Yes I've seen that mini-itx cluster before...
I think I would make it the same way.
My Router is mini-itx, all other Boards with VIA CPUs are only micro-atx boards.
They are not VIA Boards, but the VIA CPUs should make it power-saving enough.
I think I would start with a 10 motherboard cluster, to see if it works for my needs... I consider if 10 motherboards can be provided with one bigger ATX PSU, parallel connected to each motherboard.
> I think I would start with a 10 motherboard cluster, to see if it works for my
> needs... I consider if 10 motherboards can be provided with one bigger ATX
> PSU, parallel connected to each motherboard.
>
Why?
Maybe you'll tell me I am missing the point, but it seems to me that the strength of the VIA boards is the low power consumption for fixed-point / text / network applications, none of which use floating point at all.
Unless a cluster of ten uses *less* power than a single Intel/AMD motherboard there seems no point doing a ten cluster if it is *just* to run a floating point project on about the same power. Why not a single Intel/AMD motherboard?
Of course, if you need a ten-cluster for some other task and are simply looking for a productive way to use the unwanted cycles, then can see the point. Certainly a ten cluster would fly like the wind when its doing the sort of application those chips are designed for.
Please don't take this as criticism -- I am simply curious because I don't see the attraction myself.
I've been thinking about this VIA boards too, but like gravywavy said it makes no sense.
In this project you can't split the calculations per WU over the ten nodes so you basicly end up with 10 single machines (getting ten WU's) of which FPU power is low and crunching time very long. see stats on this machines.
If you run them from a big PSU, you can also run two/three AMD boards from it providid leaving as much hardware that is not required out. You can get those semprons 2400+ very cheap, complete with memory and fan for the price of one VIA board.
In another thread (maybe on another board for all I remember) there is the possibility that the Pentium M may be the "best" for electric power vs. processing speed.
For raw procesing speed alone I would have to give the nod to the Opteron, G5 or Xeon systems ... with, right now, Xeon being the king on THROUGHPUT ...
Mostly because the Xeon, in dual processor systems, allows 4 WU to be in-flight at a time.
Sorry i was in egypt for a couple of weeks. Had a great time !!without!! any computer in front of me ;)
Well as i am back i am happy to announce the expansion of the "VIA-C3 slow motion" Team. Tim Lindner joined during my absence und together we breached the 1000 credits border ;)
Welcome Tim!
Ich habe selbst nur 2 VIA 1Ghz hier zur Heimvernetzung hier rumstehen, von denen einer gerade wegen Hardwaredefekt ausgeschieden ist. Natürlich wieder in meiner Abwesenheit :-/ Wenn Du jetzt 100 dieser süssen Boards zur Verfügung hast, dann kannste natürlich einen fetten Cluster mit bauen. Aber was willst Du damit anfangen? Ich denke eine Leistungsaufnahme von ca. 2KW hätteste dann schon.
Mostly because the Xeon, in dual processor systems, allows 4 WU to be in-flight at a time.
==========
Actually you can run 4 WU's at the same time with just 1 P4 HT Processor. I discovered how to do it some time ago. The Task Manager even shows all 4 running at 25% CPU use each ... ;)
> Actually you can run 4 WU's at the same time with just 1 P4 HT Processor. I
> discovered how to do it some time ago. The Task Manager even shows all 4
> running at 25% CPU use each ... ;)
how does total throughput compare? I mean how long does it take to finish 4 WU all running together compared with running one at a time one after another?
about the same or slightly slower would be my guess, without having tried it
>
> > Actually you can run 4 WU's at the same time with just 1 P4 HT Processor.
> I
> > discovered how to do it some time ago. The Task Manager even shows all 4
> > running at 25% CPU use each ... ;)
>
> how does total throughput compare? I mean how long does it take to finish 4
> WU all running together compared with running one at a time one after
> another?
>
> about the same or slightly slower would be my guess, without having tried it
===========
I never ran 4 WU's at a time long enough to really see what the difference was, all I did was run a few Pirate WU's that way to see if it would finish them.
At first I thought I was on to something because of the Time the BOINC MGR was giving me for completing them, according to the times I was running 4 as fast as I could run 2 ... But I kept thinking something wasn't right here so I got my Stop Watch out and started up 4 Pirate WU's.
Well for every 20 seconds of Time the BOINC MGR claimed I ran 1 Minute would elapse on 1 set of WU's. The other set of WU's was fairly close though on the time the BOINC MGR Claimed I ran...I want to try it again and get some exact figures for running the completed WU's.
I figured the last time there was no advantage of running 4 verses 2 at a time in that it was taking twice as long to run 4 as compared to running 2 then 2 more ... But like I said I want to get some more figures ...
Don't think it's the same CPU
)
Don't think it's the same CPU as started this thread, but look at what you can do with the these low power chips.
Yes I've seen that mini-itx
)
Yes I've seen that mini-itx cluster before...
I think I would make it the same way.
My Router is mini-itx, all other Boards with VIA CPUs are only micro-atx boards.
