
Blogger Mad Mod Mike said...

Dude, I wish my Dual-Core 3.0GHz Opteron 165 would give me 667 megabit per second! All I can squeeze out is 7.6GB/s, really disappointing for 2 channels of RAM :(.

lol this is too great, further shows the imcompetency of Intel.

3:39 PM, July 03, 2006  
Anonymous Anonymous said...

Yea, because we all know the marketing pawn that writes those up really know what he's talking about.

Way to cite an obvious typo to defend your position that AMD's chips are better. You know as well as I that it was supposed to be 667Mhz.

4:10 PM, July 03, 2006  
Anonymous Anonymous said...

why is it that intel fanboys can't take a joke when they're the ones talking smack most of the time. i wonder how much faith intel employees have in their own company. i also wonder how much intel employees know about their company and the product besides the engineers. i have to say, being that i attended all of amd's tech tours, their employees seem enthusiastic and passionate about their company. they also know their company's products.

5:36 PM, July 03, 2006  
Anonymous Anonymous said...

I have got to ask, how much time do you invest in finding these things?

As much effort you put into making Intel look bad you could have solved world hunger and the aids epidemic.

6:25 PM, July 03, 2006  
Blogger nyx said...

Keep in mind that the article is from INTEL. If INTEL is capable of such oversights in their spec listings, how can we believe they set any standard of excellence towards the production of their CPUs? This type of carelessness and stupidity might be acceptible from a third world circus show, but I expect more from corporations with such influence on technological direction. INTEL is supposed to be a corporate contender on the processor front, but they place about as much emphasis on creativity, inventiveness, quality, and efficiency as one would expect from a room full of demented irate monkeys with sledge hammers.

7:57 PM, July 03, 2006  
Anonymous Anonymous said...

Wait until you see Core Duo DX2. It will have 166MB/sec memory and make AMD chip-men cry like little boys.

8:10 PM, July 03, 2006  
Anonymous Graham said...

Way to cite an obvious typo to defend your position that AMD's chips are better. You know as well as I that it was supposed to be 667Mhz.

That is because Shari-fraud and Fat Mod Mike are out of ammo and just waiting for their precious AMD to get fragged.

8:57 PM, July 03, 2006  
Anonymous Anonymous said...

This typo by an _Elpida_ employee most certainly spells the end of Intel! Good God, the fact that the person didn't qualify further the metric (per chip? Per lane?) or possibly made a mistake and meant to say 667 MHz surely means the end of the Evil Empire!

All hail your brilliant detective work in finding this meaningless piece of fluff!

10:06 PM, July 03, 2006  
Blogger Entrophos said...

That is because Shari-fraud and Fat Mod Mike are out of ammo and just waiting for their precious AMD to get fragged.

I think I hear your mommy calling. Kids and their potty mouths nowadays. Tsk tsk.

10:29 PM, July 03, 2006  
Anonymous Richard said...

Let's all take him serious shall we.

No really let's; let's continue bagging him, go on, it makes you feel better doesn't it.

Sure does first thing I wake up in the morning.

Anyway, Sharikou, are you able to get an AMD vs Intel shootout or not? I'm guessing you won't be, but it's worth a shot.

12:35 AM, July 04, 2006  
Blogger Sharikou, Ph. D said...

Anyway, Sharikou, are you able to get an AMD vs Intel shootout or not?

I will definitely do it. But it may take a while. Woodcrests are hard to find these days.

12:42 AM, July 04, 2006  
Anonymous Richard said...

Will it be before or after the release? It's only 20 days away.

12:55 AM, July 04, 2006  
Anonymous BigBadWolf said...

"As much effort you put into making Intel look bad you could have solved world hunger and the aids epidemic."


In respose to above,
"Intel epidemic" is much worse than that.
Keep up the good work Shari.

1:56 AM, July 04, 2006  
Anonymous BigBadWolf said...

"I will definitely do it. But it may take a while. Woodcrests are hard to find these days."


I have already despatched a search party to find woodcrest.

2:03 AM, July 04, 2006  
Blogger symbiansn said...

