Posts by HAL9000

21) Message boards : Number crunching : AMD 290X vs RX 480 for seti/ DUAL NVIDIA 1070 vs Single 1080 (Message 1845177)
Posted 21 days ago by Profile HAL9000
Post:
i came back to find the cruncher on the login screen. I've checked events and I couldn't see any crash, it just rebooted with critical event that it wasn't shut down properly... is there anything in the commandline that might lead to this? the screen is connected to the iGPU which isn't used for seti.

I will leave it running and see if this happens again.

I have found instances when closing BOINC that happens. Normally it is when I have multiple clients running on a system and I close them all at once.
22) Message boards : Number crunching : AMD 290X vs RX 480 for seti/ DUAL NVIDIA 1070 vs Single 1080 (Message 1845074)
Posted 21 days ago by Profile HAL9000
Post:
Heres my final dilemma. I got saved up to buy two gtx 1080 or 4 rx480. Or three 1070. Also got the rx480 does the 4gb vs 8gb make big difference?
What should I do?

Either 4GB or 8GB will be fine for SETI@home work. The mos I have used for a SETI@home task on my R9 390X is ~400MB and that was when using command line options that increased the memory usage of the app.

If you want to go with Nvidia or Radeon you can pick between better FLOPs per Watt or FLOPs per $.
On paper the specs look something like this:
GTX 1080 8228 GFLOPs, 180W, $700 x2 = 16456 GFLOPs, 360W, $1400 or 45.71 GFLOPs/W, 11.75 GFLOPs/$
GTX 1070 5783 GFLOPs, 150W, $450 x 3 = 17349 GFLOPs, 450W, $1350 or 38.55 GFLOPs/W, 12.85 GFLOPs/$
RX 480 5161 GFLOPs, 150W, $200 x 4 = 20644 GFLOPs, 600W, $800 or 34.40 GFLOPs/W, 25.80 GFLOPs/$

In practice there isn't a direct correlation between FLOPS and doing SETI@home work. Especially between different vendors. Sometimes it isn't even useful to compare different generations of GPUs this way.
You can have a look at Shaggie's GPU FLOPS: Theory vs Reality thread to see a performance comparison using credit.

This doesn't tell the complete story for all GPUs. As there are so many different manufactures and app configuration options.
In comparing my R9 390X to hosts with a R9 Fury X and tasks with a similar AR. They are pretty much on par. With the worst case being my R9 390X about 9% slower than a hosts with highly optimized settings.

EDIT:
I was just comparing the run times of Michel Makhlouta's 1070's to my R9 390X. It looks like my 390 completes similar tasks bout 25% faster and the 390 is only rated about 2% higher in GFLOPS. GTX 1070 - 5783 GFLOPs vs R9 390X - 5913 GFLOPs


You looking at the cards, from your PC vs Michaels
46,247.89 [2] NVIDIA GeForce GTX 1070 (4095MB) driver: 376.33 OpenCL: 1.2
/2
23,123.945 about per card
vs
AMD Radeon R9 390X (Grenada XT) (8192MB) driver: 1912.5 OpenCL: 2.0
21,884.42

The amd card is very close the nvidia card, but the amd card costs about 60% less or am I missing something here?

It looks like you are comparing RAC of the two systems. I was comparing the run time of similar tasks.

I just started using my R9 390X on about 2017-01-21. So the system RAC is still rather low. The current daily numbers looks to be in the 30-33k range. Before using the GPU the system was running 2 CPU tasks at a time and had a RAC of ~5500.
23) Message boards : Number crunching : AMD 290X vs RX 480 for seti/ DUAL NVIDIA 1070 vs Single 1080 (Message 1845070)
Posted 21 days ago by Profile HAL9000
Post:
The amd card is very close the nvidia card, but the amd card costs about 60% less or am I missing something here?

The R9 390X appears to be rated for around 250W. The GTX 1070 is rated at 150W, up to 180W for overclocked versions.
I'd rather pay more up front for a card that will cost a lot less to run, than pay less up front for a card that will cost a lot more to run. Particularly when they produce similar amounts of work.

When running SETI@home work GPUz normally shows between 130-150w. The current average after about 9 days shows 142w .
The power usage is about 60% of the TDP. Which is similar to the power usage for the 750 Ti I have. So I would guess a 1070 runs around 90w or so?

Normally I don't run MB work on the 390x. Since the CPU in the system is more power efficient at doing SETI@home MB work. The last time I was using it was before we had GBT work. Plus I'm running the ATI SoG app. Which Raistmer suggested may not be the most efficient for the GPU. Once I had a good baseline I'll switch to the ATI HD5 app.
24) Message boards : Number crunching : Panic Mode On (104) Server Problems? (Message 1844634)
Posted 23 days ago by Profile HAL9000
Post:
Just adding an idea here.. but the feeder can only re-fill so fast/frequently/often, so what if instead of letting someone get lucky and get all 200 tasks that it has in one request.. the feeder gets limited to assigning 20 or 50 tasks at a time? I know there will be more groaning and griping about that, too, but if the feeder is more likely to have tasks in it, then there should--theoretically--be less people who get absolutely nothing because it is empty.

