Best computer for Seti

Questions and Answers : Preferences : Best computer for Seti
Message board moderation

To post messages, you must log in.

AuthorMessage
Avenger

Send message
Joined: 16 Mar 16
Posts: 3
Credit: 22,157
RAC: 0
Estonia
Message 1772392 - Posted: 18 Mar 2016, 12:22:26 UTC

Hi, I am new in this thing. I am wondering, what is best computer for these tasks. Is there important powerful gpu or cpu or something else and is there some list of top computers, from where I can compare my machine with others? When I check computer information, there is 'measured floating point speed' and 'measured integer speed'. Are these important numbers? How are these numbers calculated, what is key to increase these numbers? Thanks.
ID: 1772392 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3717
Credit: 8,851,593
RAC: 607
Bulgaria
Message 1772395 - Posted: 18 Mar 2016, 12:52:10 UTC - in response to Message 1772392.  

is there some list of top computers, from where I can compare my machine with others?

On the HOME page look at Top-Right
http://setiathome.berkeley.edu/top_hosts.php


Is there important powerful gpu or cpu or something else

You may compute on whatever you have (even Android phone)
http://setiathome.berkeley.edu/apps.php


When I check computer information, there is 'measured floating point speed' and 'measured integer speed'. Are these important numbers? How are these numbers calculated

Not really "important numbers" - this is from BOINC benchmark (which is based on very old code)
These numbers are used by BOINC to decide initially how many tasks to ask from servers and what estimated times to show
(estimated times are very wrong at the beginning, may be 5x more of the real needed time)

You can do BOINC benchmark manually from Tools menu:
http://boinc.berkeley.edu/wiki/Advanced_view#BOINC_Manager_Menus


The more real numbers are from SETI@home applications
On your computer page:
http://setiathome.berkeley.edu/show_host_detail.php?hostid=7957662

if you click:
Application details Show
http://setiathome.berkeley.edu/host_app_versions.php?hostid=7957662

you will see:
Average processing rate



- ALF - "Find out what you don't do well ..... then don't do it!" :)
ID: 1772395 · Report as offensive
rob smith
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 14907
Credit: 230,003,890
RAC: 386,514
United Kingdom
Message 1772399 - Posted: 18 Mar 2016, 13:12:10 UTC

As a guide:
Intel i7 processor
Motherboard capable of supporting at least four double width cards
4 of Nvidia GTX980ti or 4 of Nvidia Titan X

Add to that Windows 7 Pro, 16GB RAM and a reasonable hard drive, run "optimised" applications, and three tasks per GPU and you would find that computer well up the first page. You would need a decent sized (>1600W) PSU, and not expect to run much in the way of room heating....

Same hardware, but running Linux, using Petri's "magic" app and it would be in the top three.

Cost?? Several thousand dollars in hardware, plus running costs at about 30kWHr/day

Of course you could run with what you've got just now and wait for the new Nvidia GTX1080 which is due on the market in a couple of months time and is supposed to have better performance and lower power consumption than the GTX980ti, cost - I'd guess $1000 each, but lower power consumption....
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1772399 · Report as offensive
Avenger

Send message
Joined: 16 Mar 16
Posts: 3
Credit: 22,157
RAC: 0
Estonia
Message 1772407 - Posted: 18 Mar 2016, 13:31:17 UTC - in response to Message 1772392.  

Before I created this post of course I checked all menus in this program and page. This Top hosts list seems to be just users, who have top scores with credits, not computers, because when I compare top users numbers with my machine, it is not that bad at all. That's why I asked, is there some actual list of computers, listed by calculation power.
Also, I understand, you can use server rack with many machines to be faster, but my target is not to build machine specially for Seti, but just compare home computers used by people. What processors are better and so on.
ID: 1772407 · Report as offensive
AMDave
Volunteer tester

Send message
Joined: 9 Mar 01
Posts: 234
Credit: 6,246,099
RAC: 12,174
United States
Message 1772417 - Posted: 18 Mar 2016, 15:02:05 UTC - in response to Message 1772399.  

...wait for the new Nvidia GTX1080...

Published Friday, March 18th 2016 01:01
NVIDIA's Next Flagship Graphics Cards will be the GeForce X80 Series
There is a graph that can be enlarged by clicking on it. The specs on the Geforce X80 Titan are impressive.
ID: 1772417 · Report as offensive
Avenger

Send message
Joined: 16 Mar 16
Posts: 3
Credit: 22,157
RAC: 0
Estonia
Message 1772426 - Posted: 18 Mar 2016, 15:33:57 UTC - in response to Message 1772392.  

Also, what means average upload/download rate and why it is so low(39KB/sec-353KB/sec)? Is Seti network connection bandwidth full, my actual connection is much-much faster?
ID: 1772426 · Report as offensive
rob smith
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 14907
Credit: 230,003,890
RAC: 386,514
United Kingdom
Message 1772434 - Posted: 18 Mar 2016, 15:49:14 UTC

In your clicking around you missed:
http://setiathome.berkeley.edu/top_hosts.php
which is the list of top computers.
The top computer belongs to Petri33, and is "nothing that special", being a 3.2GHz i7-3930K with four GTX980, but a very special application build that screws the last dime out of those four GPUs.


