Splash: What are you using for new systems at work?

[Santilli]

OK:
I see the point. Do the Xeon's have to be matched to work together, like the AMD processors :?: :cursin:

 

[Santilli]

By the way, how do Tyan and Supermico compare with updating their boards for new processors?

In other words, if I buy a Supermicro board, dual for two Xeon 3.06, and, next week, they come out with 3.5ghz models, will supermicro support those with a bios update?

What are the probabilities of a new version of the Xeon, and how high do you think the Xeon chip can be scaled,before Intel changes the pin design for the chip and mobo?

Thanks

gs

 

[Santilli]

Hmm. Another issue about Supermicro is they are one of the few US made boards.

Another one is they are designed to work with multiple scsi cards, and, my Asus is blue screening every once and awhile, after I put the Adaptec UW card in it. That's 3 scsi cards in the same machine, at 3 speeds...

Just took it out. I guess I'll try it in the P3 and backup to the drives that way...

Much slower, however.

gs

 

[Explorer]

So what's the point in getting a dual processor board?

It's the only way to get a mobo with an "advanced" expansion bus (i.e. -- PCI-X). Also, remember that the better of these sophisticated mobos with dual Ultra-320 SCSI channels, like the Supermicro X5DA8 mentioned earlier, has the SCSI channels talking to PCI-X/133, not the 64-bit - 33MHz/66MHz PCI segment.

The onboard GbE adaptors also talk to PCI-X. From there, you can plug in other PCI-X or compatible PCI adaptors into the PCI-X expansion bus. It's too bad that PCI-X has never been interposed into a chipset for the single processor Pentium 4. I suspect that ServerWorks would do something like this if there was any serious demand for mobos with ordinary P4 processors talking to a PCI-X bus (probably not).

By the way, how do Tyan and Supermico compare with updating their boards for new processors?

I can tell you for certain that Supermicro generally does NOT update their BIOS downloads much, because they hardly ever need updating. They do update the BIOSes of affected mobos, though, when a new major stepping occurs and/or processor microcode becomes available so that they will run with these updated processors. Otherwise, I've seen very little in the way of bug fixes because the BIOSes are pretty damned well tested beforehand (i.e. -- not using customers as guinea pigs).

In other words, if I buy a Supermicro board, dual for two Xeon 3.06, and, next week, they come out with 3.5ghz models, will supermicro support those with a bios update?

Only if the chipset on that mobo will work with a 3.5 GHz Xeon. Of course, it won't, because the end of the line for the current Xeon family ("Northwood") is at 3.2 GHz, where yet another Xeon will take over ("Prescott") at 3.4 GHz. "Prescott" will require a whole new mobo for many many reasons.

What are the probabilities of a new version of the Xeon, and how high do you think the Xeon chip can be scaled,before Intel changes the pin design for the chip and mobo?

"Prescott" will require a whole new mobo for many many reasons, and one of those many many reasons is that IT WON'T EVEN HAVE PINS! So, you have this newfangled Land Grid Array (LGA) socket for "Prescott" with more contacts than current "Northwood" has pins, and then you also have new lower operating voltages for the processor.

And just extrapolating a bit here... I suspect that the current Opteron processors are fast approaching half life, meaning that they will be yesterday's news around about may or June of 2004. Somewhere around then or a little later, a new Opteron will have to appear, an Opteron that supports DDR2 memory. So, when those show up, you'll have to get a whole new mobo for the new processor and the new memory that it supports.

Do the Xeon's have to be matched to work together, like the AMD processors

This is certainly nothing exclusive to the Athlon. Yes, a pair of Xeons must be of the same stepping before SMP will occur, otherwise only Processor #1 will be used by Win2K, WinXP, WinNT, BeOS, Linux, or UNIX. On a related note, after a couple of weeks of testing, I finally picked up a pair of factory-matched small diaphragm condensor microphones this weekend. Each and every Oktava microphone is hand assembled in Russia by skilled craftspeople from hand selected parts...

