Bruce, why the hell aren't you doing this???? :-).
Yeah they could call it the Pulsar Core - or Pulsar Core Duo for a binary system discovery. But I guess Bruce did his PhD in physics, not in marketing - and more power to him! :-)
Alternatively, he may well do all of that - but for GW detections ..... yeah, 'The Gravitar' or 'The Magnetar'. Jet black case, red(shifted) LED's, large diameter fans. Quite invisible on your desktop, but the pens, pencils and stapler will slide toward it. Cat on the keyboard becomes a cat in orbit ......
maybe something more low key perhaps, like "I found a new pulsar - have you?" printed on a tee shirt front - with the discovery-stats/detection/confirmation plots on the back. Plus 'Intel Inside' logo of course. Personally signed, validated, certificated etc. Top notch nerd cred value! :-)
Cheers, Mike.
( edit ) Why stop there ? Get Nvidia, Kingston, Corsair, Gigabyte, Asus, Western Digital, Seagate, Netgear, Dell, even LG ..... :-)
( edit ) Maybe M$ ?
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Just looking at your host, with a same-generation CPU with half the cores, and with likely lower overhead from non-CPU parts, I think your power measurements may well turn out correct. (I've had four meters of this general class, three models from two makers--all have matched to within a watt or two at the 100W level, and have passed a sanity check using an incandescent light bulb).
I tried a second meter and it gives exactly the same results. I had always imagined that all the components on the mobo would probably burn as much power as the CPU and if the CPU was rated @ 65W, surely the full draw would be close to double that. Then considering the extra for the HDD and CDROM and the PSU efficiency, surely it would be drawing at least 150W from the wall. That's why I was so surprised it was only 80W @ 100% CPU load.
The E3200 @ 3.2GHz has continued to draw around 80W. Today it was flicking between 78-79 everytime I observed it. The task times are reasonably repeatable now with very little variation and averaging 4.29hrs as opposed to 4.41 for the previous ones. The app version for all these tasks crunched under Linux is 1.09.
So I decided to do a comparison with running the 3.11 app version under Windows XPP under identical conditions. The power draw now is flicking between 80-81 and the first couple of results have taken 4.75 hours. I'll need to wait for more to be sure but using the same performance metric as previously it looks like I'll be getting 2x250/(81x4.75) = 1.30 credits/watthour, as opposed to 1.417 under Linux. My experience has always been that the Linux app is probably around 5-10% faster than the Windows app and it looks like it may use a smidgen less power as well. So all these measurements were on the same hardware, same BIOS settings and same (1st) power meter.
I setup an E6300 Pentium dual core (45nm Wolfdale) on the second power meter. It had been running solidly for a couple of months and was upgraded exactly as described for the E3200. So the only difference was the CPU chip and the fact that it had been run in @100% continuous load for a couple of months. It had been running @ 10.5x344 = 3.612GHz and various voltage options had been increased to ensure stability. The APB1 app version is again 1.09.
The first thing I did after setting it up on the power meter was to set all BIOS options for voltage back to default and see if it would run properly. It booted up fine and I stopped BOINC to see what the idle power draw was. It sat rock steady on 53 watts. I then re-started BOINC and allowed things to run for a few minutes. The watts flicked between 81-82. It was still running after about 20 minutes with the same wattage. After about an hour it had locked up so I rebooted and set the Vcore only +100mV. There are also voltage options for RAM, MCH chipset, ICH chipset but I left those at default. It has now run for quite a few hours and results have finished and validated and the wattage has been flicking between 88-89. I will allow this to continue running for a few days and monitor the results. The history of results shows very consistent times of 3.93 hrs and the new ones just finished are very much the same. So the calculated performance is 2x250/(89x3.93) = 1.43 credits/watthour. This is pretty much the same figure that I got for the E3200. It looks like I would improve it a bit if I lowered the overclock until it would run with stability at default Vcore. It might be worth seeing if a drop from 3.6GHz to say 3.5GHz would achieve that. I kinda like 3.6GHz though :-).
