I have a computer I built with an AMD FX 8350 and an ATI R9 270. On Seti that computer is lucky to do 30k a day, but on Milkyway that same computer will do over 80K a day. Seti does not seem to work well with ATI or AMD. I have the processor overclocked to 4.81ghz and the GPU at 1050mhz. Well over stock. I pretty much quit on Seti because it seems like a huge waste of time and power to accomplish so little. I am just curious if this is just how it is? Nividia or waste time?

You realize they seem to use two different forms of granting credit? Everything in milky way seems to either grant 20 or 106.88 credits, regardless of how long the wu ran. You can never compare credit granted between 2 projects and get any relation to relative performace...

I run several AMD cards and they are quite competitive. They can be a tad slower than NVIDIA on multibeam sometimes but they are just as fast or faster than them with the new Greenbank data. Likewise they scream with astropulse work units. Take a look at my machines, my Macs run 3 at a time per card and my one AMD Windows machine runs at a time. They are all very competitive with 2 card machines, particularly given they are all based on AMD tech from 2011...

Chris

---

I am not concerned with RAC, but the number of floating point calculations. If I look at what I did not MW and Seti, my MW calculations are far higher in a shorter amount of time. 1.28 quintillion floating-point operations on Seti and 4.69 quintillion floating-point operations on MW, and I spent less time on MW. I am going by the certificate for that. I am not sure how else to compare the 2. I do know MW seems to load my GPU far more than Seti does. I run 2 at a time on MW like I did on Seti.

Well you computers are hidden on seri so I can see what your run times are like exactly. Where are you seeing these flop counters?

On milky way your apr (average processig rate) is between 40 and 96 GFLOPs, which you can roughly double if you are running two at a time. That is unlikely to be hugely different than what you are getting on seti... Now I have no idea if those two values are comparable between projects.

That said, I get effective rates of between 290 and 520 GFLOPS depending on the capability of the particular card on multibeam and between 1800 and 2100 GFLOPS for astropulse.

Have you tried tweaking you command line settings to get some more performance on seti?

Chris

---

You should be able to see my computers now. Not sure why I unchecked that.

Were are you seeing the average compute rate? Forget it. I just found it.

Have you tried tweaking you command line settings to get some more performance on seti?


Not sure what you mean?

I have a computer I built with an AMD FX 8350 and an ATI R9 270. On Seti that computer is lucky to do 30k a day, but on Milkyway that same computer will do over 80K a day. Seti does not seem to work well with ATI or AMD. I have the processor overclocked to 4.81ghz and the GPU at 1050mhz. Well over stock. I pretty much quit on Seti because it seems like a huge waste of time and power to accomplish so little. I am just curious if this is just how it is? Nividia or waste time?

I'm sorry for saying this but if you have a look through some of the other threads here you will notice that the operation of the credit system is, depending on how strong your view is, bonkers (that may be to kind for some).

There has been a lot of angst expressed over the last 3-4 years on it's behaviour and whilst many have complained, there seems to be little to no desire to fix it.

As to cross project standardisation of recognition, I think you can kiss that one goodbye.

My suggestion would be if you want to do seti, then do so in the knowledge that within seti your recognition may be rated at X, and for other projects it may be rated as a multiple of X.

cheers and happy hunting

---

Credit is worth exactly nothing, so I am not too worried about that. I know MW tends to "pay" more than Seti. It is the number of calculations done that I am more concerned with. If I run my system for 1 week on Seti and one week on MW, it will accomplish more work on MW going by the number of floating point operations accomplished. That is what I am trying to figure out. Why there seems to be such a huge difference for me. Not in the amount of credit each project gives. There are a few projects on Boinc I feel are worth my time and power bill, but I want to get the most bang for the buck do to speak.

I built my system because I enjoy overclocking. Now I want to put it to use doing something worthwhile.

I'm not sure that SETI & Milkyway can be compared directly. SETI uses Single Precision operations while Milkyway does Double Precision, at least for GPU apps & I would guess CPU apps as well. SETI does FFT calculations. I don't know what Milkyway does.

A comparison of the two with my Radeon R9 390X
Running 2 Milkyway tasks at a time: The GPU power usage is ~125w range & tasks run ~35 seconds.
Running 1 SETI task at a time: The GPU power usage is ~230w & tasks run about 6 minutes.

