I have noticed no change in my RAC. In another project when things were changed, you noticed a large change in RAC, quickly. Since this one has been running for several days, and now most of my machines are on S5, I am still getting the same RAC.
I believe that they have the numbers running well. No one else is complaining of loss of RAC, that I have seen, either.
Actaully Pooh Bear I have seen a drop in my per hour credit return. But expected this as the WU's are longer, and the S-5 App is still in its first rollouts. It will get better and the return will go back up. Most of my Machines on S-4 Akos were returning around 20 credits per hour. Now they are returning around 8 credits per hour. Time and further optimizations by the Devolpment team will pull that back up for all of us. At least here there is a staff that talks to you. Trys to explain what is going on. And is open to suggestions. Unlike another project a lot of us know.
Here's my RAC ( I luv the 5.4.9 stats pane! ):
You'll note the plummet - it's from the aircon, UPS, breakers etc ... outages. Also the S4 is running down and S5 ramping up, my home farm looks like this ( BoincView screenshot )
which indicates I'm well into the Hanford S5 thumpers, so the effect of that ( accounting for my quorum partner(s) in those specific analyses ) before validation and granting of actual credit, and then funneling that into the RAC will probably take a 'while'. I'll see. But it's all really going into the bank though! :-)
Cheers, Mike.
(edit) 'Est. credits' is calculated on work done so far for a given WU. May I recommend BoincView by the way, an excellent tool even if you don't have a big farm!
(edit) 'baby' is an old baby AT board machine, the others are mostly P4's at 3GHz with HT, 'walkies' a laptop.
(edit) Also 'Elizabeth' is my daughter's machine ( I named her after it! Or was it the other way around? ), and doesn't get left alone much during school holidays! :-)
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
It hasn’t been reported yet, but my first “faster� WU took 3.2 hours with v4.05 on my G4/733. Since its name includes “0384.0� it appears to be a “short� one. BOINC Manager is now estimating 4.3 hours each for the rest of the batch.
Time goes up from 3500 Sec to 26700....Celeron M (Dothan)
Fun, credit and science....i work with 2 computers, a Laptop with a 1.7 P-M and my privat computer with the celi. I like to see at least one Wu/Day on my laptop, if i have to wait 2 or 3 Days and see no realy progress it is boring - sorry, thats what i think.
Ok, so with my "puny" 800 Mhz machine my current S5 WU is showing a bit over 62% done and a total completion time of 28 hours and 25 minutes and increasing!!
I haven't done any s5 on my other computer but I'm *patiently waiting*. I'm interested to see how this works out and even though my RAC has plunged straight to the bottom of a very deep hole; I'm totally doing this for the science not the credits. My girlfriend thinks I'm a geek, I'm pretty sure she is right, good thing too...
So im getting a 23.9% improvemnt in crunch times using HT.
This tell me a lot.
Hi Akosf... I am confused somewhat by your reply. Are you saying that the S5 app should run faster single threaded vs. HT ? Like U41.05 or similar? Being able to run HT efficiently across multiple projects is a concern I have as at times I run more than just E@H and would like to carry the HT advantage across the board.
So im getting a 23.9% improvemnt in crunch times using HT.
This tell me a lot.
Hi Akosf... I am confused somewhat by your reply. Are you saying that the S5 app should run faster single threaded vs. HT ? Like U41.05 or similar? Being able to run HT efficiently across multiple projects is a concern I have as at times I run more than just E@H and would like to carry the HT advantage across the board.
what he's saying is that the S5 app has a decent ammount of room for improvement. The 25% gain from HT means that the stock app is only using at most ~80% of the CPUs capacity. Rewriting hte inner loops to hit 100% like his s4 optimized apps do would completely eliminate the HT advantage.
So im getting a 23.9% improvemnt in crunch times using HT.
This tell me a lot.