A years ago, Intel's then highly touted SSE2 was mispelled somehow on their own page, and there was some other typo too, but were corrected short after I sent'em an email pointing the errors. There was a nice reply from Intel thanking me too.

You know, a marketing company company cares most about posting numbers than words, or than doing smart interconnect technology, or IMC, or else, by the way :)

3:57 AM, July 04, 2006  
Anonymous george said...

all i could get out of my athlon 64 x2 is 7.6GB/s. i think intel still does not have anything like hypertransport. and still has the memory bandwith dunce cap.

8:04 AM, July 04, 2006  
Anonymous Anonymous said...

First of all, the comment is about FB-DIMM, not Woodcrest. Secondly, it's about the transfer rate, not bandwidth. Any Ph.D. worth his salt (and sometimes even high-school students) knows that bandwidth=rate*width. So, for the math-impaired, FB-DIMM channel bandwidth = 667MHz*8B*1.5=8GB/s. And that's in the quote as well but, of course, Sharikou's highlighter ran out.

9:25 AM, July 04, 2006  
Anonymous Anonymous said...

"why is it that intel fanboys can't take a joke"

I got to admit. This is the best megabit/s joke I've ever read. I can't wait to tell this to my girlfriend. Any day now.

9:37 AM, July 04, 2006  
Anonymous Anonymous said...

"Intel epidemic" is much worse than that.

Another smart person backing up the doctor, good to see what kind of intelligence is on your side...lol

9:41 AM, July 04, 2006  
Anonymous george said...

BULL AMD's bus is dedicated while intels bus is shared single duplex crap plus 2 hops from northbrige to memory, northbrige to cpu. that is the secret, how could you get that much bandwith with a fsb arch?! intel is getting creamed by amd! and let's say you got a 4x4 system or a opteron the memory bandwith doubles with each procesor added. that means that amd's procesors scale better. an amd system has 3 busses 1 for procesor, 1 for io and another one for memory.

10:16 AM, July 04, 2006  
Anonymous george said...

REMEMBER IT IS PRACTICLE BANDWITH NOT IDEAL BANDWITH THAT MATERS AND AMD IS THE KING OF PRACTICLE BANDWITH!!!!!!!!!!!!!!!!!!!!!!!!!!!

10:21 AM, July 04, 2006  
Anonymous Anonymous said...

I met two kinds of people who try to sound smart. One kind is smart.. The other points to other people shortcomings to look good..

The most recent articles on this comic journal has nothing to do with its objective, "Pervasive 64-bit computing". It is purely Intel bashing.. I wish it has more "sceintific" analysis about archetictures, road maps, .. The key word is "sceintific"

THis Journal looks more like a fantacy journal for immature little people.. So I suggest you rename your journal.. May I suggest "My daddy never played with me", or "I am a bitter because no buddy hugs me"

10:58 AM, July 04, 2006  
Anonymous Anonymous said...

Anonymous said
"First of all, the comment is about FB-DIMM, not Woodcrest. Secondly, it's about the transfer rate, not bandwidth. Any Ph.D. worth his salt (and sometimes even high-school students) knows that bandwidth=rate*width. So, for the math-impaired, FB-DIMM channel bandwidth = 667MHz*8B*1.5=8GB/s. And that's in the quote as well but, of course, Sharikou's highlighter ran out."

Isn't comical how the wanna be Duc made fool of himself and his fanboys... LOL

May be somebody should tell the inquirer!

11:06 AM, July 04, 2006  
Anonymous Anonymous said...

DR. SH.,

I think you should retract your story of describing Intel to Nazi Germany.. The article below states no evidence of your claims.. I think apology is due if you are a big man!


http://theinquirer.net/default.aspx?article=32802

11:15 AM, July 04, 2006  
Anonymous Anonymous said...

Sharikou should definitely do the shootout.

It will have about as much credibility as a USA vs. Iran missile contest judged by Osama Bin Laden.

11:39 AM, July 04, 2006  
Blogger Mad Mod Mike said...