• If you can just hold on for about a week or so, I will be getting in several more Xeon processors in at work. These Xeons are 2.4 GHz, 3.06 GHz, and 3.06 GHz + 1 MB L3 cache processors -- all running @ 533 MHz FSB.
  
  I will benchmark them to see *approximately* what their VALUE is -- their bang/buck ratio. In other words, benchmark score (where more is better) divided by cost = value. This will also be an SMP versus non-SMP test for value. Even though all but two of these boxes are destined to run Windows 2003 Server or Windows 2000 Server, I will only use Windows 2000 Workstation (Win2K Pro) / SP-4 as the host operating system and I will only use an X5DA8 workstation mobo (not the X5DP8-G2 or X5DPi-G2 mobos).

_________________________________________________________________________


Failed sending email ::

DEBUG MODE

Line : 225
File : /home/handruin/public_html/forum/includes/emailer.php

 

[Santilli]

I wait with baited breath...

Funnier then you know. If Bill gets around to posting them, I've been trying to get the cat to play fetch, and she does, when she wants too.

She's actually very good at it. However, at some point she decides fetch should be hide the mouse, and have Dad look for it.

Currently, that's the game.

OK. Yes, I'm not in any hurry, since my Toyota just ate about 500 bucks more then I expected, needing front axles.

gs

 

[Santilli]

Now, the DVD player arrived. How do I figure out what SCSI id's are taken by the plextor stuff? I think I left the burner, and the reader on default.
4 and 6? Can't remember...

gs

 

[Santilli]

Stupid Question. It's 3&4, and I've got it.

Thanks

gs

 

[.Nut]

Stupid Question. It's 3&4, and I've got it.

SCSI "6" has highest SCSI device priority and is often used for CD/DVD.

SCSI scanners sometimes use "6."

 

[Jan Kivar]

Stupid Question. It's 3&4, and I've got it.

SCSI "6" has highest SCSI device priority and is often used for CD/DVD.

SCSI scanners sometimes use "6."

Hmm... I thought that ID7 has the highest priority (assigned to the HBA), and then it goes from ID0 to ID6, ID6 having the lowest priority.

PLEASE correct me if I'm wrong.

Cheers,

Jan

 

[Platform]

Hmm... I thought that ID7 has the highest priority (assigned to the HBA), and then it goes from ID0 to ID6, ID6 having the lowest priority.

PLEASE correct me if I'm wrong.

No need to mention "7" in the PC world, since we know a SCSI peripheral isn't normally setup for using "7" (older SGI boxes reserve "4" for its SCSI controller).

Once SCSI-2 was introduced, highest to lowest in SCSI priority became:

7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8.

 

[Jan Kivar]

No need to mention "7" in the PC world, since we know a SCSI peripheral isn't normally setup for using "7" (older SGI boxes reserve "4" for its SCSI controller).

Once SCSI-2 was introduced, highest to lowest in SCSI priority became:

7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8.

Why all the SCSI drives ship with ID0? [as You can see, I've never owned a SCSI drive, nor have much experience configuring them...]

Jan

 

[Handruin]

That's very interesting... I found more (of the same) info here.

Jan, the link I posted says:

Another consideration is that if you have any devices that absolutely cannot tolerate delays in receiving their stream of data--such as a CD recording drive or a video encoder--they should be given top priority on the bus. Many people also like to make the host adapter the highest-priority device on the bus, which is why host adapters will often default to a SCSI ID of 7. It should be noted that some older host adapters can be finicky about device IDs. Some will only boot a hard disk if it is set to device ID 0. (This is inflexible and has been basically done away with in newer hardware.)

 

[Jan Kivar]

That's very interesting... I found more (of the same) info here.

