Seeing as how the new VIA Nano-ITX motherboards are tiny, and can be passively cooled, but only support up to 1GHz CPU's, would it be possible to use them in a parallel processor set up that would still keep things very quiet yet provide a higher performance level? For example, four of them wouln't take up much room and in combination with 2.5 HDD's and slim optical drives could still be in the SFF size category. I'm too much of a newbie to understand how it might be accomplished but am sure that out there in the SPCR community there are those who could create such a system. Any comments?
http://www.viaembedded.com/product/epia ... oardId=221
Parallel Processing with VIA Nano-ITX
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Yes and no. Yes, it should be possible to run them in parallel. Ramlow Design, the resellers of the Mappit claim to have done this, and I have no reason to doubt them.
However, software architechture often does not scale well. Let's assume that a 1GHz Via processor was equivalent to a 500MHz Celeron. You can't connect 4 of them together and expect it to act like a 2GHz Celeron (and thus play Quake 3 at normal speeds). Most software is not written to take advantage of multiple processors. About the best thing you CAN often do with this stuff is that you can run multiple programs and not take a performance hit. For example, on my Dual Athlon, I'm running 2 instances of Folding @ Home.
So, yes it can be done, but it won't be as good as having a single 2.0GHz processor.
However, software architechture often does not scale well. Let's assume that a 1GHz Via processor was equivalent to a 500MHz Celeron. You can't connect 4 of them together and expect it to act like a 2GHz Celeron (and thus play Quake 3 at normal speeds). Most software is not written to take advantage of multiple processors. About the best thing you CAN often do with this stuff is that you can run multiple programs and not take a performance hit. For example, on my Dual Athlon, I'm running 2 instances of Folding @ Home.
So, yes it can be done, but it won't be as good as having a single 2.0GHz processor.
Actually, having 4 Nano-ITX boards is having 4 separate computers, which is even more complicated than having a quad-CPU motherboard. At least with the quad-CPU motherboard, the memory, hard drives, networking, etc., are all shared resources by virtue of the fact that they are all the same motherboard. With 4 Nano-ITX computers, you would need to figure out a way to interconnect them all and have them work as one cohesive unit.
In general terms, what you are thinking of doing is called "clustering". Windows Server 2003 does have some provisions for clustering, and you can read about them here:
http://www.microsoft.com/windowsserver2 ... fault.mspx
A very well known implementation of clustering in Linux is known as "Beowulf". If you want to read more about that, I would suggest doing a Google search on "Linux Beowulf cluster". As is often the case with Linux, it is somewhat difficult for an outsider to figure out what they are talking about.
The types of applications that tend to run well on clusters tend to be those that are highly parallelizable (is that a word?
). For example, certain scientific calculations, animation projects, and sometimes databases. Ordinary single user applications tend to suffer, though, because of the overhead involved in connecting and managing the clustered computers.
In general terms, what you are thinking of doing is called "clustering". Windows Server 2003 does have some provisions for clustering, and you can read about them here:
http://www.microsoft.com/windowsserver2 ... fault.mspx
A very well known implementation of clustering in Linux is known as "Beowulf". If you want to read more about that, I would suggest doing a Google search on "Linux Beowulf cluster". As is often the case with Linux, it is somewhat difficult for an outsider to figure out what they are talking about.
The types of applications that tend to run well on clusters tend to be those that are highly parallelizable (is that a word?
). For example, certain scientific calculations, animation projects, and sometimes databases. Ordinary single user applications tend to suffer, though, because of the overhead involved in connecting and managing the clustered computers.mini-itx.com has an article on a cluster of C3 boards.
I am considering moving away from the one powerful PC to using clusters for video editing.After a bit of research it seems that Linux is the way to go for clusters.
http://www.linuxdevcenter.com/pub/a/lin ... lerra.html
However after a fortnight of using Fedora on one pc it does seem a bit overwhelming. There is loads of info about but sifting and making sense of it is another matter.
However I like the idea of having multiple silent stand alone computers that can be added of split as the task warrents as well as seperate hard drives in their own silent enclosures, ditto dvd writers and other periferals either linked with usb/firewire or wireless.
I am considering moving away from the one powerful PC to using clusters for video editing.After a bit of research it seems that Linux is the way to go for clusters.
http://www.linuxdevcenter.com/pub/a/lin ... lerra.html
However after a fortnight of using Fedora on one pc it does seem a bit overwhelming. There is loads of info about but sifting and making sense of it is another matter.
However I like the idea of having multiple silent stand alone computers that can be added of split as the task warrents as well as seperate hard drives in their own silent enclosures, ditto dvd writers and other periferals either linked with usb/firewire or wireless.
In addition of Beowulf there is also OpenMosix, which is a single system image cluster (ssi). This has the advantage that most software will not need to be modified to run on an ssi cluster. However as most programs consist of a single process they can only run on a single cpu at a time.alglove wrote: A very well known implementation of clustering in Linux is known as "Beowulf".