Hi Akosf... I am confused somewhat by your reply. Are you saying that the S5 app should run faster single threaded vs. HT ? Like U41.05 or similar? Being able to run HT efficiently across multiple projects is a concern I have as at times I run more than just E@H and would like to carry the HT advantage across the board.
what he's saying is that the S5 app has a decent ammount of room for improvement. The 25% gain from HT means that the stock app is only using at most ~80% of the CPUs capacity. Rewriting hte inner loops to hit 100% like his s4 optimized apps do would completely eliminate the HT advantage.
Does that mean that lets say Crunchr3's app for Seti is not fully optimized? As HT with his app's are still faster. Or are we talking apples and oranges here? That Einstein will always be different in this regard than all the other Boinc projects
what he's saying is that the S5 app has a decent ammount of room for improvement. The 25% gain from HT means that the stock app is only using at most ~80% of the CPUs capacity. Rewriting hte inner loops to hit 100% like his s4 optimized apps do would completely eliminate the HT advantage.
Does that mean that lets say Crunchr3's app for Seti is not fully optimized? As HT with his app's are still faster. Or are we talking apples and oranges here? That Einstein will always be different in this regard than all the other Boinc projects
Yes and no. If the app has greater throughput with HT it's not using a full 100% of the CPU capacity. That doesn't guarantee that there's room for a speedup though. I don't know what seti does under the hood, so I can't comment on which class it falls into.
Speaking for a strictly hypothetical project/system combination, so don't complain if there's no meaningful explanation for the behavior (a meaningful exmaple would be much more complex to construct).
The project runs a loop where it does 5 individually independent calculations (A, B, C, D, E), but needs all 5 results before it can begin doing anything in the next iteration.
If the CPU can do a single calculation at a time the runing of two iterations of the program loop would look like this, and use 100% of the cpu capacity, leaving no potential gain for HT:
A1
B1
C1
D1
E1
A2
B2
C2
D2
E2
If a second CPU can do 5 calculations at a time the two iterations would look like this and again use 100% of the CPU's capacity, again leaving no potential gain for HT.
A1 B1 C1 D1 E1
A2 B2 C2 D2 E2
IF you run the app on a machine than can do 3 calulations at once though what you get is this (XX = do nothing), and only use 83% of the CPU capacity. Remember that the 1st step of iteration 2 can't start until all 5 steps of 1 are finished.
A1 B1 C1
D1 E1 XX
A2 B2 C2
D2 E2 XX
Short of rewriting the application in a way that would allow a more flexible ordering of operations (which may or may not be possible) there's no way to use the last 17% of the CPU capacity. WIth HT you can however (V...Z are the 2nd thread's steps)
In this case, 5 single steps of a 2nd thread were slotted into the 5 empty spaces left by the first one. After 10 cycles the 1st thread has ran 5 interations of the loop, and the 2nd 1 iteration. A real HT system wouldn't give a preference to one thread over the other, and would have both running at euqal rates (excluding noise factors).
RE: RE: I have noticed no
)
Here's my RAC ( I luv the 5.4.9 stats pane! ):
You'll note the plummet - it's from the aircon, UPS, breakers etc ... outages. Also the S4 is running down and S5 ramping up, my home farm looks like this ( BoincView screenshot )
which indicates I'm well into the Hanford S5 thumpers, so the effect of that ( accounting for my quorum partner(s) in those specific analyses ) before validation and granting of actual credit, and then funneling that into the RAC will probably take a 'while'. I'll see. But it's all really going into the bank though! :-)
Cheers, Mike.
(edit) 'Est. credits' is calculated on work done so far for a given WU. May I recommend BoincView by the way, an excellent tool even if you don't have a big farm!
(edit) 'baby' is an old baby AT board machine, the others are mostly P4's at 3GHz with HT, 'walkies' a laptop.