Jan, the link I posted says:

Another consideration is that if you have any devices that absolutely cannot tolerate delays in receiving their stream of data--such as a CD recording drive or a video encoder--they should be given top priority on the bus. Many people also like to make the host adapter the highest-priority device on the bus, which is why host adapters will often default to a SCSI ID of 7. It should be noted that some older host adapters can be finicky about device IDs. Some will only boot a hard disk if it is set to device ID 0. (This is inflexible and has been basically done away with in newer hardware.)

Thanks, Handy. Me no good with SCSI.

Cheers,

Jan

 

[mubs]

Anytime the issue of a dual-proc system comes up, there seems to be much gnashing of teeth.

Even if every program one runs is SMP-unaware, is there not a benefit to having a second CPU? A CPU eating program like Santilli's possessed Photoshop can hog one processor, and the OS and all other progs like mail client, browser, etc. can time-slice on the other. Will this not make for a more responsive system overall? I would even say that in a situation like this, there is a benefit to having a program not being multi-threaded because then it can't hog the second processor as well.

So long as the OS can use dual CPUs, I think there is still a benefit. Apparently, Opterons scale 80-90% compared to Xeons which scale at only 20-30% (saw this somewhere - Inquirer?).

Do I make sense or am I blabbering? Flame away!

 

[Corvair]

Anytime the issue of a dual-proc system comes up, there seems to be much gnashing of teeth...

Possibly to some folk's chagrin, I focused upon *value* over mere capability: SMP boxes match up much better to multithreading / multiprocessing applications such as Photoshop or Premiere, and even more so with deeply multithreading / multiprocessing applications such as Oracle DBMS and MS-SQL because there is a greater return on investment, hence, the value of SMP with capable applications, or conversely, the diminished value of SMP with nothing more than simple applications. In the end, one needs to understand the holy triangle of value, requirements, and BUDGET.

...Even if every program one runs is SMP- un aware, is there not a benefit to having a second CPU?

Yes, a small benefit, but at a significant cost -- meaning lower value than a single processor.

On an SMP box, a modern operating system that supports multithreading and multiprocessing will always take advantage of its own capabilities, by running O/S background processes such as print spooling on PROC #2 when PROC #1 is running a Lotus 1-2-3 calculation. The gain that a computer jockey notices from an SMP box over a mono-processing box, using the same microprocessor in both cases, is simply going to be minimal given today's processing speeds -- and this is not mentioning that the Northwood P4 and Xeons as well as the upcoming Prescott family of microprocessor in these mono-processor boxes are capable of hyper-threading.

...A CPU eating program like Santilli's possessed Photoshop can hog one processor,...

I always thought the CPU eating program was Pac-Man, where it trots across yer screen eating CPU cycles. I digress, but go ahead please.......

...and the OS and all other progs like mail client, browser, etc. can time-slice on the other. Will this not make for a more responsive system overall? I would even say that in a situation like this, there is a benefit to having a program not being multi-threaded because then it can't hog the second processor as well.

A multiprocessing O/S will control every application's access to CPU time, memory usage, and I/O usage. So, it's pretty much dependent upon how well the multiprocessing O/S handles these simple application programs as far as the priority level an application runs at, or how much "preemptive-ness" the multiprocessing O/S uses on an application that acts like a CPU hog in order to maintain balanced processing throughput. By the way, there have also been numerous Asymmetrical multiprocessing designs in computing -- mostly in the past, on mainframes -- where each processor runs its own house and addresses partitioned memory, not the shared memory of the symmetrical multiprocessing architecture. As far as I know, asymmetrical multiprocessing hardware and operating systems are nowadays an extreme niche product.

...So long as the OS can use dual CPUs, I think there is still a benefit. Apparently, Opterons scale 80-90% compared to Xeons which scale at only 20-30% (saw this somewhere - Inquirer?)...

I'm not sure I would believe anything they say in the Inquirer, much less the National Enquirer. Isn't the Inquirer some sort of bad joke???

 

[Fushigi]

...So long as the OS can use dual CPUs, I think there is still a benefit. Apparently, Opterons scale 80-90% compared to Xeons which scale at only 20-30% (saw this somewhere - Inquirer?)...

