Unless the optimized MAC code employs a technique to lower the memory bandwith required, then I do not believe the performance gains are all code related. I have yet to see a 5000x based system running XP or Vista that can beat the memory performance of a tuned dual channel DDR2 system. Though I have no basis for this assumption, I think the MAC Pro running OSX has broken through the 10GB/sec memory bandwith to feed those 8 cores more data than XP or Vista.

But then again I am frequently wrong.

Can someone running OSX on the MAC Pro please post some memory bandwith and latency test results?

For what it's worth, again... the higher performance of the MAC app IS ONLY code related since the mac pros are actually nothing else than this.

@JDWahle why don't you give it a shot and port that code to windows ?

---

Join BOINC United now!

Unless the optimized MAC code employs a technique to lower the memory bandwith required, then I do not believe the performance gains are all code related. I have yet to see a 5000x based system running XP or Vista that can beat the memory performance of a tuned dual channel DDR2 system. Though I have no basis for this assumption, I think the MAC Pro running OSX has broken through the 10GB/sec memory bandwith to feed those 8 cores more data than XP or Vista.

But then again I am frequently wrong.

Can someone running OSX on the MAC Pro please post some memory bandwith and latency test results?

For what it's worth, again... the higher performance of the MAC app IS ONLY code related since the mac pros are actually nothing else than this.

@JDWahle why don't you give it a shot and port that code to windows ?

In that review, it is stated that, [the] "V8 delivers considerable torque to the rear wheels thanks to 16MB of total L2 cache, dual 1333MHz front-side buses, a staggering 21GB/s of memory bandwidth, and eight Core-microarchitecture CPU cores running at 3GHz each". Yet in the shootout when they pitted quad-core Xeons against Opterons, the V8 choked on the Memory bandwith tests, yielding a paltry 4GB/sec for the dual XEON 5355 system. Seems that benchmark was running two threads, running on the same chip, thus utilizing only one of the available FSBusses. Even by doubling the that figure to compensate, bandwith might have reached 8GB/sec, still less 37% efficiency!!!

Does anyone know of similar benchmarks on the MAC Pro running OSX with all 8 slots filled with matching FB-DIMMS? Only when I can see those OSX benchmarks will I believe that this performance discrepancy is totally code related.

I apologize if this an old subject that has been discussed before. With me being rather new to these boards if someone could point me to any prior discussion I'd be much appreciative. TIA.

Can anyone come up with a 4 core MAC Pro to compete against this 4 core Vista box currently in 17th place on the Top Hosts list running Crunch3r's Vista build? Show me one with higher RAC and I might then believe that it is the code. This one doesn't count because he is a cherry picker!

@Crunch3r - I've only just began to look at the stock S@H code, Do you know how I can get it to build using MS Visual C++ 2005 Express Edition? (Where is the .prj file?) I would like some time gaining familiarity and experimenting before I even think about porting code.

---

@Crunch3r - I've only just began to look at the stock S@H code, Do you know how I can get it to build using MS Visual C++ 2005 Express Edition? (Where is the .prj file?)

You don't need a .prj file simply doubleclick on the seti_boinc.sln (solution file) it'll load all the projects.

I would like some time gaining familiarity and experimenting before I even think about porting code.

The last time i tried, about 4 moths ago VS 2005 could't compile the stock source (V6)... However that might have changed now.

Personaly i wouldn't waste my time on the stock code at all since what you've been looking for (Why the osx code is faster) is in here.

---

Join BOINC United now!

JDWhale wrote:

Can someone running OSX on the MAC Pro please post some memory bandwith and latency test results?

Tell me where to find the testing programs you would like to see benchmarks from...and I will try to get you some results to compare...

---

JDWhale wrote:

Can someone running OSX on the MAC Pro please post some memory bandwith and latency test results?

Tell me where to find the testing programs you would like to see benchmarks from...and I will try to get you some results to compare...

@UL1 - Thanks, I wish I knew.