"So, for the math-impaired, FB-DIMM channel bandwidth = 667MHz*8B*1.5=8GB/s. And that's in the quote as well but, of course, Sharikou's highlighter ran out."

For the looking impaired, using DDR2-667 Quad-Channel on a Woodcrest platform only gives 5.5GB/s Memory BANDWIDTH...yes, it is very sad and pathetic.

11:44 AM, July 04, 2006  
Anonymous Anonymous said...

"First of all, the comment is about FB-DIMM, not Woodcrest. Secondly, it's about the transfer rate, not bandwidth. Any Ph.D. worth his salt (and sometimes even high-school students) knows that bandwidth=rate*width. So, for the math-impaired, FB-DIMM channel bandwidth = 667MHz*8B*1.5=8GB/s. And that's in the quote as well but, of course, Sharikou's highlighter ran out."

So basically, you are telling us that the description in Intel's web page is accruate and true, right ?

mmm.. I believe you then.

;-)

11:45 AM, July 04, 2006  
Anonymous Graham said...

I will definitely do it. But it may take a while. Woodcrests are hard to find these days.

It is no harder that ordering from Dell.com. They have Xeon 5100 systems avaialable for order now. Have you ordered one? I didn't think so. Pardon me if I don't hold my breath waiting for your results.

11:49 AM, July 04, 2006  
Anonymous Anonymous said...

"667MHz*8B*1.5=8GB/s "
theoretical bandwidth is not actual its more like 5Gb/s for one core then down to 2.5 Gb/s when doing real work.

such is the FSB madness

5:38 PM, July 04, 2006  
Anonymous Anonymous said...

Dr... when will you learn to be a little more objective.

For now INTEL will have the premier dual-core desktop, laptop and server products for 6-9 months. There really isn't anything to debate. The wind is just blowing the other way. The past two years its been AMD's day. Now it'll be INTEL's.

They will make billions with their new Merom. What will be ironic is that they will make less proft and Gross Margin with COnroe/Woodcrest/Merom even though they are better across the board then they did with Penntim IV/Netburst against a far superior Opetron.

At 45nm node with their next generation Nehalem.. they will really kick AMD back to the loss leader again.

The joke will be on those with it too Piled High and Deep to see the truth

6:31 PM, July 04, 2006  
Anonymous Anonymous said...

dude, what's your middle name? AMD? lol..i think you should change your name from Sharikou, PhD to Sharikou, AmD...makes me wonder where you got your PhD from...AMD campus? Also, your entire blog site can be summarizes to 3 letters: AMD..why spend all the time? Every body knows now that AMD's current FX62 got its butt handed over to the Woodcrests, both in Performance/Watt, and overclockability. Reverse HT is cool, if they can actually show it's working right now in AM2 systems that you can buy today.

9:20 PM, July 04, 2006  
Anonymous Anonymous said...

EETIMES article headline “ AMD, IBM climb in supercomputer ranks; Intel falls”
It seems scientists know what possessor works.

http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=189800109

1:20 PM, July 05, 2006  
Blogger DBA said...

Are you guys waiting for Intel's new server chip? think again.

Every one is surprised Intel can pull out a new CPU so fact. On the other hand, it is not a surprise to see Intel makes a fool of itself.

http://www.theinquirer.net/default.aspx?article=32842

US government unit throws Intel out over RAID problems. Maybe we should only consider talk about the Woodcrest or its benchmark 6 months later.

10:31 AM, July 06, 2006  
Anonymous Anonymous said...

Ouch, another fork in the road for Woodcrest, what a pity. ;^) Where's that old commercial for RAID bug spray when you need it...

http://www.theinquirer.net/default.aspx?article=32842

11:40 AM, July 06, 2006  
Anonymous Anonymous said...

"dude, what's your middle name? AMD? lol..i think you should change your name from Sharikou, PhD to Sharikou, AmD...makes me wonder where you got your PhD from...AMD campus? Also, your entire blog site can be summarizes to 3 letters: AMD..why spend all the time? Every body knows now that AMD's current FX62 got its butt handed over to the Woodcrests, both in Performance/Watt, and overclockability. Reverse HT is cool, if they can actually show it's working right now in AM2 systems that you can buy today."