Answering your question about transfer speeds - Its down to the way the internet works, along with the overheads of transferring small files around.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1772434 · Report as offensive
AlProject Donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1386
Credit: 229,540,042
RAC: 512,290
United States
Message 1772795 - Posted: 20 Mar 2016, 12:25:12 UTC - in response to Message 1772434.  

Wow, just looked at the link, that's Nuts! I mean those cards are nothing to sneeze at, but still, very impressive results. He's got some serious Juju going on there, I hope that he shares it with the rest of the world eventually!

I've had this question for a while now, looking at this chart: http://api.viglink.com/api/click?format=go&jsonp=vglnk_14584755824519&key=2b0adaafa9ad8a29fede7758fada1730&libId=im0hcom5010000ws000DA1lk9bpncmyace&loc=http%3A%2F%2Fwww.pcworld.com%2Farticle%2F2936000%2Fwhy-graphics-card-lust-is-the-cruelest-obsession-relentless-upgrades-take-no-prisoners.html&v=1&out=https%3A%2F%2Fcms-images.idgesg.net%2Fimages%2Farticle%2F2015%2F06%2F980tichart-100591314-orig.png&title=Why%20graphics%20card%20lust%20is%20the%20cruelest%20obsession%3A%20Relentless%20upgrades%20take%20no%20prisoners%20%7C%20PCWorld&txt= (sorry, didn't know how to attach the pic directly into the post, that may have been better) I noticed that you have core count, clock rates, memory width, single and double precision values, etc.

So... What Really matters to BOINC/SETI crunching? I have believed for quite some time that more CUDA cores are always better, and obviously faster (pretty much anything) is better than slower, but if you are lets say buying a Not brand new card, and comparing the specs on the chart above for in help making a wise decision, what are the #1,2 and 3 things to look at when trying to evaluate them? I'd really like to know, so I can file it away for future reference.

ID: 1772795 · Report as offensive
rob smith
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 14907
Credit: 230,003,890
RAC: 386,514
United Kingdom
Message 1772805 - Posted: 20 Mar 2016, 14:00:33 UTC

Sadly your link is broken :-(

As to what is best?
Your assumption is correct within a family of GPUs, and this is broadly reflected by the final two digits of the card number (for Nvidia GTX cards). So a GTX710 is less capable than a GTX760, and less capable than a GTX980. It gets a bit more confusing when trying to compare performance between "families" - generally the higher the first digit, the better the performance, so a '250 will not perform as well as a '450, and that in turn won't perform as well as a '750, which should (in theory) perform as well as a '950.

It is also worth thinking about the power consumption of the cards. The older card families (2xx to 6xx) are generally more power hungry than the newer cards (7xx and 9xx) for a given value of xx - an outstanding example of this is the GTX750 returns a disproportionately good credit/watt.

An observation - In one of my computers I used to run a pair of GTX690s - that is a a pair of dual processor cards (based on the GTX680), these drew about 300W each, and returned a lower credit/day than the pair of GTX780 that currently sit on the same motherboard & processor - the GTX780s are single processors and only draw about 250w... I know which ones I prefer to keep fed with electricity!!!
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1772805 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3717
Credit: 8,851,593
RAC: 607
Bulgaria
Message 1772806 - Posted: 20 Mar 2016, 14:12:14 UTC - in response to Message 1772795.  

... didn't know how to attach the pic directly into the post ...

Open the image in new tab (have to be direct link to picture, not .html) - e.g. by Right-Click on picture -> Open image in new tab

(Note: api.viglink.com seems to not be "ideal" site, was blocked by 2 lists I use:
"uBlock Origin has prevented the following page from loading ...
Because of the following filter
||api.viglink.com^
Found in: hpHosts’ Ad and tracking servers • Dan Pollock’s hosts file
"
Also on WOT it is not too good:
https://www.mywot.com/en/scorecard/api.viglink.com
)


I think this is the picture you want?
http://core0.staticworld.net/images/article/2015/06/980tichart-100591314-orig.png
[img]http://core0.staticworld.net/images/article/2015/06/980tichart-100591314-orig.png[/img]

Use [img] the same way as you use [url]






- ALF - "Find out what you don't do well ..... then don't do it!" :)
ID: 1772806 · Report as offensive
AlProject Donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1386
Credit: 229,540,042
RAC: 512,290
United States
Message 1772973 - Posted: 21 Mar 2016, 2:40:12 UTC
Last modified: 21 Mar 2016, 2:59:58 UTC

Sure is, thank you very much! Rob, thanks for the reply, I kind of understood the generalities of it, but was hoping by looking at the above chart, along with personal experiences like yours, I could get some informed opinions of what was really what. For example, right now I could buy a used Titan Black for the same price as a GTX 980TI ACX 2.0+, both around $650 or so, and here is a comparison of the 980 lineup: http://www.evga.com/Products/Product.aspx?pn=06G-P4-5998-KR. Which is 'better'? In this case I'd probably the say the latest 980Ti, but the 'normal' 980? I don't know. And on paper, the Titan X looks 'better'. But is it? One's a little faster, the other has 2x the memory.