All in all the early numbers make me quite happy with the humble E3200 as a budget choice. However it's very early days yet so I'll keep playing and reporting.
Just to insert a coin of my 5 cents into discussion. So, guys, why not to try new MLC-SDDs in these machines? I will soon get a couple of them. But they will work in file server, so I will not be able to measure the power draw, but I can make a picture of them to see their parameters. We here do not need a huge disk space for BOINC only machines, so the lowest capacity SSD drive (just 4 Gbs like in ASUS EEE PC 901) I think will be enough. If someone says that they are so expensive these days than I can offer another solution - COMPACT FLASH card with EIDE-adapter :)
P.S. Yesterday I've seen guys who are crunching E@H on barebone system on Intel Atom N270, altered for use in car instead of regular music box and using free WiFi-spots for the internet where available.
So, guys, why not to try new MLC-SDDs in these machines?
While I read that there can be power savings it is far from automatic depending on model. I happen to use a professional sound recorder for which recently solid-state drives have been offered as alternatives to hard drives. In that specific application there was no power saving.
On the other hand a friend of mine who is an Intel employee loaned me one of their well regarded drives about a year ago, and based on the temperature of the housing after it'd been in my computer for a week, and also readings on my system-level power meter, I think it used several watts less than my usual hard drive choices. So I certainly agree it is possible. It should not be assumed.
I have not seen pricing on solid-state drives attractive for economy cruncher use. Can you point to some?
A well-made SSD would likely also be a system level reliability upgrade.
So, guys, why not to try new MLC-SDDs in these machines?
I have not seen pricing on solid-state drives attractive for economy cruncher use. Can you point to some?
A well-made SSD would likely also be a system level reliability upgrade.
I've found something in Kingston's site.
I've seen these drives here (South Urals, Russia) for the prices about $200 USD depending on the size and manufacturer. So, it's quite useful instead of low-power 1TB drive if the capacity is not an argument.
I've seen these drives here (South Urals, Russia) for the prices about $200 USD depending on the size and manufacturer. So, it's quite useful instead of low-power 1TB drive if the capacity is not an argument.
Right now, for new drives at comparable shopping carefulness, here in the US an 80 or 160 Gbyte HD is about $40, while a 32 or 40 Gbyte SSD is about $100. I doubt there is enough power consumption savings to pay that back on almost anyone's billing rate.
What you say should be possible, as the clearest natural SSD advantages over HD are price at the low capacity end, and speed in the sense of latency (no rotational delay). But the SSD makers don't see a prompt road to profit in making, say 8 Gbyte drives for $10 right at the beginning. When you are only making millions, rather than hundreds of millions of the things, you don't get those nice low costs right at the beginning, so they are looking for places where they can offer a compelling enough advantage to earn a huge price premium instead. Right now that is folks who prize the speed advantage, or possibly reliability or power consumption advantages. With luck, perhaps they will pursue the low end later.
The question of SSD drives is unlikely to even be on my radar any time soon. I have plenty of existing drives of various types and sizes and, more importantly, plenty of a particular model that seems very robust and very suited to the job.
Over the years of playing with surplus equipment, you get to form an opinion about which models are good and which are terrible. For HDDs I would place the Maxtor 541DX 20GB slimline drive in the terrible category. When first acquired, many had bad sectors and those that seemed OK initially, failed with monotonous regularity.
When I purchased the first of what would turn out to be my tualatin celeron 1300 workhorse, I noted that they also featured a slimline drive - the seagate ST320014A - which (apart from the label) looked very similar to the Maxtors. I was initially sceptical of how they would perform (being a slimline drive) but my experience has been nothing but good. Even though they date back to the 2002/03 era, I can't recall a single failure in service. They run cool to the touch whereas the Maxtors ran quite hot. I'm very happy to keep using them in all my upgraded machines. Now that I've got the ability to measure it, I guess I'll get around to measuring how much they actually draw in normal running. It can't be very much and must be less than the Maxtor on the basis of operating temperature alone.