US government unit throws Intel out over RAID problems!

http://www.theinquirer.net/default.aspx?article=32842

It is not sure if the Woodcrest is the fastest. But one thing for sure is, Woodcrest is a shitty CPU. And it is offical (from US government).

11:59 AM, July 06, 2006  
Anonymous Anonymous said...

"It seems scientists know what possessor works."

Granted Intel fell, but they still have 301 versus AMD's 81, thats still quite a gap.

I would be curious if some of that is related to capacity, AMD may not be able to produce enogh chips.

Don't those super computers have tens of thousands of chips?

12:52 PM, July 06, 2006  
Anonymous Anonymous said...

In other news looks like Intel may have had other plans for the money they got from Marvel...

EEtimes

12:56 PM, July 06, 2006  
Anonymous Mr. PoPo said...

Explain this :

http://www.dailytech.com/article.aspx?newsid=3170

Conroe is top dog until Q4 2k7
K8L appear in Q4 2k7
45nm Conroe is top dog till Q3 2k8
Nehalem is top dog from Q3 2k8

Umm...where is AMD ?

1:33 PM, July 06, 2006  
Anonymous Anonymous said...

not only scientists, but our own government of all places reject intel's product. i know this is not the most reliable source of news, but still a good read.

http://www.theinquirer.net/default.aspx?article=32842

1:51 PM, July 06, 2006  
Blogger Ron said...

http://www.theinquirer.net/default.aspx?article=32855

Dig this

2:21 PM, July 06, 2006  
Anonymous Anonymous said...

mad-mod-mike said:
"For the looking impaired, using DDR2-667 Quad-Channel on a Woodcrest platform only gives 5.5GB/s Memory BANDWIDTH...yes, it is very sad and pathetic.

According to what? Your blog? You didn't even get the bus rate right. It's 1333, not 1067. Also, you don't grasp the fact that most I/O traffic doesn't go thru the bus; the snoop filter blocks that.

Here's your homework. Read this.
Please come back and report the findings on peak bus bandwidth (slide 49), what a snoop filter is (slides 31,32), and then show us how you arrived at 5.5 GB/s. While you're at it, look at slide 33 comparing real loaded latencies of X3 versus Opteron.

And talking about pathetic... Did you realize that the FB-DIMM-667 bus runs at 4GHz? How does Hypertransport 1GHz compare in your opinion? I know, greatest thing since sliced bread.

6:44 PM, July 06, 2006  
Anonymous BigBadWolf said...

Half of Intel employees on PR campaign , the rest half Wankers.

No wonder their processors suck a@@.

8:02 PM, July 06, 2006  
Anonymous BigBadWolf said...

"dude, what's your middle name? AMD? lol..i think you should change your name from Sharikou, PhD to Sharikou, AmD."



And what is your middle name?

Netburst, Northbridge or the upcoming Intel abrotion Itanic .

8:20 PM, July 06, 2006  
Anonymous BigBadWolf said...

http://www.theinquirer.net/default.aspx?article=32842

HEHE... So much for woodys crappy azz.
Eat the dust Intel fanboys.

8:46 PM, July 06, 2006  
Anonymous Anonymous said...

There will be no debate when Woodcrest and Conroe CPUs are widely available with superior performance and competitive pricing. There will be a debate as long as these CPUs are not widely available, or real-world performance (not just SuperPI) is not as good as has been claimed for same-cost processors.

9:00 PM, July 06, 2006  
Anonymous Anonymous said...

AMD's Q2 revenue will be 9% lower than expected: http://www.theregister.co.uk/2006/07/07/amd_drops_q2/

10:52 PM, July 06, 2006  
Anonymous Anonymous said...