An ssi cluster still has the advantage that the processes can be migrated to nodes that are less heavily loaded.
Then there is the price question. 4 mini-itx boards with accesories will easily cost the same as an Athlon 64 system, while being restricted in their usefulness. I can only recommend this for the coolness factor
We have some more different experience with the VIA 1GHz machines. A single machine is adleast 8 times slower than a 2.4GHz P4 (running at 2.7GHz). When you do really heavy duty calcs with software that is actually optimised for Intel architecture for instance, then you can get up to a factor 20+. We here at Ramlow Design do majorly 3D animations and once we use something like Radiosity (indirect lighting) the above P4 thinks about a normal screen resolution image for about 3 hours. the VIA is still thinking after more than 60 hours. If you have VIA boards laying around in your backyard, then they are great for clustering, as they have no big power demands - our farm of 10 machines was running of the same power as the P4 !!! So if you are going to use something like 100 machines or more then you will have to consider to tap straight into the overhead powerline outside your home to supply a farm like that made up of P4s. This might darken the lights in your town a bit too ...jojo4u wrote:I think this has not really any use. Because 4x VIA 1Ghz is as fast as a Athlon 64 system and the power draw should be equal.
Anyway, we used that VIA renderfarm only for testing purposes for the manufacturer of the Mappit A4F to demonstrate that the A4Fs can be run for a long time at 100% CPU load without any trouble - with great success.
-
silvervarg
- Posts: 1283
- Joined: Wed Sep 03, 2003 1:35 am
- Location: Sweden, Linkoping
That was a very unfair test. The VIA processor is quite bad at floting point calculations since it is not designed for that kind of market. So to test it with 3D graphics calculations you really abuse the processor.We have some more different experience with the VIA 1GHz machines. A single machine is adleast 8 times slower than a 2.4GHz P4 (running at 2.7GHz). When you do really heavy duty calcs with software that is actually optimised for Intel architecture for instance, then you can get up to a factor 20+. We here at Ramlow Design do majorly 3D animations and once we use something like Radiosity (indirect lighting)
It is like taking a 4WD vehicle for a speed test on a formula 1 race track.
It will still run, but it is not the place that does it justice.
A fair test between AMD and Intel processor with 3D animations would be much more interesting. Perhaps including some SGI machine as well...
Well the main intend was to test the stability of the fanless Mappit A4F system under full load for extreme long times. Obviously an Intel is faster because it got the sets of SSE commands, which I believe the Via doesn't have and most high performance 3D packages an other heavy duty calculation software make use of those commands.silvervarg wrote:That was a very unfair test. The VIA processor is quite bad at floting point calculations since it is not designed for that kind of market. So to test it with 3D graphics calculations you really abuse the processor.
It is like taking a 4WD vehicle for a speed test on a formula 1 race track.
It will still run, but it is not the place that does it justice.
A fair test between AMD and Intel processor with 3D animations would be much more interesting. Perhaps including some SGI machine as well...
We still use some of the A4Fs for normal rendering and processing - for the more heavier things we try to avoid using them and run it on P4s.
And our kind of use is not the only use. I believe the VIA processors are great for internet servers or in a database cluster - what matters is the low running cost and the high reliability.
ROFL!!!So if you are going to use something like 100 machines or more then you will have to consider to tap straight into the overhead powerline outside your home to supply a farm like that made up of P4s. This might darken the lights in your town a bit too ...
But you have hit the point there:
...since i using VIA machine (1200Mhz Nehemiah one) fanlessly and as web server, testing machine and downloading machine - and so onI believe the VIA processors are great for internet servers or in a database cluster - what matters is the low running cost and the high reliability.
My uptime is month (now must be restarted, becuase we installed Bluetoth USB adapter, just to discover the range are about 3m (9 feets), witch suxx) and it don't seems to have any problem at all.The speed is a question for itself - there are certain SIMPLE tasks, like copying file over network and so on, where the 1200Mhz VIA smoke 900Mhz CeleronII used before in same machine!
Thats sweet.
Most simple operations outperform similary clocked intel/amd cpus.
BUT!
As soon, as you get into any calculations, the speed is very very slow. I tried to run Prime95, wich working acceptable fast (to my surprise), but folding is very very very slow... Math is simply not for these cpus at all, and 3D is mostly about math
Anyway, you are wrong about the lack of SSE - Nehemiah core have it. Enzo-T (or what the older one is...) don't, but have 3DNow. I still suspect that the difference between the cores are just the microcode, but...
Who cares as long it working well and mainly fanless and fast for simple operations - and that is for what i need my server.
So, making a cluster out of these lack practical purpose, IMHO.
Not to mention that there are certain folks on the VIA arena forum, witch claim that the parallel operation is hard to realize - and mostly not working at all - well, that's what Im hear there
Dunno if it get fixed, or someone come with mobo, on what it working? It sill need different OS that windows and best recompilled applications to take more advantage of dual cpus.
So if anyone want fast and powerfull machine, stick please with one CPU and fanles watercooling, like mine