I'm not sure I would believe anything they say in the Inquirer, much less the National Enquirer. Isn't the Inquirer some sort of bad joke???

Well, it is possible to have highly efficient scalability with SMP. Dunno what Intel is doing but in AS/400-land, we had a 2-way machine with a performance rating of 1050 CPWs (IBM variant on TPC-C). Upping to a 4-way brought the perf rating to 2000. About 90% scalability.

On a current iSeries 825, the 3-way is rated at 3600CPWs and the 6-way is rated at 6600. The 3-way is made up of 3 POWER4 dual-core CPUs with 1 core disabled but full cache access and the 6-way just enables the dormant cores. The scalability is still very good at 83%.

 

[Mercutio]

Are AS400s NUMA systems?

 

[Fushigi]

Are AS400s NUMA systems?

I don't believe they are; I'll have to look further. A few years ago when IBM was buying SEQUENT, I was at a user conference. The IBMers there were hyped because IBM was said to be just developing their NUMA tech and hoped for a good jump start from teh SEQUENT tech. So I'll say early boxes were not but I can't yet say about current machines.

 

[Santilli]

Which handles SMP better, 2000 or XP?

gs

 

[Mercutio]

Are AS400s NUMA systems?

I don't believe they are; I'll have to look further.

Hm. Have to wonder how they get near-linear scalability using shared memory. Not that I know enough about the electrical engineering aspects of design to comment further.

 

[Mercutio]

Which handles SMP better, 2000 or XP?

I'd suggest something with a name that ends in "Server", actually, since those products are a little better tuned for the large amounts of RAM that tend to go hand-in-hand with MP machines.

 

[Onomatopoeic]

Which handles SMP better, 2000 or XP?

I'd suggest something with a name that ends in "Server", actually, since those products are a little better tuned for the large amounts of RAM that tend to go hand-in-hand with MP machines.

The only other bad thing about a WinServer product (Win2K and Win2K3), other than the price, is that some software packages refuse to run or install on Server because it detects that it is indeed Server that it is being installed on -- not 2K Workstation, XP Professional, or XP Home -- as they have a "Server" version of said software, or you have to additionally purchase client pack software to add some number of licenses before you can install it or run it.


================================================


On an important tangent here:

Something that I meant to bring up earlier was PCI-X versus PCI Express. I've known for a while what PCI Express is, but what I still haven't gotten any info on -- until today -- is just what sort of PCI Express configurations will be offered on upcoming desktop mobos.

I held off talking about PCI Express until I could give a local source a telephone call, someone who could likely be able to tell me some quasi-secret details about PCI Express products (mobos) coming within the next several months. (Unfortunately, another local source who was always pretty damned good about funneling me accurate information on upcoming hard drives and tape drives well in advance of their release is no longer employed in the reseller industry.)

So, the following is what I got today, and it basically isn't anything proprietary or secret, instead it simply hasn't been discussed in detail by Intel and other, as far as I know:

The first iteration of PCI chipsets -- for the desktop -- will all give you a 16X PCI Express slot for a compatible graphics adaptor. Additional PCI Express slots will be optional, but unfortunately, none of these slots will be any faster than 1X (1X = a single serial PCI link) for desktop mobos. That's not to say that a 1X PCI Express slot is slow, as it is equivalent in speed to a conventional 32-bit / 66 MHz PCI slot, but I was hoping that 1X, 2X, and/or 4X PCI Express slots along with add-in 2X and/or 4X PCI Express cards that could handle a 4X PCI Express connection (1024 MB/s) would be available early on. In other words, PCI Express slots on par with PCI-X/133 (1066 MB/s) will not be available on desktop-class mobos for a while. On upcoming server mobos and their chipsets, it will be a different situation for PCI Express.

The number of PCI Express slots on desktop mobos, beyond the 16X slot for PCI Express graphics adaptors, will probably be no more than 4 slots (4-each 1X slots). One very nice thing about PCI Express is that it has very low latency, because each PCI Express segment is a full-duplex serial connection that is connected *directly* to the memory controller hub and not to a PCI bridge.