This should be the proper link to the "AMD, IBM climb in supercomputer ranks; Intel falls" article.

http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=189800109

7:04 AM, July 07, 2006  
Anonymous muziqaz said...

hm conroe may be faster brutally than A64, but woodcrest won't be considered as real server chip, until Intel comes up with something of the size of HT.
We all can take celeron and put some server OS, and say we have server at home.Same with woodcrest.Major compnies who need scalability will choose AMD offerings, because it is real server chip with good architecture.
Amd for those who hate sharikou's opinion: why bother come here?Is it hard just skip this site? Go to Intel.com and read some PR stuff how woodcrest/conroe/merom will change your life.

7:56 AM, July 07, 2006  
Blogger TheKhalif said...

The most recent articles on this comic journal has nothing to do with its objective, "Pervasive 64-bit computing". It is purely Intel bashing.. I wish it has more "sceintific" analysis about archetictures, road maps, .. The key word is "sceintific"


I can see where you would think that but Pervasive means to cover fully. The only PERVASIVE 64bit architecture is AMD64. it will soon be required for use with MS server products.

12:38 PM, July 07, 2006  
Anonymous Anonymous said...

"For now INTEL will have the premier dual-core desktop, laptop and server products for 6-9 months."

Laptops, yes. Desktop, maybe. Server... definitely not. I think you're forgetting that Intel can't scale their processors past 2P and once you start running multiple apps that tax the L2 on woodcrest it really doesn't perform that much better than a competetive AMD Opteron.

1:04 PM, July 07, 2006  
Blogger Mad Mod Mike said...

'And talking about pathetic... Did you realize that the FB-DIMM-667 bus runs at 4GHz? How does Hypertransport 1GHz compare in your opinion? I know, greatest thing since sliced bread."

I'm sorry you are a moron and don't know wtf you're talking about.

I didn't mention Intel's FSB in that article you linked to idiot, and since you can't understand AMD HyperTransport, and you seem to think DDR2-667 some how runs at 4Ghz (oh lord you are stupid) I won't bother to respond to your idiotic comments.

9:11 AM, July 08, 2006  
Anonymous Anonymous said...

mad-mod-mike, of course you cannot answer being the ignorant fool you are.

Sorry I asked you to read that presentation. It's advanced material and it might confuse your one-track mind. I think we should start with learning how to use Google. It's obvious you have no idea what FB-DIMM is.

Here:

"FB-DIMMs have been designed to provide significant performance gains over current Registered DIMM solutions, and they will offer unprecedented speeds up to 4.8 Gigabits per second (Gbps) data rate, equivalent to 6.4 Gigabytes per second (GB/s) data bandwidth."

9:50 AM, July 08, 2006  
Anonymous Anonymous said...

Here's something more of the real story about today's FB-DIMM world:

"FBDIMMS will most likely be expensive at first, has a little more latency, but better throughput. Sounds like RAMBUS a few years ago."

George, if you want to take this offline ping me.

Basically, the FBDIMM performance is a matter of physics and the economics is a straight inequality.

The latency is dictated by the spec, and it's not "a little more." It's the buffer delay for each DIMM in the chain (twice), plus the controller buffer delay. If you don't have at least three FBDIMMs, there's no point in using them -- plain DIMMs are cheaper and faster. The added latency is not trivial. Unlike RDRAM, it can't ever get better because FBDIMMs use the same DRAM as plain DIMMs.

As for the cost, the same inequality applies. FBDIMMs require a minimum two extra PCB layers, many more vias, an expensive controller chip, and additional cooling in addition to the same bill of materials as a plain DIMM. The plain DIMM will always be cheaper.

What's more, economies of scale will always favor the plain DIMM because there are too many applications (and more each day) where the plain DIMM will do the job. That pushes the FBDIMM out to the low-volume end of the curve, which means that economies of scale make it even more expensive -- the vicious cycle.

Buffering a DIMM is a good thing in a lot of ways, but direct-attach RAM is always going to be cheaper and faster for the low end of the pyramid. Don't ask Intel for the roadmap on this one, ask the memory vendors. Off the record, Mian will tell you straight, for instance.

12:16 PM, July 08, 2006  
Blogger Mad Mod Mike said...