Once I work out how much the hard disk draws, I may well consider doing away with them altogether and booting from a live CD with the BOINC stuff hosted on a network share. When added up over all the crunching machines, that might represent a worthwhile saving in power.
An HD shouldn't draw much. IIRC desktop drives draw about 10W when spinning and 1W at idle. Laptop drives are significantly lower.
Quote:
I may well consider doing away with them altogether and booting from a live CD with the BOINC stuff hosted on a network share. When added up over all the crunching machines, that might represent a worthwhile saving in power.
Gary has his hosts hidden, but given his high RAC, recent use of antiques, and penchant for economy builds, I'd guess it is quite a few. Also, I think recall that BOINC writes to a state file rather frequently, so depending on how long one's OS preferences allow a disk to wait before going down to idle, it may well be that the drive on which the state file lives never idles. 10 watts times a lot of hosts just might add up in a year. Even if it is 5 watts. And given the drives are sensibly warm to the touch (by and large), I'll wager the real time average is not likely much less than 5 watts.
I can't address the relative value of Gary's time to do the network drive idea versus his local cost of power, but over his large empire, I suspect the savings opportunity is considerable. For the rest of us, not so much.
I'll wager the real time average is not likely much less than 5 watts.
5 watts is pretty much what I would expect for a 'just warm' drive. As some drives run rather hotter than 'just warm' perhaps some get closer to 10 watts.
Quote:
I can't address the relative value of Gary's time to do the network drive idea versus his local cost of power, but over his large empire, I suspect the savings opportunity is considerable. For the rest of us, not so much.
At the peak, I had about 240 machines running. Today I have around 55. Of those about 7 are 'oldies' that will soon either be shut down or upgraded. Then I should be finished for a while.
Electricity costs me approx $USD0.15 per Kilowatthour. If I could save 5 watts on 50 machines, my annual saving would be 5x24x365x50x0.15/1000 = $USD328. Since electricity will rise steeply in coming years, it may be worthwhile investigating further. The first step is to actually measure the watts per drive.
It would actually be a bit tedious to convert 50 machines like this. However with Linux, I can set up a single machine with everything installed and configured and all scripts in place. Then I can remaster the machine to make a new live CD of the working installation. The conversion of any single machine should then be as simple as stopping BOINC, moving the complete BOINC installation to the appropriate place on the network drive, disconnecting the hard drive and then booting up from the new live CD which will know where to look for the BOINC stuff on the network.
RE: Bruce, why the hell
)
Yeah they could call it the Pulsar Core - or Pulsar Core Duo for a binary system discovery. But I guess Bruce did his PhD in physics, not in marketing - and more power to him! :-)
Alternatively, he may well do all of that - but for GW detections ..... yeah, 'The Gravitar' or 'The Magnetar'. Jet black case, red(shifted) LED's, large diameter fans. Quite invisible on your desktop, but the pens, pencils and stapler will slide toward it. Cat on the keyboard becomes a cat in orbit ......
maybe something more low key perhaps, like "I found a new pulsar - have you?" printed on a tee shirt front - with the discovery-stats/detection/confirmation plots on the back. Plus 'Intel Inside' logo of course. Personally signed, validated, certificated etc. Top notch nerd cred value! :-)
Cheers, Mike.
( edit ) Why stop there ? Get Nvidia, Kingston, Corsair, Gigabyte, Asus, Western Digital, Seagate, Netgear, Dell, even LG ..... :-)
( edit ) Maybe M$ ?
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Just looking at your
)
I tried a second meter and it gives exactly the same results. I had always imagined that all the components on the mobo would probably burn as much power as the CPU and if the CPU was rated @ 65W, surely the full draw would be close to double that. Then considering the extra for the HDD and CDROM and the PSU efficiency, surely it would be drawing at least 150W from the wall. That's why I was so surprised it was only 80W @ 100% CPU load.
The E3200 @ 3.2GHz has continued to draw around 80W. Today it was flicking between 78-79 everytime I observed it. The task times are reasonably repeatable now with very little variation and averaging 4.29hrs as opposed to 4.41 for the previous ones. The app version for all these tasks crunched under Linux is 1.09.