Something similar was done with AP work requests a few years ago. At least it seemed that way once no more than ~7 tasks were assigned per request.
If hitting an empty feeder is the only issue it seems like more users would be seeing their queues dropping & that toggling settings wouldn't fix the users getting no work.
At another project there was an issue with the feeder running dry. So the admin adjusted some settings for it. I think they said they increased the number the feeder held at once, but they may have increased how often it was filled.
25) Message boards : Number crunching : Panic Mode On (104) Server Problems? (Message 1844154)
Posted 26 days ago by Profile HAL9000
Post:
. . Strange, out of my 3 rigs only one has tasks in this state but it has something like 45 of them. So if 1 in 3 has this problem then that, as you say, would add up to an awful lot of tasks in limbo on a global scale. I hope they are aware of the problem and again like you I hope they fix it tomorrow ...
Stephen
.

I'm seeing it on 3 of 5 rigs, all on jobs returned 6 Dec 2016, and all have been in that state since 6 Dec.
(i.e., both wingmen returned the work on 6 Dec)
Given that, no maintenance tomorrow will resolve this unless it is specifically addressed, I would guess.
Hopefully there's someone reading this thread that can get the report to the right folks ...

Out of ~1400 pending tasks I have ~60 from that time period pending. Out of that 14 that are awaiting validation with two results that were returned on the 6th or 7th.

There is a process that will check the tasks again when their original Report deadline passes. Then they should be sent to the validator.
26) Message boards : Number crunching : Should I get a new Power Supply? Math help needed (Message 1843695)
Posted 28 days ago by Profile HAL9000
Post:
To figure out my cost per KW/h I just divide the amount I'm changed by the KW/h. So for 1285 KW/hs I paid $208.41, or $0.162 KW/h. Since the rates change during the year I keep a yearly average. Which is closer to $0.18 KW/h.

Going from a Bronze to Gold PSU would save about 6.5w for every 100w of load. If we figure a 250w system load about 16.25w would be saved, or 142.35 KW/h a year. With my cost of $0.18 KW/h that would save about $25 a year and it would takes nearly 4 years for the PSU to pay for itself.

I really only compare the cost of more efficient PSUs when I am already replacing it for another reason. Like when I bumped my gaming system from a 650w to a 750w PSU. I only had to factor the difference in cost between the more more efficient models. Instead of the whole cost of the PSU. The EVGA SuperNOVA 750W Gold and Platinum PSUs were only about $10 different and I calculated I would save about that in a year. So I opted for the Platinum model. The Titanium model would have saved <$20 a year and it was $50 more than the Gold model.
27) Message boards : Number crunching : Another dumb question...................RAM (Message 1843097)
Posted 19 Jan 2017 by Profile HAL9000
Post:
There are some projects than use quite a lot of ram. There was one I was running that used ~4GB per task. Currently SETI@home the memory requirements are pretty low. Normally <50MB per task for CPU tasks. GPU tasks can use more, but that varies based on the specific app. I think 100-200MB per GPU tasks might be a good rule of thumb.

If the systems were using all of the RAM and doing a lot of disk thrashing then the increased memory will defiantly help.
28) Message boards : Number crunching : AMD 290X vs RX 480 for seti/ DUAL NVIDIA 1070 vs Single 1080 (Message 1842983)
Posted 19 Jan 2017 by Profile HAL9000
Post:
I was just comparing the run times of Michel Makhlouta's 1070's to my R9 390X. It looks like my 390 completes similar tasks bout 25% faster and the 390 is only rated about 2% higher in GFLOPS. GTX 1070 - 5783 GFLOPs vs R9 390X - 5913 GFLOPs

How well the application is able to make use of the hardware is what really determines whether your card will produce a lot of work or not. Doesn't matter how much capability the card my have, if the software can't take advantage of it.

Indeed. I added that to emphasize the comment I made a few lines before that.
In practice there isn't a direct correlation between FLOPS and doing SETI@home work. Especially between different vendors. Sometimes it isn't even useful to compare different generations of GPUs this way.
29) Message boards : Number crunching : Panic Mode On (104) Server Problems? (Message 1842954)
Posted 19 Jan 2017 by Profile HAL9000
Post:
I use 7.6.33 with the manager from 6.10.48. I don't need a notices tab and I use the messages tab enough that having to open Event Log is simply annoying.

The showstopper for me is the lack of app_config with the older manager.

