PDA

View Full Version : Understanding the Mac Pro's FB-DIMM memory architecture


MacBytes
Aug 17, 2006, 10:54 PM
http://www.macbytes.com/images/bytessig.gif (http://www.macbytes.com)

Category: Tips and How To's
Link: Understanding the Mac Pro's FB-DIMM memory architecture (http://www.macbytes.com/link.php?sid=20060817235405)
Description:: It's not all recordbreaking speed and low latency with Apple's newly chosen memory architecture....

Posted on MacBytes.com (http://www.macbytes.com)
Approved by Mudbug

Anonymous Freak
Aug 17, 2006, 11:22 PM
http://www.macbytes.com/images/bytessig.gif (http://www.macbytes.com)

Category: Tips and How To's
Link: Understanding the Mac Pro's FB-DIMM memory architecture (http://www.macbytes.com/link.php?sid=20060817235405)
Description:: It's not all recordbreaking speed and low latency with Apple's newly chosen memory architecture....

Posted on MacBytes.com (http://www.macbytes.com)
Approved by Mudbug

One note: The article mentions that a Woodcrest workstation based on DDR2 would likely be faster. That is true, but there are no DDR2 chipsets for Woodcrest right now. And given DDR2's apparent limit of 2 DIMMs per channel, 2 channels max, that would leave a workstation with a (current) maximum of 4 DIMM sockets, compared to the Mac Pro's 8. For a true 'workstation' level computer, more memory is better.

For a 'desktop', DDR2 is better, and a fast Conroe (a.k.a. Core 2 Duo or Core 2 Extreme,) would be the king. But if you need more than 8 GB of RAM, or more than 2 processor cores, Woodcrest/FB-DIMM is not only your only choice, but really is the better choice.

WildPalms
Aug 17, 2006, 11:40 PM
One note: The article mentions that a Woodcrest workstation based on DDR2 would likely be faster. That is true, but there are no DDR2 chipsets for Woodcrest right now. And given DDR2's apparent limit of 2 DIMMs per channel, 2 channels max, that would leave a workstation with a (current) maximum of 4 DIMM sockets, compared to the Mac Pro's 8. For a true 'workstation' level computer, more memory is better.

For a 'desktop', DDR2 is better, and a fast Conroe (a.k.a. Core 2 Duo or Core 2 Extreme,) would be the king. But if you need more than 8 GB of RAM, or more than 2 processor cores, Woodcrest/FB-DIMM is not only your only choice, but really is the better choice.


ehurtley, you are absolutely right. FB-DIMM is about great amounts of working memory for workstations, albeit with a small tradeoff in overall memory speed. The speed hit is very very minimal considering the benefits achieved to enable real time modelling, cad cam, etc which ideally work best with greater amounts of ram than the average application.

cube
Aug 18, 2006, 04:13 AM
Another reason why Apple needs to provide a midrange Core 2 machine. I was waiting for FB-DIMM. I knew it was going to be slower, but this is total crap.

cube
Aug 18, 2006, 04:28 AM
For a 'desktop', DDR2 is better, and a fast Conroe (a.k.a. Core 2 Duo or Core 2 Extreme,) would be the king. But if you need more than 8 GB of RAM, or more than 2 processor cores, Woodcrest/FB-DIMM is not only your only choice, but really is the better choice.

The better choice is Opteron.

BenRoethig
Aug 18, 2006, 08:33 AM
The better choice is Opteron.

Which is significantly slower and more power hungry than the Xeon 5100 series.

cube
Aug 18, 2006, 09:07 AM
Which is significantly slower and more power hungry than the Xeon 5100 series.

http://www.sun.com/servers/x64/x2200/

RUAerospace
Aug 18, 2006, 10:00 AM
http://www.sun.com/servers/x64/x2200/

http://www.anandtech.com/showdoc.aspx?i=2793&p=3
http://www.techreport.com/etc/2006q2/woodcrest/index.x?pg=1
http://www.hexus.net/content/item.php?item=6256&page=6

Generally, its wiser to look for benchmarks from a source OTHER than the company thats trying to sell and market the product. ;)

cube
Aug 18, 2006, 10:28 AM
http://www.anandtech.com/showdoc.aspx?i=2793&p=3
http://www.techreport.com/etc/2006q2/woodcrest/index.x?pg=1
http://www.hexus.net/content/item.php?item=6256&page=6

Generally, its wiser to look for benchmarks from a source OTHER than the company thats trying to sell and market the product. ;)

Those are comparisons against OLD Opteron systems, not the new Socket F which are about 21% faster.