Not being familiar at all with MAC, I don't even know if such test programs exist (but I think they must). So, I am appealing to those knowledgable about benchmarking MAC under OS X to share memory bandwith & latency test programs.

It seems that all the benchmarks focus on video editing, encoding, (etc.), which as resolutions and video quality rise, can put more demand on memory performance when using an application that runs across all 8 cores.

I'm still looking,
John

---

LOL...

Funny you ask. When these discussions began about the Intel 5000 series chipsets I started looking around for some benchmarks (in reviews and elswhere) for the new MacPro's to compare to the Intel boards. I'm not saying there aren't some hard numbers out there, but they sure aren't easy to find! ;-)

That being said, as Crunch3r mentioned, I always assumed the MacPro was essentially the same as an Intel S5000XVN. However, when I was researching the FBDIMM memory architecture I discovered some interesting things about it. For one, it is possible the Northbridge/MCH is an Apple customized version of the 5000 which has different firmware for the MCH more optimized for workstation applications which is the MacPro's main market niche. Apple has developed a kind of 'insider' relationship with Intel in recent years, so I wouldn't be surprised if Intel doesn't give them some 'special' consideration on matters like this.

The Intel S5000XVN OTOH is a more general purpose board targeted towards small servers as well as higher end workstations, and thus the MCH may be setup differently to afford somewhat less efficient performance but more suitable for a wider spectrum of applications.

One other observation about Skulltrail. In virtually all the technical discussions and analyses about FBD I've read, having one slot per channel is probably the worst way to design the board. IIU the articles correctly you would have to make sure your DIMMs were dual rank and populate all 4 slots in order to get any tangible benefit from the architecture over DDR2.

Alinator

Alinator wrote:

LOL...

That being said, as Crunch3r mentioned, I always assumed the MacPro was essentially the same as an Intel S5000XVN. However, when I was researching the FBDIMM memory architecture I discovered some interesting things about it. For one, it is possible the Northbridge/MCH is an Apple customized version of the 5000 which has different firmware for the MCH more optimized for workstation applications which is the MacPro's main market niche. Apple has developed a kind of 'insider' relationship with Intel in recent years, so I wouldn't be surprised if Intel doesn't give them some 'special' consideration on matters like this.

Thanks for that report, Alinator.

I saw this Mac Pro wiki earlier today showing the MacPro use of Extensible Firmware Interface (EFI). Available when booting OSX & linux, but not available when booting WinXP using Bootcamp.

Apple's Boot Camp provides backwards-BIOS-compatibility, and dual and triple boot configurations can be achieved.[24] These operating systems are easily installable on Intel x86 based Apple computers:

• Mac OS X 10.4.7 and later (native, OS supports EFI)
  
• Microsoft Windows XP and Vista 32-bit & 64-bit (hardware drivers are included in Boot Camp) (only supports BIOS)
  
• Linux (most late 2006 or 2007 distributions) (supports both BIOS and EFI)
  

I wonder if maybe that is where the high performance memory/chipset drivers are hiding? Somewhere else Linux founder, Linus T. had some dissenting remarks about EFI. Goggle if interrested.

I hope that we're making some progress on this subject. Maybe if we can get the EFI available when booting WinXP on the Mac Pros, or maybe pushed somehow on other D5000x MOBOs, so we can do some "Apples-to-Apples"(pun intended) comparisons.

---

I'm curious if anybody following this thread either has or has access to the Dell Precision T7400 and is willing or able to run some benchmarks... SiSoftware SANDRA (Lite version is free for personal use), or maybe Everest Home Edition(Sorry I can't find link to that specific product today, maybe not as reliable, but easier to run from a USB device without doing an install) read/write/latency tests.

Maybe Dell is a similar partner with Intel and can get their hands on the "good stuff".

Cheers,
John

---

I wonder if maybe that is where the high performance memory/chipset drivers are hiding?

There is no hiding...
The Mac pros ("old" ones) have the same performance for memory/chipset like the s5000xvn...

why are you so obsessed about that one ?

However look for yourself ... LINK

The whole key IS the source code.... nothing else.

---

Join BOINC United now!