Ah, I use notepad for that.

It is odd that some users with the issue are having luck toggling their web preferences back and forth. While others with the same settings, and haven't touched their preferences in months, are not having any issues.
My SETI@home prefs are either Yes, Yes, No or No, Yes, No and look to have been that way since at least April of last year.
Are the only hosts effected those that have or did have "accept work from other applications" set to Yes before the issue effecting that setting occurred?
30) Message boards : Number crunching : Panic Mode On (104) Server Problems? (Message 1842948)
Posted 18 Jan 2017 by Profile HAL9000
Post:
I use 7.6.33 with the manager from 6.10.48. I don't need a notices tab and I use the messages tab enough that having to open Event Log is simply annoying.
31) Message boards : Number crunching : Cruncher Down (Message 1842947)
Posted 18 Jan 2017 by Profile HAL9000
Post:
But I have never had a bad ASUS.


Mine went from bad to worse... received the RMA replacement, put everything together and then "the lights went on but no one was home": when it blew, it took out the AMD FX 8-core CPU. This is exactly what the MSI boards kept doing. And its warranty had expired.

So, about $300 (Canbucks) later as I bought a three-year in-store replacement warranty, I have a new CPU on the RMA board in case it dies the same way and fries another one. Funny... the fellow at the store mentioned just this, the some people just have more luck with certain brands of boards and swear by them whereas others have endless problems. I hope this was my only ASUS lemon.

I used to swear by ASUS boards and now I just swear at them. My lat Asus MB board was for a Core 2 Duo system ~2009. After replacing it with a Gigabyte MB & taking a hammer to the Asus MB I no longer had issues & felt better.
Over the years I've transitioned from Abit > Asus > Gigabyte. With the occasional Intel or Supermicro MB. I do have an ASRock MB I bought not to long ago and I have to say their quality has gone up quite a bit since Asus sold them off.
32) Message boards : Number crunching : AMD 290X vs RX 480 for seti/ DUAL NVIDIA 1070 vs Single 1080 (Message 1842940)
Posted 18 Jan 2017 by Profile HAL9000
Post:
Heres my final dilemma. I got saved up to buy two gtx 1080 or 4 rx480. Or three 1070. Also got the rx480 does the 4gb vs 8gb make big difference?
What should I do?

Either 4GB or 8GB will be fine for SETI@home work. The mos I have used for a SETI@home task on my R9 390X is ~400MB and that was when using command line options that increased the memory usage of the app.

If you want to go with Nvidia or Radeon you can pick between better FLOPs per Watt or FLOPs per $.
On paper the specs look something like this:
GTX 1080 8228 GFLOPs, 180W, $700 x2 = 16456 GFLOPs, 360W, $1400 or 45.71 GFLOPs/W, 11.75 GFLOPs/$
GTX 1070 5783 GFLOPs, 150W, $450 x 3 = 17349 GFLOPs, 450W, $1350 or 38.55 GFLOPs/W, 12.85 GFLOPs/$
RX 480 5161 GFLOPs, 150W, $200 x 4 = 20644 GFLOPs, 600W, $800 or 34.40 GFLOPs/W, 25.80 GFLOPs/$

In practice there isn't a direct correlation between FLOPS and doing SETI@home work. Especially between different vendors. Sometimes it isn't even useful to compare different generations of GPUs this way.
You can have a look at Shaggie's GPU FLOPS: Theory vs Reality thread to see a performance comparison using credit.

This doesn't tell the complete story for all GPUs. As there are so many different manufactures and app configuration options.
In comparing my R9 390X to hosts with a R9 Fury X and tasks with a similar AR. They are pretty much on par. With the worst case being my R9 390X about 9% slower than a hosts with highly optimized settings.

EDIT:
I was just comparing the run times of Michel Makhlouta's 1070's to my R9 390X. It looks like my 390 completes similar tasks bout 25% faster and the 390 is only rated about 2% higher in GFLOPS. GTX 1070 - 5783 GFLOPs vs R9 390X - 5913 GFLOPs
33) Message boards : Number crunching : Panic Mode On (104) Server Problems? (Message 1842762)
Posted 17 Jan 2017 by Profile HAL9000
Post:
[. . Hey guys,
. . Can I do a quick poll?
. . What setting is in your "get additional days of work field"? Both for those getting work and those having problems getting any.
Stephen.

7.6.22 with no issues, setting are:
Store at least 10 days of work
Store up to an additional 10 days of work

Doesn't matter in my case, as those parameters are never enforced.
The hard limit of 100 tasks per CPU and 100 tasks per GPU keep me limited to a total of less than 12 hours work for most of the machines here.

