• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Iwill Dual-Opteron SFF approaches

SUOrangeman

Diamond Member
2CPU article

Am I supposed to say something about BBQ and OMG? 🙂

A vendor on the forums there just received his ZMAXdp.

By the way, a dual nForce4 mobo would be alot of fun as well!

-SUO
 
Rather then start a new thread, I will just loft this one back into orbit. SUO, I did a search on the forums for zmaxdp, and you poste din pretty much every thread that came up. I am just looking for more details on a release date or price. If you get any other info, post or PM me. Thanks
 
I do not understand the point of purchasing a dual Opteron machine that is castrated by manufacturer to have memory only on one CPU. Having separate memory banks for different CPUs happens to be one of Opteron's biggest advantages.
 
Maybe because it still is a screamer, and dual memory bank boards are a lot more expensive, not to mention they won't fit in a small form factor. Some users dont really need the benefit of having each CPU have its own memory to access. I mean I want it for multitasking goodness, and help with encoding and compiling(if I ever try gentoo again). I'm not gonna be doing anything to intensive.



Originally posted by: Burbot
I do not understand the point of purchasing a dual Opteron machine that is castrated by manufacturer to have memory only on one CPU. Having separate memory banks for different CPUs happens to be one of Opteron's biggest advantages.

 
Maybe because it still is a screamer, and dual memory bank boards are a lot more expensive, not to mention they won't fit in a small form factor. Some users dont really need the benefit of having each CPU have its own memory to access. I mean I want it for multitasking goodness, and help with encoding and compiling(if I ever try gentoo again). I'm not gonna be doing anything to intensive.
The only people that will spend gobs of cash on a dual processor workstation are those who need performance or memory capacity. Thus, a dual processor workstation manufacturer should make sure their machine *provides* performance and memory capacity. Users that do not really do anything too intensive are already well served by a cheap Sempron/Celeron machine and thus are not a part of target market.
 
Burbot, I agree. By hanging all the memory off one CPU, you've defeated the biggest performance advantage the Opteron offers (ie. the onboard memory controller). HT is fast, yes, but forcing one CPU to perform all its memory accesses over HT is a shame. Iwill isn't the only manufacturer doing this, either.
 
Where exactly do you see the added benefit of the dual memory banks? According to this review at gampc. They reviewed 2 dual opteron boards, one single bank memory, the other with dual banks. There was almost no differance in performance between the 2. So what applicatiosn really take advantage of the extra bank of memory. For someone who wants to use the box as a media server, or divx encoder, or who just wants to run 20 apps at a time, its not gonna make a huge differance. ANd yes there are a lot of people out there that spend the cash on a machine like this, just for those tasks.


I mean i could be totally wrong, if I am please show me the benches that prove otherwise. I would like to know for sure. I am just going off what I have seen and know, and with that you senario doesn't hold up.



I mean sure the guys at NASA and the peopl that will string together 50 of these to make a mini super computer may have a need for the extra bank of memory. But the way i see it, your average SMP power user doesn't,




Originally posted by: Burbot
Maybe because it still is a screamer, and dual memory bank boards are a lot more expensive, not to mention they won't fit in a small form factor. Some users dont really need the benefit of having each CPU have its own memory to access. I mean I want it for multitasking goodness, and help with encoding and compiling(if I ever try gentoo again). I'm not gonna be doing anything to intensive.
The only people that will spend gobs of cash on a dual processor workstation are those who need performance or memory capacity. Thus, a dual processor workstation manufacturer should make sure their machine *provides* performance and memory capacity. Users that do not really do anything too intensive are already well served by a cheap Sempron/Celeron machine and thus are not a part of target market.

 
It's interesting that gamepc published another article where they compared NUMA vs non-NUMA performance and when in NUMA configuration (ie. each CPU has its own memory), the memory bandwidth went through the roof. See for yourself.

This suggests that something was wrong with their NUMA configuration for the October, 2004 article. In fact, their benchmarks suggest that perhaps they weren't running NUMA at all.


 
Granted, the setup will be limited by only having two memory slots, but with the size constraints I don't see any way they could have implemented seperate memory banks for each CPU. I used to have an Opteron dually based on the MSI board (which had a similar shared-memory setup) and I sure didn't notice any performance hits. That system absolutely flew!

If I had the fundage, this Iwill SFF would just about perfect for me, especially paired with a couple of the lower-wattage (55w) HE 246's. Just about all the benefits of a dually (assuming you don't need PCI-X slots for SCSI cards and such) in a small, convenient package.
 
Uh, in a SMP system all CPUs use the same RAM.

You can optimize access with more busses somewhat but this is a pretty fundamental thing and not much you can do about it. You can't just start having the CPUs operate in different memory without massive support from the OS.
 
coolred
I share opinion of arcas about some kind of configuration screwup. Let's assume that Sandra memory bandwidth scores actually mean something useful. Then NUMA configuration, by virtue of having 2x theoretical memory bandwidth, is going to give you 2x score. This effect can be seen on gamepc.com newer test, but is curiously absent from October 2003 one. Thus, it looks like during Oct 2003 test they were putting all memory into one bank even on K8W, and all advantages of NUMA did not really show up because they *did not use it*.