Some of the deluxe desktop mobos may have a mix of PCI Express and "old" 32-bit parallel slots (probably @ 33 MHz). The conventional parallel PCI slots, however, will be bridged off of a PCI Express segment, which probably translates into a maximum of pair of 32-bit / 33 MHz PCI slots per 1X PCI Express segment. So, we might very well see a "deluxe" mobo with 3-each PCI Express slots and a 2-each parallel 32-bit/33MHz PCI slots, or 2-each PCI Express slots and 4-each parallel 32-bit/33MHz PCI slots -- and, of course, the PCI Express graphics slot.

The server-class PCI Express mobos will come along after the desktop-class PCI Express mobos. All of this will happen in 2004, with the desktop mobos possibly trickling out as early as December 2003 (probably late December, and then probably just samples and review units).

 

[mubs]

Corvair,

There's no doubt at all that I defer to your deeper knowledge and experience. I know a little bit here and there, just enough to be dangerous without being useful.

Having said that, my earlier post was in no way intended to be critical of your posts. I was just sharing my subjective experiences.

In terms of a good OS doing pre-emptive scheduling, I've experienced what you said on Unix systems. But it appears Windows is not very good with pre-emptive scheduling. My good friend Cas will no doubt disagree. If Windows was good, why does Santilli's Photoshop hog the processor? In his own words, he is unable to do anything at all on his machine with Photoshop running. When this kind of proggie becomes unresponsive, it is many times impossible to even shutdown/restart windows; one is forced to hit the reset button. When you put a CD into your CD-ROM drive, Windows is unresponsive for some time. Whether it is a problem with the OS or poorly written device drivers, the problem exists (from my layman's point of view). I've seen low-end single-processor Intel servers running Unix doing amazing amounts of work, and dual-processor higher-end (comparitively) Intel servers running NT running like molassess doing just print-file sharing. My subjective experience has been that Windows needs a lot more iron to run decently.

I built my current PC 12/2000; Tyan Tiger 100, 2xP3-800, W2k Pro. Though the procs are laughably slow by today's standards, it fairly responsive. I believe it is mainly because even if one proc is busy or is "captured" by a bad/unresponsive program, the OS responds on the other.

Again, these are my subjective experiences. Cost-benefit is an entirely different matter, of course, and I'm not even touching on that here. And since I haven't used anything faster than my current machine, I am totally clueless as to what a fast, contemporary system would feel like.

Oh, about the Inquirer. I'm sure you know they neither have a lab to do benchmarking, nor the required technical expertise. I was referring to a benchmark they linked to. And no, I don't believe everything I read on the 'net, but this issue of Opterons scaling well has cropped up in many places, with AMD winning those deals with Cray etc. I guess time will tell how good they really are.

 

[Santilli]

Tyan vs. Supermicro...

Well, the discussion goes on.

I've lost my entire data, on array, and I'm not sure why.

I'm considering an upgrade, but can't make up my mind.

I currently have, in PCI slots, an atto UL3d, Adaptec AIC-7850, Texas insturments Firewire card, Philips PSC 706 audio card, an ATAPI cd rom drive, 56 max, Lacie Grou SA La Sie 1394 CDRW, Pioneer DVD rom-305, Plextor CD rom PX 40TS SCSI, and a burner, Plextor, I don't have connected.

Any ideas how I could set all this up on a mobo with current support for Xeon processors, either supermicro, or Tyan?

gs

 

[Mercutio]

That all looks very run-of-the-mill to me, Greg.

 

[Santilli]

Mercutio:

It is. The problem is migrating to the new standards, and most of the mobos don't have many pci slots, and, those that do, usually don't have that many 32 bit/33mhz slots.
Machine currently has 5 pci 32/33mhz pci slots.