I wonder if maybe that is where the high performance memory/chipset drivers are hiding?

There is no hiding...
The Mac pros ("old" ones) have the same performance for memory/chipset like the s5000xvn...

why are you so obsessed about that one ?

However look for yourself ... LINK

The whole key IS the source code.... nothing else.

LOL...

How did I miss that one! :-)

This compares virtually the same to the reviews I've seen for the S5000XVN. So I guess it's safe to say the memory system is not the reason for any significant differences and the answer must be software. ;-)

Alinator

... So I guess it's safe to say the memeory system is not the reason for any siginificant differences and the answer must be software. ;-)

Finaly :)
just sad that i had to keep posting it over and over again...

OK... once more
IT'S THE CODE and NOT the "holy apple sticker" on the case that makes the difference!

---

Join BOINC United now!

... So I guess it's safe to say the memeory system is not the reason for any siginificant differences and the answer must be software. ;-)

Finaly :)
just sad that i had to keep posting it over and over again...

OK... once more
IT'S THE CODE and NOT the "holy apple sticker" on the case that makes the difference!

Agreed. This isn't to say that there aren't applications where the S5000XVN will outperform the Mac due to hardware, but SAH isn't one of them. ;-)

Alinator

Crunch3r wrote:

Finaly :)
just sad that i had to keep posting it over and over again...

OK... once more
IT'S THE CODE and NOT the "holy apple sticker" on the case that makes the difference!

Thanks Crunch3r, for your persistence. After I've had the bruise on my forehead a while, I might learn to open the door before trying to exit the room :)

I haven't had time to DL the V8 code you pointed me toward yet, but I'll assume that part of the hand massaging in that code reduces main memory reads and does more with the data each time it's cache. Working smart not hard is the better path.

Thanks for being persistent with me, now it's time for me to go deep into my cave and hibernate.

Ciao,
John

---

LOL...

No problemo, and don't feel bad.

A lot of us are from Missouri. ;-)

Alinator

LOL...

A lot of us are from Missouri. ;-)

Descended from indeed. Though I've never been there, Daddy-O was born & raised near Bunceton, MO. :-D

I guess it shows :)

---

Hey it's not sticker thankyou!

It's etched!

---

Since we now figured out that the speed of the mac app is due to the code beeing used, i want to get back to the "bandwith issue" previously metioned here.
So i ran a test with the latest RightMark Memory Analyzer.


The first pic shows total bandwith reading memory ... running 8 threads simultaniously.




The second one shows bandwith reading/writing simultaniously to the memory using 8 threads for each core.



MB: S5000XVN
MEM: 4x1 GB DDR2 FB-DIMM 667 MHz


Now it would be interesting to see some test results of a Core2Quad using DDR2 800 MHz DIMMS. (Especialy the second test would be interesting RW)

---

Join BOINC United now!

Since we now figured out that the speed of the mac app is due to the code beeing used,

Sorry, I missed that one. What's the extra optimisation for the Mac?

i want to get back to the "bandwith issue" previously metioned here.
So i ran a test with the latest ...

The first pic shows total bandwith reading memory ... running 8 threads simultaniously. ...

The second one shows bandwith reading/writing simultaniously to the memory using 8 threads for each core. ...

MB: S5000XVN
MEM: 4x1 GB DDR2 FB-DIMM 667 MHz

What is that showing exactly for those numbers? And why those numbers?

Mmmm... Me puzzled!

Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Crunch3r wrote:

Now it would be interesting to see some test results of a Core2Quad using DDR2 800 MHz DIMMS. (Especialy the second test would be interesting RW)

Core2 Quad Q6600 3.2GHz w/ DDR2 800MHz @ 1000MHz

Read on 4 cores...



and reading/writing simultaniously...





MB: Gigabyte P35-DS3L
MEM: 2x1 GB DDR2 DIMM 800MHz @ 1000MHz


Sorry, I can't shut her down to reset memory OC right now :(

---

More of the same.


:-}

I wonder if there is any difference between versions 1.0 & 1.1 :o

---