Anonymous Freak
Aug 18, 2006, 11:10 AM
Those are comparisons against OLD Opteron systems, not the new Socket F which are about 21% faster.

If it really is, then great. The competition will keep Intel on their toes, and let us have faster processors (from both companies,) more cheaply.

But when the only benchmarks for the new product are from a company SELLING the product, I don't trust them. I didn't trust Intel's benchmarks for Conroe late last year (before independent reviewers had had a chance to test it,) I don't trust the new Opteron numbers from Sun. I'll wait for an independent source to confirm how fast the new processors are.

(I love Sun's 'power saving' numbers. The 'high power' Opterons draw 120W compared to the high end Xeon's 80. The 'low power' Opterons draw 55W compared to the middle-of the road Xeon's 65, or the low power Xeon's 40.)

piltupso
Aug 18, 2006, 01:06 PM
Ok I admit I am not and engineer, could anybody give a quick digest of the article. From my read what I got was that the memory could be slightly slower but the author wasn't sure how slow or how often. Am I close? Also it seemed to say the other features of the Xeon and motherboard gave gains in other areas as a trade off to using the more expensive slower memory?

amin
Aug 18, 2006, 01:30 PM
I too don't know what to make of all this. Is the Mac Pro quad core 2.66GHz model with 4x512MB FB RAM (which I just ordered :rolleyes: ) significantly slower in general than a dual core 2.66GHz Conroe machine with 2 x 1GB regular DDR2 RAM? I guess it won't matter much to me. The fastest computer I have ever owned is my G4 Powerbook, so I'm guessing I will be happy with the performance!

mkaake
Aug 18, 2006, 02:00 PM
I too don't know what to make of all this. Is the Mac Pro quad core 2.66GHz model with 4x512MB FB RAM (which I just ordered :rolleyes: ) significantly slower in general than a dual core 2.66GHz Conroe machine with 2 x 1GB regular DDR2 RAM? I guess it won't matter much to me. The fastest computer I have ever owned is my G4 Powerbook, so I'm guessing I will be happy with the performance!

Yeah, it's pretty easy to loose sight of that... people will throw their arms up about the slower memory access times, but ignore that the machine is still faster that the Quad G5...

Anyhew, FB-DIMM technology will only move forward from here, and it provides the expandibility required to keep our memory hungry machines filled up...

cube
Aug 18, 2006, 03:38 PM
Anyhew, FB-DIMM technology will only move forward from here, and it provides the expandibility required to keep our memory hungry machines filled up...

An Opteron machine does not need this crap because you get a dual channel memory controller on each CPU.

cube
Aug 18, 2006, 03:40 PM
(I love Sun's 'power saving' numbers. The 'high power' Opterons draw 120W compared to the high end Xeon's 80. The 'low power' Opterons draw 55W compared to the middle-of the road Xeon's 65, or the low power Xeon's 40.)

You should love Intel's. Intel's TDP is "typical", while AMD's TDP is maximum.

RUAerospace
Aug 18, 2006, 04:13 PM
Those are comparisons against OLD Opteron systems, not the new Socket F which are about 21% faster.

Actually the only applicable difference between the models is a bumping up of the clockspeed from 2.6 GHz to 2.8 GHz which at most, amounts to a 7% increase in performance (the move to DDR2 is equivalent to the AM2 platform change of the Athlon64; no real gains and some losses actually). It really makes no difference in the comparison considering A) Woodcrest typically beats the Opteron by far more than 7% in the majority of the benchmarks and B) AMD has already had equally specced (DDR2 equipped) 2.8 GHz Athlon64 FXs out for months now that were subsequently demolished by Conroe in benchmarks.