My default BOINC web settings are also 10/10 days of work. I did have a mix of 7.2.42, 7.4.42, 7.6.23, & 7.6.33 on my machines, but recently switched all of the Windows machines to 7.6.33. I wasn't having any issues getting work. I just wanted to bring them all up to the same version.
34) Message boards : Number crunching : Panic Mode On (104) Server Problems? (Message 1842525)
Posted 16 Jan 2017 by Profile HAL9000
Post:
My settings are yes, yes, and yes.

That appears to be it.
Try changing them to accept v8 work only and see if they stop getting work for a while.

Settings that used to work, no longer do. Sound familiar?

Most of my machines are in a venue that is set to only allows MB v8 work. I have 2 hosts that are in a venue with AP v7 enabled.
So that setting hasn't effected my ability to get work.
35) Message boards : Number crunching : BOINC odd network traffic (Message 1842415)
Posted 15 Jan 2017 by Profile HAL9000
Post:
Doing some tests I can confirm that resource monitor does show increased network activity depending on what is being viewed in BOINC Manager.
Transfers or disk tab
http://i.imgur.com/iGSCt0v.png
Projects tab
http://i.imgur.com/MPcwuVG.png
Tasks tab: View active
http://i.imgur.com/iJ9iMZ4.png
Tasks tab: View all
http://i.imgur.com/7c5WLwQ.png

However I could not confirm the displayed activity effected my network/internet bandwidth.
I restricted the internet bandwidth for the machine I was testing to 1024KB in my router.
With BOINC set to view all in the task tab I went to speedtest.net and ran a test.
Resource monitor displayed 2MB Network I/O & the results for the speedtest were 1MB/1MB.
36) Message boards : Number crunching : Panic Mode On (104) Server Problems? (Message 1842398)
Posted 15 Jan 2017 by Profile HAL9000
Post:
In checking the logs across my machines for the past few days the only time I have got a Scheduler request completed: got 0 new tasks response it was followed by This computer has reached a limit on tasks in progress.
37) Message boards : Number crunching : RAC & Badges (Message 1842166)
Posted 14 Jan 2017 by Profile HAL9000
Post:
As far as I'm aware badges are based on total contribution not RAC.

No, it's RAC.
I'm often going from no badge, to now a silver badge, and in a week or so if I continue to crunch like this a gold badge, all depending on how much I crunch. When I only run my tablet for a while, my badge quickly goes down to silver, and then bronze, and then disappear totally.

For example, this guy has a bronze badge with only Total credit 549,922, but a RAC of 1,195.67: https://setiathome.berkeley.edu/show_user.php?userid=10259455

And this guy with a Total credit 7,327,754, but a RAC of 1,039.18, doesn't have any badge at all: https://setiathome.berkeley.edu/show_user.php?userid=149109

Viewing Top participants by total credit will currently show similar information. With some users without and a mix a gold and silver badges.

Alto worth noting in that sometimes the script that updates the RAC badges doesn't run or gets stuck. So if you are near a transition from one to the other your badge may not update until someone pokes the script.
38) Message boards : Number crunching : RAC & Badges (Message 1841964)
Posted 13 Jan 2017 by Profile HAL9000
Post:
Last week you could get a Silver badge with a RAC in the high 5k, now you need mid 7k. Did I miss the memo?

Since the number of users and their participation is variable there will never be a fixed value for the % badges.
39) Message boards : Number crunching : Looking for used laptops (Message 1841413)
Posted 11 Jan 2017 by Profile HAL9000
Post:
GPUs take power from the 6 or 8 pin power cables (if GPU has them) plus up to 75 Watts from the PCI bus.

That why when you use PCIe extender cables they require (most of them) the additional power to be supplied from either a 4 pin Molex (HD power plug) or SATA power plug. The included SATA to 4 pin Molex adaptor is just to make sure you have an available power source.

I imagine they do that because most new PSUs have more SATA power connectors than 4-pin molex. At least that is the case with all of the PSUs I've bought in the past several years.
40) Message boards : Number crunching : Pine A64+ 2GB running 24/7 (Message 1840773)
Posted 8 Jan 2017 by Profile HAL9000
Post:
If you have the time that would be great.

I'm at work right now. But when I get home, I'll power it down, move it to a power meter, restart everything, and let it run for a few hours. Once power consumption stabilizes at a constant running temperature, I'll post an average reading from the meter. It will be just the Pine A64+ It will not have any loads from fans on it. The fan will be on a separate circuit.

Ok. I let it run overnight and it is now pulling down 5.7 Watts according to the meter. https://flic.kr/p/QNCWq2
[img ]https://c2.staticflickr.com/1/563/32032890761_50085bd8b2_n.jpg[/img]

Since you probably can't run it 100% here without the fan, you should add what the fan draws too, to that.

A typical fan for these kind of devices normally runs in the 0.25-0.5W range.


Previous 20 · Next 20


 
©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.