AMD CPUs generally have weaker ALUs compared to their Intel counterparts. When it comes to SETI performance that aspect is important.

I would say you should run whichever projects you feel are worthwhile for your processor time.

Also are you extracting the estimated FLOPs from the tasks sent to your system for each project, or calculating them some way?

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Credit is worth exactly nothing, so I am not too worried about that. I know MW tends to "pay" more than Seti. It is the number of calculations done that I am more concerned with. If I run my system for 1 week on Seti and one week on MW, it will accomplish more work on MW going by the number of floating point operations accomplished. That is what I am trying to figure out. Why there seems to be such a huge difference for me. Not in the amount of credit each project gives. There are a few projects on Boinc I feel are worth my time and power bill, but I want to get the most bang for the buck do to speak.

I built my system because I enjoy overclocking. Now I want to put it to use doing something worthwhile.

"Number of calculations" done devices awarderd from credit granted, they interconnected. So one could not say that doesn't care about credit but worries about "number of calculations". It's the same thing (as BOINC computes it) in essence.

One does not usually compare credits between projects....
All projects value the work done differently.

You can compare your computor to a similiar one here on seti to see if your computor are doing as good as it should.

I'd say your computor does what it should here on seti.

One does not usually compare credits between projects....
All projects value the work done differently.

Although if Credit New worked properly, and every body used it, you could compare credit between projects because that's what it was meant to allow...

---

Grant
Darwin NT

I have a Windows PC with an AMD A10-6700 at 3.7 GHz and a GTX 750 Ti OC graphic board.I have also a Linux box with an Opteron 1210 at 1.8 GHZ and a AMD/ATI HD 7770. I am running SETI@home stock apps on both and I see no great difference. On the Linux box I am running also SETI Beta, alternating it with SETI@home, and CPDN. The Windows PC runs all CERN Projects with Virtual Box on its larger RAM, 24 GB, while the Linux box has only 8 GB.
Tullio

---

One does not usually compare credits between projects....
All projects value the work done differently.

You can compare your computor to a similiar one here on seti to see if your computor are doing as good as it should.

I'd say your computor does what it should here on seti.

IMHO BOINC should be concerned with how each project allocates credit otherwise scheduling among project will be wrong. For example I also am trying to split contributions on single machine between MILKYWAY and SETI. In my case MW is processing 9x "work" vs. SETI daily average basis.

I also noticed that "deadline" for MILKYWAY is early than for SETI by two weeks! All MILKYWAY scheduled work is due by 20 May 2016 while on SETI all work is due in June!

SO BOINC runs more MILKYWAY WU(tasks) than SETI. Currently 4 MILKYWAY active tasks to one SETI.

In project page MILKYWAY "resource share" is 33% while SETI "resource share" is 67%. Not quite the way active tasks, work units, are currently running.

Is this an issue for BOINC, SETI or MILKYWAY to address via current allocation method for work?

---

One does not usually compare credits between projects....
All projects value the work done differently.

You can compare your computor to a similiar one here on seti to see if your computor are doing as good as it should.

I'd say your computor does what it should here on seti.

IMHO BOINC should be concerned with how each project allocates credit otherwise scheduling among project will be wrong. For example I also am trying to split contributions on single machine between MILKYWAY and SETI. In my case MW is processing 9x "work" vs. SETI daily average basis.

I also noticed that "deadline" for MILKYWAY is early than for SETI by two weeks! All MILKYWAY scheduled work is due by 20 May 2016 while on SETI all work is due in June!

SO BOINC runs more MILKYWAY WU(tasks) than SETI. Currently 4 MILKYWAY active tasks to one SETI.

In project page MILKYWAY "resource share" is 33% while SETI "resource share" is 67%. Not quite the way active tasks, work units, are currently running.

Is this an issue for BOINC, SETI or MILKYWAY to address via current allocation method for work?

What you are seeing with the resource share has nothing to do with credit the more with the time debt system that the BOINC client uses to try balancing resources. The BOINC client aims for more of a long time resource allocation balance. With your resource share settings the client could run SETI@home for 60 days & then run Milkyway for 30 days.
The number of tasks completed doesn't really equate to amount of work done. Different projects perform different types of calculations. How efficient the hardware and software you use to perform those calculations determines how much actual work your computer(s) are doing.