Ethernet, and firewire, and scsi, could all be taken care of on the motherboard....
GS

 

[Santilli]

The other problem is the case I have is full. No more more drive bays.

Could go external firewire one of these days, or, I've been looking at removaable drives for backup.

Are the DVD RW format a viable solution for large storage?

The firewire writer I've been using is really cool, allowing a regular CD-R to be rewritten a number of times, for document backups. Does the DVD RW format work the same way, and is it fast enough?

Thanks

gs

 

[Fushigi]

Before my last upgrade I was using 5 PCI cards:

- SB Live
- SCSI
- Promise card for large ATA drive support
- NIC
- Modem

With the last upgrade I ditched the modem and, through mobo integration, got down to just:

- SCSI

I don't see why you wouldn't achieve similar results.

BTW, the new mobo was a budget FIC AU13 with 6 32/33 PCI slots + 8X AGP.

 

[Pradeep]

Are the DVD RW format a viable solution for large storage?

The firewire writer I've been using is really cool, allowing a regular CD-R to be rewritten a number of times, for document backups. Does the DVD RW format work the same way, and is it fast enough?

Thanks

gs

The +RW formats are up to 4x speed now, which means a write speed of a bit over 5MB/sec. That means a full 4.37 GB in about 15 mins. Sure beats swapping CD-R's

I guess you have come across the reliability factor of running four drives in RAID 0. So you are going to go the Xeon route?

 

[Santilli]

I'm wondering if my hard drive setup is the real culprit. I wonder if one of the cleaning utilities, or a bug, got in, and deleted a few essential files neccessary for 2000 to boot. It was a corrupted .cab file, IIRC.

Also, the floppie drive may not be functioning right, or that neccessary file has been deleted from my floppy install disks, since, after two attempts installing the os from floppy, the same cab file was missing, keeping the machine from booting.

I did install a 56X generic CD- ide reader, and, on one of the ide channels, it works, kind of slowly, in pio mode 3, but it works. I installed 2000 from it, and so far, no problems.

The other installs could have been defeated by either a poorly seated scsi card, or, a cheap scsi cable, now replaced with a granite digital, that I had with only one connector, reworked by GD to add their terminator, and connectors added.

One of the problems I had was there was only one scsi boot option, and, with blank hard drives, that was taken by the atto card. The 2906 has no boot designation in the bios, or, I believe, it's own software, either, but I could be wrong on the second.


Only time will tell.

I use a lot of stuff on the Asus motherboard, that really isn't designed to run on it, and, my stuff is more aimed at server quality stuff, rather then games, etc. except for my current video card.

My problem is space, or the need for another raid box.

If I take the 4 drives, put them inside the Supermicro chasis Gary is using, along with the 3 or 4 cd burners/readers, then I've used the avaliable 7 bays in the tower.

I would like to be able to use the SCA mini-backplane that holds 3-5 drives, depending upon format, in the tower.

SCA drives are cheaper, and would work just fine for swappable storage.
Plus, you could easily configure a raid with such a setup.

What I really need is to sit down with Hypermicro and go over their product line, and figure out what I want, and what I can afford, and what I can carry forward.

Then look at Tyan, and see if I could put the Tyan motherboard in the
Supermicro chassis, and if it would work as well.

The only thing that bothers me about Tyan was the initial release of the Tiger pretty much sucked.

That was a new chipset, as well.

I could also look at an Intel mobo, and, finally, I wonder if the generation of stuff after this current batch would be the way to go. Provided my existing rig functions, and lasts.

Budget wise, I would like to build a new computer this year, due to taxes, but, a digital camera for work might drain my tax money, and I do need one for work records, which is driving all this stuff.

gs

 

[Pradeep]

Actually the Tiger MP worked well, it was the initial Thunder K7 that had some issues. The follow up to the Tiger MP, the Tiger MPX in fact had more problems, such as non-functioning USB, piss poor 32bit PCI performance etc.