If looking at the top of the line (980 Series), in that link, there are 2 cards - the 06G-P4-4998 & 06G-P4-4996 - exact same perf specs Base/Boost 1190/1291, but different Power Phase - 14+3 vs. 8+2 - and Max Power Draw 250w vs. 300. Is the thought that the 14+3 would have more stable power and hence overclockability? Otherwise why spend a little more for the card and a bit more long term for powering it if they perform exactly the same?

As to my earlier question about using that chart to try and figure out what the most bang/buck would be at any given price point between used cards, what are the most important things to look at when comparing cards for crunching? Cores? Raw clock speed? Something else? Does the available memory allow more WU's to process simultaneously?

And of course, there is the ever present Next Gen coming down the pike sooner or later (and seems like it's usually sooner, especially if you just bought the current gen top of the line), which will usually bring more performance and less power consumption, or at least more performance.

ID: 1772973 · Report as offensive
rob smith
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 14907
Credit: 230,003,890
RAC: 386,514
United Kingdom
Message 1773005 - Posted: 21 Mar 2016, 6:20:36 UTC

When advising folks I tend to avoid looking at specific card manufacturers sites - they change their products more often than the chip-set manufacturers, and they are aiming at the gaming market, not the 24/7 computing market - decide on the chip set that looks best to your needs, then have a look at the prices from the top board people (Asus, EVGA, Gigabyte and a couple of others).

For every user, apart from those with unlimited budgets purchase price and running cost come into the equation for "better". Here's a chart I've found useful over the years, it is Nvidia's view of the relative performance f their GPUs:


Ignore the numbers up the lefthand side, compare the heights of the bars - you will see that the Titan-X, GTX980ti and the Titan-Z are all very similar., and all three are faster than the GTX980. If you now look at your favourite shop site you will get the purchase price, and if you look at the Nvidia page for each of GPU in turn you will see the power draw: Titan-Z = 375W, GTX980ti = 250W, Titan-x = 250W, while the GTX-980 only draws 165W. You've now got the basic figures for "stock" cards. So you can do your sums and work out the cost per year of purchasing and running them, and, if you take a look at my computer with the two GTX980 (http://setiathome.berkeley.edu/results.php?hostid=6452693) you will be able to guess at the RAC each configuration will get. For my money I wouldn't go for a Titan-Z as it costs a lot to run, or a Titan-x as it costs a lot to buy. I already had the two GTX980 when the GTX980ti came out. (The usual, a good upgrade comes along just after one has splashed the cash....)

If I were building a system from the floor up today I would wait the couple of months until the new Nvidia X-80 (or whatever they are going to call it) comes out, as that would give me a bit of a chance to save a little more and not have to "bend the plastic" so hard.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1773005 · Report as offensive
rob smith
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 14907
Credit: 230,003,890
RAC: 386,514
United Kingdom
Message 1773012 - Posted: 21 Mar 2016, 8:12:45 UTC

One thing I haven't mentioned yet - cooling...
Unless you are lucky enough to have a dry-air freezer with enough spare space and capacity you will have to be careful with how you cool your system. This starts with having a case that is both big enough and with a decent designed in air flow - cold in on one face, and hot out on the opposite face, with the ability to direct cold air onto the RAM, bridges and the like, that has enough space for all the bits - I've got one case that is big enough for the motherboard, but I can't use one of the slots because the case gets in the way. Then you have the call "water or air" - do you cool everything with air, do you use hybrid, or do you go all water. each has its advantages and disadvantages.
Most of my systems have hybrid for the CPU and air for the GPUs, one is all air. Whatever you choose you do need to make sure you get the hot air out of the case and not just pump it around inside - I've gone for the "force air in" approach, with more fans pushing air in than dragging it out (I used to work on plant where the off-air was at about 1000C, so putting a fan on the hot side of that was "not a good idea", and the habit has carried on). I did have water on the pair of GTX690s, but that died when the computer got flooded (from an external source).
With GPUs it is worth noting that many of them have multiple fans, and so dump the heat into the case - which is OKish when there's only one in the case, but when you get three or four this becomes quite a problem. Most standard Nvidia cards are designed with a single fan towards the inboard end, and force air out of the case, which is good, because that gets the heat out, but not so good as the air flow doesn't support the 24/7 heat generation that we run.
Now, breaking my own rule, Asus do a series of GPUs where the fans draw air in from both sides of the board. I've two pairs of these in service, a pair of GTX960 and a pair of GTX970, it would appear that these run a bit cooler than I would expect, and certainly a lot cooler than the pair of GTX460 that were replaced by the GTX960s.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1773012 · Report as offensive

Questions and Answers : Preferences : Best computer for Seti


 
©2017 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.