The general concensus right now among the tech community is that clock-to-clock, Woodcrest typically demolishes the Opteron in integer performance while providing marginal to considerable (2-15%) gains in floating point and vector. Unless AMD seriously changes the core in some way, it is doubtful that the Opteron can (generally) equal Woodcrest's performance without scaling to over 3 GHz. Of course, thats not taking into account power and heat dissipation and the fact that woodcrest is able to scale to higher clockspeeds far more easily.

RUAerospace
Aug 18, 2006, 04:16 PM
You should love Intel's. Intel's TDP is "typical", while AMD's TDP is maximum.

I think theres little doubt among anyone with real knowledge that Core 2 consumes far less power and dissipates far less heat than Opteron or Athlon64.

cube
Aug 18, 2006, 04:33 PM
Actually the only applicable difference between the models is a bumping up of the clockspeed from 2.6 GHz to 2.8 GHz which at most, amounts to a 7% increase in performance

21% (http://sharikou.blogspot.com/2006/08/ibm-socket-f-opteron-21-faster.html)

RUAerospace
Aug 18, 2006, 05:05 PM
I think we have vastly different conceptions of what accounts as evidence Cube. Instead of actual benchmarks, all you've done was use a corporate marketing quote stating that new IBM servers can perform up to 21% over previous revisions under the same power envelope (extrapolating from AM2, it is largely a given that the Revision F Opterons will consume considerably less power). This, you have further misconstrued into the statement that the Revision F Opterons are generally 21% faster than their predecessors (using that kind of misrepresentation, I might as well be saying that the Core 2 Duo are generally 80% faster than the opteron, well it probably is if you take into account power consumption). Frankly this doesn't stack up against the evidence I have already given. If you read the press releases (http://www.dailytech.com/article.aspx?newsid=2971) for the Revision F Opteron, you would've seen that the only difference between a Revision F Opteron and a Revision E was the inclusion of DDR2 support. For any significant purpose, that doesn't change a thing performance-wise (http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2762&p=12). In fact, all the modifications in the history of the Athlon64 wouldn't add up to a 21% increase in typical clock-to-clock performance!

cube
Aug 18, 2006, 05:10 PM
SPEC figures are generated by the vendors. Sun's numbers are being verified by SPEC. Socket F was only launched 4 days ago!

RUAerospace
Aug 18, 2006, 05:18 PM
SPEC figures are generated by the vendors. Sun's numbers are being verified by SPEC. Socket F was only launched 4 days ago!

The article you've cited doesn't mention SPEC in any way. Frankly, I don't think theres anything else to discuss at this point. I think we're pretty clear on whats true and whats not.

mkaake
Aug 18, 2006, 06:02 PM
An Opteron machine does not need this crap because you get a dual channel memory controller on each CPU.

great!

but anyway, I think it's a good direction for technology to move... the amount of space saved on the mobo is pretty nice, and once they speed it up, it'll be even better. New tech will only get better.

amin
Aug 18, 2006, 08:42 PM
I have now read posts by a number of folks stating that Conroe desktops will perform better than Woodcrest desktops due to this RAM latency issue. Is this really true despite the extra dual-core processor, faster FSB, and greater memory bus bandwidth on the Mac Pros? Are there any practical, head-to-head, hands-on comparisons of a quad Woodcrest system with a dual Conroe system? I'm wondering whether this latency issue is overblown. I'm also confident that as more apps are written to take advantage of four cores, the current Mac Pros will shine even brighter.

cube
Aug 19, 2006, 04:47 AM
The article you've cited doesn't mention SPEC in any way. Frankly, I don't think theres anything else to discuss at this point. I think we're pretty clear on whats true and whats not.

See the Sun link above.

amin
Aug 19, 2006, 07:58 AM
Sorry to be rude, but would you all kindly take the Opteron supremacy debate to another thread and allow this one to be about "Understanding the Mac Pro's FB-DIMM memory architecture"?