Dell SB had 400SC servers with 2.4GHz 800FSB P4, 40GB HDD, 128MB DDR for around $345 after $100 MIR. That gets you gigabit LAN and Serial ATA. I believe the MIR expired yesterday, but even without it's pretty good.

http://configure.us.dell.com/dellstore/config.aspx?c=us&cs=04&kc=70000&l=en&oc=s21003f

Given that you would end up paying prob around $200 for a decent SCA backplane enclosure, may be an option to consider.

 

[Santilli]

What does Prescott bring to the table that would make it worth waiting for?
Dual cores on the same chip?

Do Xeons have the same features?

gs

 

[Splash]

What does Prescott bring to the table that would make it worth waiting for?
Dual cores on the same chip?

Do Xeons have the same features?

The upcoming (early 2004) Xeons will definitely not be dual-core. There are dual-core Xrons coming, but that might be 2 or 3 years from now.

As for what you are looking for, Santilli, it's definitely not a Xeon, but an Opteron -- to be more exact, the Tyan K8W (S2885).

And, if you go with the Tyan K8W, you should save yer dollar$ and get just ONE 246 or (ga$p) 248 at this time and just 1 GB of certified DDR-400 RAM. Later, after the prices drop on the O-246 (or O-248), you can buy yer second processor and "go dual." You can then also buy another GB of DDR-400 RAM.

The writing's on the wall: The Opteron rules the workstation performance roost and probably will for a few years (at least). A Xeon might still be a slightly better pick for a server in some cases, at least until the Opteron has some run-time on it.

 

[Santilli]

"The writing's on the wall: The Opteron rules the workstation performance roost and probably will for a few years (at least)."

When did this happen? The prices on the processors you mentioned are WAY up there.

I found a couple shops around here that might want to sell older boards for Xeons. Read 200-300 for very good boards, and the Xeons aren't 750 dollars each.

Does windows run as well on the Opteron as it does on the xeons?

gs

 

[CougTek]

Does windows run as well on the Opteron as it does on the xeons?

No, they run better.

And from the benchmarks I saw, the Opteron are more dominant for server stuff than for workstation use compared to Xeon. Even for the workstation market, they are at least equal and often better, but for server applications, their lead over the Xeon is almost ridicoulous. There's an article about this at Ace's hardware.

 

[Santilli]

WOW!!! :eekers:
:bounce:

It's amazing how slow the clock speed is, yet how fast the results are with the Opteron.

Why?

64 bit? So I buy a mobo and single Opteron, at 1.4 ghz and it's as fast as a 3.0 ghz, Xeon???

gs

 

[Buck]

Boy, putting together a dual Opteron 240 system with the S2885 board and a couple gigs of Reg./ECC PC2700 memory really gets expensive. Add a really good power supply and nice case, and you easily hit $2,500.00.

 

[ddrueding]

Boy, putting together a dual Opteron 240 system with the S2885 board and a couple gigs of Reg./ECC PC2700 memory really gets expensive. Add a really good power supply and nice case, and you easily hit $2,500.00.

Yeah, but a single opteron system with a single gig of RAM can be had for $1500. That's not bad at all.

 

[blakerwry]

::sigh:: remember paying $3000 for a 66MHz 486SX system...

 

[Mercutio]

WOW!!! :eekers:
:bounce:

It's amazing how slow the clock speed is, yet how fast the results are with the Opteron.

Why?

gs

Well-optimized microcode, a goodly number of execution units and a pipeline that isn't 20 stages long.

Nothing to do with 64bit-ness.

To dramatically simplify, a pentium core is 2 486 cores (2 ALUs and an FPU) each with a (?)5 or 6 stage pipeline, a p55C is 2 ALUs, an FP unit and an MMX/FPU and so on. With the P4, instead of adding more raw processing capability - additional pipelines - more stages (10 more, in fact) were added to existing pipelines, so that less work is done for every tick of the clock. The best analogy I can think of is making the road from point A to point B twice as long so you can increase the speed limit 50%.