So I decided to do a comparison with running the 3.11 app version under Windows XPP under identical conditions. The power draw now is flicking between 80-81 and the first couple of results have taken 4.75 hours. I'll need to wait for more to be sure but using the same performance metric as previously it looks like I'll be getting 2x250/(81x4.75) = 1.30 credits/watthour, as opposed to 1.417 under Linux. My experience has always been that the Linux app is probably around 5-10% faster than the Windows app and it looks like it may use a smidgen less power as well. So all these measurements were on the same hardware, same BIOS settings and same (1st) power meter.
I setup an E6300 Pentium dual core (45nm Wolfdale) on the second power meter. It had been running solidly for a couple of months and was upgraded exactly as described for the E3200. So the only difference was the CPU chip and the fact that it had been run in @100% continuous load for a couple of months. It had been running @ 10.5x344 = 3.612GHz and various voltage options had been increased to ensure stability. The APB1 app version is again 1.09.
The first thing I did after setting it up on the power meter was to set all BIOS options for voltage back to default and see if it would run properly. It booted up fine and I stopped BOINC to see what the idle power draw was. It sat rock steady on 53 watts. I then re-started BOINC and allowed things to run for a few minutes. The watts flicked between 81-82. It was still running after about 20 minutes with the same wattage. After about an hour it had locked up so I rebooted and set the Vcore only +100mV. There are also voltage options for RAM, MCH chipset, ICH chipset but I left those at default. It has now run for quite a few hours and results have finished and validated and the wattage has been flicking between 88-89. I will allow this to continue running for a few days and monitor the results. The history of results shows very consistent times of 3.93 hrs and the new ones just finished are very much the same. So the calculated performance is 2x250/(89x3.93) = 1.43 credits/watthour. This is pretty much the same figure that I got for the E3200. It looks like I would improve it a bit if I lowered the overclock until it would run with stability at default Vcore. It might be worth seeing if a drop from 3.6GHz to say 3.5GHz would achieve that. I kinda like 3.6GHz though :-).
All in all the early numbers make me quite happy with the humble E3200 as a budget choice. However it's very early days yet so I'll keep playing and reporting.
Cheers,
Gary.
Just to insert a coin of my 5
)
Just to insert a coin of my 5 cents into discussion. So, guys, why not to try new MLC-SDDs in these machines? I will soon get a couple of them. But they will work in file server, so I will not be able to measure the power draw, but I can make a picture of them to see their parameters. We here do not need a huge disk space for BOINC only machines, so the lowest capacity SSD drive (just 4 Gbs like in ASUS EEE PC 901) I think will be enough. If someone says that they are so expensive these days than I can offer another solution - COMPACT FLASH card with EIDE-adapter :)
P.S. Yesterday I've seen guys who are crunching E@H on barebone system on Intel Atom N270, altered for use in car instead of regular music box and using free WiFi-spots for the internet where available.
RE: So, guys, why not to
)
While I read that there can be power savings it is far from automatic depending on model. I happen to use a professional sound recorder for which recently solid-state drives have been offered as alternatives to hard drives. In that specific application there was no power saving.
On the other hand a friend of mine who is an Intel employee loaned me one of their well regarded drives about a year ago, and based on the temperature of the housing after it'd been in my computer for a week, and also readings on my system-level power meter, I think it used several watts less than my usual hard drive choices. So I certainly agree it is possible. It should not be assumed.
I have not seen pricing on solid-state drives attractive for economy cruncher use. Can you point to some?
A well-made SSD would likely also be a system level reliability upgrade.
RE: RE: So, guys, why not
)
I've found something in Kingston's site.
I've seen these drives here (South Urals, Russia) for the prices about $200 USD depending on the size and manufacturer. So, it's quite useful instead of low-power 1TB drive if the capacity is not an argument.