They are not VIA Boards, but the VIA CPUs should make it power-saving enough.
I think I would start with a 10 motherboard cluster, to see if it works for my needs... I consider if 10 motherboards can be provided with one bigger ATX PSU, parallel connected to each motherboard.
> I think I would start with
)
> I think I would start with a 10 motherboard cluster, to see if it works for my
> needs... I consider if 10 motherboards can be provided with one bigger ATX
> PSU, parallel connected to each motherboard.
>
Why?
Maybe you'll tell me I am missing the point, but it seems to me that the strength of the VIA boards is the low power consumption for fixed-point / text / network applications, none of which use floating point at all.
Unless a cluster of ten uses *less* power than a single Intel/AMD motherboard there seems no point doing a ten cluster if it is *just* to run a floating point project on about the same power. Why not a single Intel/AMD motherboard?
Of course, if you need a ten-cluster for some other task and are simply looking for a productive way to use the unwanted cycles, then can see the point. Certainly a ten cluster would fly like the wind when its doing the sort of application those chips are designed for.
Please don't take this as criticism -- I am simply curious because I don't see the attraction myself.
~~gravywavy
I've been thinking about this
)
I've been thinking about this VIA boards too, but like gravywavy said it makes no sense.
In this project you can't split the calculations per WU over the ten nodes so you basicly end up with 10 single machines (getting ten WU's) of which FPU power is low and crunching time very long. see stats on this machines.
If you run them from a big PSU, you can also run two/three AMD boards from it providid leaving as much hardware that is not required out. You can get those semprons 2400+ very cheap, complete with memory and fan for the price of one VIA board.
It will then crunch much more WU's for the buck.
John,
In another thread (maybe on
)
In another thread (maybe on another board for all I remember) there is the possibility that the Pentium M may be the "best" for electric power vs. processing speed.
For raw procesing speed alone I would have to give the nod to the Opteron, G5 or Xeon systems ... with, right now, Xeon being the king on THROUGHPUT ...
Mostly because the Xeon, in dual processor systems, allows 4 WU to be in-flight at a time.
Hmmm, I'll have to host up a
)
Hmmm, I'll have to host up a pic of my son's VIA C3 (M10000) based Nintendo Entertainment System (NES) PC.
Sorry i was in egypt for a
)
Sorry i was in egypt for a couple of weeks. Had a great time !!without!! any computer in front of me ;)
Well as i am back i am happy to announce the expansion of the "VIA-C3 slow motion" Team. Tim Lindner joined during my absence und together we breached the 1000 credits border ;)
Welcome Tim!
Ich habe selbst nur 2 VIA 1Ghz hier zur Heimvernetzung hier rumstehen, von denen einer gerade wegen Hardwaredefekt ausgeschieden ist. Natürlich wieder in meiner Abwesenheit :-/ Wenn Du jetzt 100 dieser süssen Boards zur Verfügung hast, dann kannste natürlich einen fetten Cluster mit bauen. Aber was willst Du damit anfangen? Ich denke eine Leistungsaufnahme von ca. 2KW hätteste dann schon.
Gruß Said
Mostly because the Xeon, in
)
Mostly because the Xeon, in dual processor systems, allows 4 WU to be in-flight at a time.
==========
Actually you can run 4 WU's at the same time with just 1 P4 HT Processor. I discovered how to do it some time ago. The Task Manager even shows all 4 running at 25% CPU use each ... ;)
> Actually you can run 4 WU's
)
> Actually you can run 4 WU's at the same time with just 1 P4 HT Processor. I
> discovered how to do it some time ago. The Task Manager even shows all 4
> running at 25% CPU use each ... ;)
how does total throughput compare? I mean how long does it take to finish 4 WU all running together compared with running one at a time one after another?
about the same or slightly slower would be my guess, without having tried it
~~gravywavy
> > > Actually you can run 4
)
>
> > Actually you can run 4 WU's at the same time with just 1 P4 HT Processor.
> I
> > discovered how to do it some time ago. The Task Manager even shows all 4
> > running at 25% CPU use each ... ;)
>
> how does total throughput compare? I mean how long does it take to finish 4
> WU all running together compared with running one at a time one after
> another?
>
> about the same or slightly slower would be my guess, without having tried it
===========
I never ran 4 WU's at a time long enough to really see what the difference was, all I did was run a few Pirate WU's that way to see if it would finish them.
At first I thought I was on to something because of the Time the BOINC MGR was giving me for completing them, according to the times I was running 4 as fast as I could run 2 ... But I kept thinking something wasn't right here so I got my Stop Watch out and started up 4 Pirate WU's.
Well for every 20 seconds of Time the BOINC MGR claimed I ran 1 Minute would elapse on 1 set of WU's. The other set of WU's was fairly close though on the time the BOINC MGR Claimed I ran...I want to try it again and get some exact figures for running the completed WU's.
I figured the last time there was no advantage of running 4 verses 2 at a time in that it was taking twice as long to run 4 as compared to running 2 then 2 more ... But like I said I want to get some more figures ...