" It's obvious you have no idea what FB-DIMM is."

I feel sorry that sharikou has to let morons like you post. Please show me where DDR2-667 runs at 4Ghz and I will bow to your greatness oh moron.

You have no idea what you're linking too, stop being a moron.

1:33 PM, July 08, 2006  
Blogger Mad Mod Mike said...

"mad-mod-mike, of course you cannot answer being the ignorant fool you are."

You don't know anything about an AMD64 processor, do you? There is absolutely NO link between HyperTransport speed and RAM speed because all AMD64 CPU's have an INTEGRATED MEMORY CONTROLLER which BYPASSES THE SYSTEM BUS (HT) so RAM speed doesn't matter.

You could have DDR2-1000 Quad-Channel on 1 CPU w/ 200MHz HyperTransport and it WOULDN'T MATTER. How many people are that stupid that try to post against me but make themselves stupid in the process because they DON'T READ or KNOW WTF THEY'RE TALKING ABOUT!

This is a message to all future retarded Intel fanboys who don't know wtf they're talking about: DON'T POST!

1:42 PM, July 08, 2006  
Anonymous Anonymous said...

Here you go. Table 2, link transfer rate, since you're kinda slow and might not find it.

BTW, both HT and FB-DIMM are based on high-speed serial interfaces. Sorry this isn't described in a two-letter acronym suitable for your understanding.

2:49 PM, July 08, 2006  
Anonymous Anonymous said...

Mad mod mike, clam down please.

If those Intel fanboys have those technical knowledge, they would be intel fanboys in the first place.

I feel sorry for sharikou to let these kind of postings on his blog.

4:16 PM, July 08, 2006  
Blogger Mad Mod Mike said...

"BTW, both HT and FB-DIMM are based on high-speed serial interfaces. Sorry this isn't described in a two-letter acronym suitable for your understanding."

Let me clear a few things up for your Ignorant Intel fanboy.

1) Link Transfer Rate != Bandwidth OR Speed

2) HT & FB-DIMM's have no correlation together

3) GT/s != GHz. It is referring to Gbps, not GHz, so please know wtf you're talking about when you post. Did I not just say that?

Jeez, Intel fanboys don't read anything...

12:07 AM, July 09, 2006  
Anonymous Anonymous said...

Good try, MMM. GT/s do, more or less, mean Ghz. GT/s are certainly not gigabits/sec.

It's just like how the FSB is quad-pumped, and a 266Mhz FSB is 1066Mhz "effective." What that chart is saying is that it is effectively 4Ghz.

Technically, you are right in the fact that GT/s does not mean Ghz , but it amounts to the same thing.

1:49 AM, July 09, 2006  
Anonymous Anonymous said...

"Good try, MMM. GT/s do, more or less, mean Ghz. GT/s are certainly not gigabits/sec.

It's just like how the FSB is quad-pumped, and a 266Mhz FSB is 1066Mhz "effective." What that chart is saying is that it is effectively 4Ghz."

You couldn't be farther from the truth. Nobody has a '4Ghz effective' bus. DDR2-667 is 667Mhz. That's right, 667Mhz, not 4Ghz. If Intel had a '4Ghz effective' bus they wouldn't have the bandwidth issues they currently have.
With DDR memory you would multiply the Mhz of the memory x 8 to get your bandwidth. DDR-400 had 3.2GB/s bandwidth. I'm assuming that multiplier is different for DDR2 or FBDIMMs since 667 x 8 = 5.336GB/s and Intel is only hyping 4.8GB/s.

12:40 PM, July 09, 2006  
Anonymous Anonymous said...

The 4 Gb/s is not about the FSB. It's about the FB link! The FB link is a serial link just like HT, but even faster.

There are separate 4Gbs read and 2Gbs write links per channel, providing 1.5x bandwidth over a direct-DDR channel. FB-DIMM does this by using a custom buffer to convert serial-to-parallel.

So, the multiplier for FB-DIMM is 12. 12*667 = 8 GB/s peak. Obviously, peak is not the same as 'typical', but that goes for any bus, including direct-DDR.