RE: I've seen these drives
)
Right now, for new drives at comparable shopping carefulness, here in the US an 80 or 160 Gbyte HD is about $40, while a 32 or 40 Gbyte SSD is about $100. I doubt there is enough power consumption savings to pay that back on almost anyone's billing rate.
What you say should be possible, as the clearest natural SSD advantages over HD are price at the low capacity end, and speed in the sense of latency (no rotational delay). But the SSD makers don't see a prompt road to profit in making, say 8 Gbyte drives for $10 right at the beginning. When you are only making millions, rather than hundreds of millions of the things, you don't get those nice low costs right at the beginning, so they are looking for places where they can offer a compelling enough advantage to earn a huge price premium instead. Right now that is folks who prize the speed advantage, or possibly reliability or power consumption advantages. With luck, perhaps they will pursue the low end later.
The question of SSD drives is
)
The question of SSD drives is unlikely to even be on my radar any time soon. I have plenty of existing drives of various types and sizes and, more importantly, plenty of a particular model that seems very robust and very suited to the job.
Over the years of playing with surplus equipment, you get to form an opinion about which models are good and which are terrible. For HDDs I would place the Maxtor 541DX 20GB slimline drive in the terrible category. When first acquired, many had bad sectors and those that seemed OK initially, failed with monotonous regularity.
When I purchased the first of what would turn out to be my tualatin celeron 1300 workhorse, I noted that they also featured a slimline drive - the seagate ST320014A - which (apart from the label) looked very similar to the Maxtors. I was initially sceptical of how they would perform (being a slimline drive) but my experience has been nothing but good. Even though they date back to the 2002/03 era, I can't recall a single failure in service. They run cool to the touch whereas the Maxtors ran quite hot. I'm very happy to keep using them in all my upgraded machines. Now that I've got the ability to measure it, I guess I'll get around to measuring how much they actually draw in normal running. It can't be very much and must be less than the Maxtor on the basis of operating temperature alone.
Once I work out how much the hard disk draws, I may well consider doing away with them altogether and booting from a live CD with the BOINC stuff hosted on a network share. When added up over all the crunching machines, that might represent a worthwhile saving in power.
Cheers,
Gary.
An HD shouldn't draw much.
)
An HD shouldn't draw much. IIRC desktop drives draw about 10W when spinning and 1W at idle. Laptop drives are significantly lower.
RE: An HD shouldn't draw
)
Gary has his hosts hidden, but given his high RAC, recent use of antiques, and penchant for economy builds, I'd guess it is quite a few. Also, I think recall that BOINC writes to a state file rather frequently, so depending on how long one's OS preferences allow a disk to wait before going down to idle, it may well be that the drive on which the state file lives never idles. 10 watts times a lot of hosts just might add up in a year. Even if it is 5 watts. And given the drives are sensibly warm to the touch (by and large), I'll wager the real time average is not likely much less than 5 watts.
I can't address the relative value of Gary's time to do the network drive idea versus his local cost of power, but over his large empire, I suspect the savings opportunity is considerable. For the rest of us, not so much.
RE: I'll wager the real
)
5 watts is pretty much what I would expect for a 'just warm' drive. As some drives run rather hotter than 'just warm' perhaps some get closer to 10 watts.
At the peak, I had about 240 machines running. Today I have around 55. Of those about 7 are 'oldies' that will soon either be shut down or upgraded. Then I should be finished for a while.
Electricity costs me approx $USD0.15 per Kilowatthour. If I could save 5 watts on 50 machines, my annual saving would be 5x24x365x50x0.15/1000 = $USD328. Since electricity will rise steeply in coming years, it may be worthwhile investigating further. The first step is to actually measure the watts per drive.
It would actually be a bit tedious to convert 50 machines like this. However with Linux, I can set up a single machine with everything installed and configured and all scripts in place. Then I can remaster the machine to make a new live CD of the working installation. The conversion of any single machine should then be as simple as stopping BOINC, moving the complete BOINC installation to the appropriate place on the network drive, disconnecting the hard drive and then booting up from the new live CD which will know where to look for the BOINC stuff on the network.
Cheers,
Gary.