SETI@home has long deadlines to allow as many hosts to participate as they consider can return work in a reasonable amount of time. Projects decide on their deadlines & size of tasks by what their needs are.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Thank for explanation. Helps a bit in understanding numbers but since I joined these projects at different time frames but NOT what I was targeting for contributions to each project.

Project Total credit Average credit Since
Einstein 2,572,507 5,593 8 Sep 2007
MilkyWay 588,994 1,059 7 Dec 2014
SETI 10,819,614 1,016 3 Oct 1999

BOINC is trying to match "resource share" to long term "total credit" not my intent. Guess I have to change "resource share" down for the other projects to match my intent to balance contribution "average daily" work since I joined projects at different dates total credits differ due to start dates.

---

Thank for explanation. Helps a bit in understanding numbers but since I joined these projects at different time frames but NOT what I was targeting for contributions to each project.

Project Total credit Average credit Since
Einstein 2,572,507 5,593 8 Sep 2007
MilkyWay 588,994 1,059 7 Dec 2014
SETI 10,819,614 1,016 3 Oct 1999

BOINC is trying to match "resource share" to long term "total credit" not my intent. Guess I have to change "resource share" down for the other projects to match my intent to balance contribution "average daily" work since I joined projects at different dates total credits differ due to start dates.

BOINC is NOT trying to match "resource share" to long term "total credit" in any way what so ever.
Resource share only refers to the amount of run time the client will give to each project. Your project credit and when you started with the project are not factors in any way what so ever.

If you are trying to adjust the resource share so you have an equal RAC for each project I wish you luck. As it has already been explained credit at one project is not comparable to another. Doing that is a bit like having a stack of 20 dollar bills, 20 Euros, & 20 Yen. Then saying they are equal because there is the same number in each stack.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

BOINC is NOT trying to match "resource share" to long term "total credit" in any way what so ever.
Resource share only refers to the amount of run time the client will give to each project. Your project credit and when you started with the project are not factors in any way what so ever.

Sadly, neither version of the story is (currently) true.

Balancing Resource Share by runtime (with the associated concept of 'debt') was abandoned in 2010. Although they wanted to replace it with a resource share based on actual granted credit (measured by RAC, rather than lifetime totals), they found that wasn't possible, either.

So they found a third way - using REC (recent estimated credit) as the basis for resource share. REC does away with all the project over/under payments, delays due to validation problems, etc. It's written up in Proposal: credit-driven scheduling.

I wanted to contribute the use of two machines and balance each machine to contribute equally to two projects, daily average. One SETI and EINSTEIN and the other SETI and MILKYWAY. They were "balanced" then starting late April SETI started falling behind based on "host average". By mid May, 11, EINSTIEN was 9x over SETI and MILKYWAY was 5x over SETI. Example: there may be 4 EINSTEIN tasks to 1 SETI running???

Trying to locate a reason for change but too many possibilities:

1: Deadline - both MILKYWAY and EINSTEIN are within 30 days while SETI is over 30 days away.

2: Resource share?

3: BOINC scheduling?

4) Project work units style - MILKYWAY work unit use all available cores/cpus and EINSTEIN work units are long 22+ hours v. SETI ~4hrs one core(cpu)??

Changed "resource share" and reset projects to see changes over next month. Any background to help understand behavior is welcome.

---

I wanted to contribute the use of two machines and balance each machine to contribute equally to two projects, daily average. One SETI and EINSTEIN and the other SETI and MILKYWAY. They were "balanced" then starting late April SETI started falling behind based on "host average". By mid May, 11, EINSTIEN was 9x over SETI and MILKYWAY was 5x over SETI. Example: there may be 4 EINSTEIN tasks to 1 SETI running???

Trying to locate a reason for change but too many possibilities:

1: Deadline - both MILKYWAY and EINSTEIN are within 30 days while SETI is over 30 days away.

2: Resource share?

3: BOINC scheduling?

4) Project work units style - MILKYWAY work unit use all available cores/cpus and EINSTEIN work units are long 22+ hours v. SETI ~4hrs one core(cpu)??

Changed "resource share" and reset projects to see changes over next month. Any background to help understand behavior is welcome.

I would recommend to dub such posts on BOINC forums. And maybe even better - to dub to BOINC alpha mail-list too. Only then they will have chance to be read by those who actually can do smth with this situation. All you could get here is explanation why it's broken.