Let me make it very simple:
FB-link = "HT for memory"
direct-DDR = "FSB for memory"

Can you see how both Intel and AMD have a leg in the past and one in the future?

4:33 PM, July 09, 2006  
Blogger Mad Mod Mike said...

"Can you see how both Intel and AMD have a leg in the past and one in the future?"

Can you see how even if that is true, that does not increase the performance over using regular DDR2-667 DIMM's and Intel is freakin' stupid because the latency of FB-DIMM's offsets any "advancements" it may have.

4:44 PM, July 09, 2006  
Anonymous Anonymous said...

"Can you see how even if that is true, that does not increase the performance over using regular DDR2-667 DIMM's"

No, actually I can't. What I can see is that 4 FB links use 276 pins (69 per link), slightly more than 1 (ONE!) DDR channel (240 pins) while providing 6 (SIX!) times more peak bandwidth. What's your choice for an integrated memory controller in the age of multicores?

8:25 PM, July 09, 2006  
Anonymous Anonymous said...

"No, actually I can't. What I can see is that 4 FB links use 276 pins (69 per link), slightly more than 1 (ONE!) DDR channel (240 pins) while providing 6 (SIX!) times more peak bandwidth. What's your choice for an integrated memory controller in the age of multicores?"

Why not increase the intelligence and channel count of the DRAM controller on the processor?

Remember many members of the DEC Alpha team work at AMD. And for the Alpha, they were planning on building a many channel RDRAM controller system ONTO THE PROCESSOR. Unfortunately Alpha died because HP was blinded by Intel's lies (and HP's stupidity).

In many ways, FB-DIMM is an attempt to create a poor man's RDRAM system. It's got some good things going on, but it is a compromised design, as all Intel stuff these days is.

If you truly want ultimate performance, you go with:

(a) bumping up the channel count of your on-chip RAM controller system.

[if AMD had a four channel DDR2/3 controller, it would be a big boost vs. the current two channel design]

(b) use extremely fast RAM designs... such as the follow-ons to RDRAM... XDR2, XDIMM, etc.

(c) integrate new XDR2/XDIMM compatible/comparable controllers onto the processor for low latency.

Using Rambus technology of today, you can have a single DRAM device with 16GB/sec of peak bandwidth.

Hence if AMD today built what Alpha would have had years ago, it would be something along the lines of an 8 channel XDR2 controller, giving each core of a quad-core processor 2 extremely capable RAM channels. This way each core would have 32GB/sec of peak bandwidth available to it.

Needless to say such a design would completely frag Intel's FB-DIMM.

11:22 PM, July 09, 2006  
Blogger Mad Mod Mike said...

'slightly more than 1 (ONE!) DDR channel (240 pins) while providing 6 (SIX!) times more peak bandwidth."

DDR2-667 for FB-DIMM's offer the same performance as DDR2-667 DIMM'....where the hell is this 6x coming from? Are you on crack you stupid Intel fanmonkey?

That great "buffer" on the FB-DIMM's actualyl consumes more power and causes increased latencies over standard DIMM's...it doesn't matter how many wires it takes because you can't have 12 sets of Quad-Channel RAM on a motherboard you moron, it wouldn't physically fit idiot!

11:45 AM, July 10, 2006  
Blogger Mad Mod Mike said...

I think Intel fanmonkeys need to stop saying "peak bandwidth!" because using QUAD-CHANNEL FB-DIMM DDR2-667 w/ DUAL 1333MHZ FSB's a Woodcrest system ONLY GETS 5.5GB/S REAL BANDWIDTH! That is SAD and PATHETIC!

On a Dual Opteron64 platform, using 3-3-3-8 DDR400 Registered DIMM's, I've seen up to 12GB/s which utilizes over 90% of the PEAK bandwidth...while Woodcrest can't even use 1/4 of the PEAK bandwidth.

Stop defending a worthless company that has inferior designs you stupid intel fanmonkeys.

11:48 AM, July 10, 2006  

Post a Comment

Links to this post:

Create a Link

<< Home