For someone who wants to use the box as a media server, or divx encoder, or who just wants to run 20 apps at a time, its not gonna make a huge differance.
Running a lot of "generic not too heavy" apps at a time requires sufficient memory and moderately reasonable CPU. Dual processor is definitely an overkill here.
DivX encoding? How many DivX movies do you have to encode per day to make dual CPU machine purchase reasonable?
Media server? I do not think you need SMP in those, either.

You try to make it sound like this machine is for normal folks who want a fast machine for general purpose computing. It's not. Normal folks will have a heart attack and die under the table when they see Opteron (or, to be fair, Xeon) pricing. Home-caliber power users are probably going to feel just fine with single-processor A64 barebones. Worstation users will not buy it, since it does not provide things that workstation users need (such as large disk and memory capacity). So who would get it? Somebody who has lots of cash to burn and wants to get an SMP machine to be cool?
 
Either way, benchmarks are all nice an pretty looking, but show me where this extra bandwidth shows a real world performance gain.


Originally posted by: arcas
It's interesting that gamepc published another article where they compared NUMA vs non-NUMA performance and when in NUMA configuration (ie. each CPU has its own memory), the memory bandwidth went through the roof. See for yourself.

This suggests that something was wrong with their NUMA configuration for the October, 2004 article. In fact, their benchmarks suggest that perhaps they weren't running NUMA at all.

 
Thats not entirely true. When running any 2 applications, they are gonna run faster on an SMP rig. Granted running to word processors isn't gonna show any real world performance. The media server or lan gamer can benefit from dual CPU's as well. Probablly more so in the case of the LAN gamer. Say they are playing a game utilizing most if not all of thier CPU. And they also have 4-5 people copying files off of thier computer. This could also require CPU time, thus taking that time away from the game. With dual CPU's game performance would not suffer due to file swapping.

I will concede that maybe they were not using a NUMA OS during the review I posted, I have no idea. But either way, benechmarks do not show real world performance. I want to see where exactly this added memory bandwidth shows a real performance gain.

I am not saying this machien is for normal folks. It is for many things, but not normal folks. Obviously it is small, so it would make a nice server cluster. But it also looks to be aimed squarely at the LAN gamer. And there are more then a few people on the forums who use dual processors rigs as their daily rigs, and love the performance they get from them. So it may not be for normal folks, but power users can definately see the advantage of dual processors.

Especially those with dual monitors, who like to multitask, they will definately see a benefit from dual processors as they tend to have more stuff going on all at once then the normal user would.

Originally posted by: Burbot

coolred
I share opinion of arcas about some kind of configuration screwup. Let's assume that Sandra memory bandwidth scores actually mean something useful. Then NUMA configuration, by virtue of having 2x theoretical memory bandwidth, is going to give you 2x score. This effect can be seen on gamepc.com newer test, but is curiously absent from October 2003 one. Thus, it looks like during Oct 2003 test they were putting all memory into one bank even on K8W, and all advantages of NUMA did not really show up because they *did not use it*.

For someone who wants to use the box as a media server, or divx encoder, or who just wants to run 20 apps at a time, its not gonna make a huge differance.
Running a lot of "generic not too heavy" apps at a time requires sufficient memory and moderately reasonable CPU. Dual processor is definitely an overkill here.
DivX encoding? How many DivX movies do you have to encode per day to make dual CPU machine purchase reasonable?
Media server? I do not think you need SMP in those, either.

You try to make it sound like this machine is for normal folks who want a fast machine for general purpose computing. It's not. Normal folks will have a heart attack and die under the table when they see Opteron (or, to be fair, Xeon) pricing. Home-caliber power users are probably going to feel just fine with single-processor A64 barebones. Worstation users will not buy it, since it does not provide things that workstation users need (such as large disk and memory capacity). So who would get it? Somebody who has lots of cash to burn and wants to get an SMP machine to be cool?

 
What did you use your system for?

Originally posted by: Mloot
Granted, the setup will be limited by only having two memory slots, but with the size constraints I don't see any way they could have implemented seperate memory banks for each CPU. I used to have an Opteron dually based on the MSI board (which had a similar shared-memory setup) and I sure didn't notice any performance hits. That system absolutely flew!

If I had the fundage, this Iwill SFF would just about perfect for me, especially paired with a couple of the lower-wattage (55w) HE 246's. Just about all the benefits of a dually (assuming you don't need PCI-X slots for SCSI cards and such) in a small, convenient package.

 
Just general purpose stuff. Internet, gaming, DC. It was my daily rig, as you put it. I used to have several applications going at once, and I noticed a real difference when I had to step down to my current P4C system.
 
Thats all I need to hear. Like I said, it doesn't take a NASA scientist to notice the added benefit of dual processors.
 
The Inq has some news about these machines, starting mass production now. They also say price expects to be well over 500.00 possiblly as high as 700.00. Thats more then i was expecting. I was think 500.00 max. So looks like I will be passing on this.

Just thought I would let you know.
 
Monarch Computer has these listed at $768 and arriving in January. Most of my chatter has been in a thread over at 2CPU.com.

-SUO
 
Originally posted by: coolred
Thats all I need to hear. Like I said, it doesn't take a NASA scientist to notice the added benefit of dual processors.

I have a dually as my daily rig!! 🙂 2 xeons and love it, but I also do a lot of design work on it as well, and some 3d animation stuff, so the dually holds it own!
 
Good to know, I got most of the parts for my dually rig, went for the cheaper Xeons as well, as opposed to the opterons.
 
Back
Top