(edit) Also 'Elizabeth' is my daughter's machine ( I named her after it! Or was it the other way around? ), and doesn't get left alone much during school holidays! :-)
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
It hasn’t been reported
)
It hasn’t been reported yet, but my first “faster� WU took 3.2 hours with v4.05 on my G4/733. Since its name includes “0384.0� it appears to be a “short� one. BOINC Manager is now estimating 4.3 hours each for the rest of the batch.
Time goes up from 3500 Sec to
)
Time goes up from 3500 Sec to 26700....Celeron M (Dothan)
Fun, credit and science....i work with 2 computers, a Laptop with a 1.7 P-M and my privat computer with the celi. I like to see at least one Wu/Day on my laptop, if i have to wait 2 or 3 Days and see no realy progress it is boring - sorry, thats what i think.
Regards
Ouch...I just got my first S5
)
Ouch...I just got my first S5 WU...
Two and a half hours later and its crunched 15%...
Oh well...
What's the deal with optimized apps, still indevelopment for s5?
Human Stupidity Is Infinite...
Ok, so with my "puny" 800 Mhz
)
Ok, so with my "puny" 800 Mhz machine my current S5 WU is showing a bit over 62% done and a total completion time of 28 hours and 25 minutes and increasing!!
I haven't done any s5 on my other computer but I'm *patiently waiting*. I'm interested to see how this works out and even though my RAC has plunged straight to the bottom of a very deep hole; I'm totally doing this for the science not the credits. My girlfriend thinks I'm a geek, I'm pretty sure she is right, good thing too...
Tim
RE: Ouch...I just got my
)
The current apps has about 50% of the speedup akos got with s4. The work units are just alot bigger now.
RE: I found a good
)
Hi Akosf... I am confused somewhat by your reply. Are you saying that the S5 app should run faster single threaded vs. HT ? Like U41.05 or similar? Being able to run HT efficiently across multiple projects is a concern I have as at times I run more than just E@H and would like to carry the HT advantage across the board.
RE: RE: I found a good
)
what he's saying is that the S5 app has a decent ammount of room for improvement. The 25% gain from HT means that the stock app is only using at most ~80% of the CPUs capacity. Rewriting hte inner loops to hit 100% like his s4 optimized apps do would completely eliminate the HT advantage.
RE: RE: RE: I found a
)
Does that mean that lets say Crunchr3's app for Seti is not fully optimized? As HT with his app's are still faster. Or are we talking apples and oranges here? That Einstein will always be different in this regard than all the other Boinc projects
RE: RE: what he's saying
)
Yes and no. If the app has greater throughput with HT it's not using a full 100% of the CPU capacity. That doesn't guarantee that there's room for a speedup though. I don't know what seti does under the hood, so I can't comment on which class it falls into.
Speaking for a strictly hypothetical project/system combination, so don't complain if there's no meaningful explanation for the behavior (a meaningful exmaple would be much more complex to construct).
The project runs a loop where it does 5 individually independent calculations (A, B, C, D, E), but needs all 5 results before it can begin doing anything in the next iteration.
If the CPU can do a single calculation at a time the runing of two iterations of the program loop would look like this, and use 100% of the cpu capacity, leaving no potential gain for HT:
A1
B1
C1
D1
E1
A2
B2
C2
D2
E2
If a second CPU can do 5 calculations at a time the two iterations would look like this and again use 100% of the CPU's capacity, again leaving no potential gain for HT.
A1 B1 C1 D1 E1
A2 B2 C2 D2 E2
IF you run the app on a machine than can do 3 calulations at once though what you get is this (XX = do nothing), and only use 83% of the CPU capacity. Remember that the 1st step of iteration 2 can't start until all 5 steps of 1 are finished.
A1 B1 C1
D1 E1 XX
A2 B2 C2
D2 E2 XX
Short of rewriting the application in a way that would allow a more flexible ordering of operations (which may or may not be possible) there's no way to use the last 17% of the CPU capacity. WIth HT you can however (V...Z are the 2nd thread's steps)
A1 B1 C1
D1 E1 V1
A2 B2 C2
D2 E2 W1
A3 B3 C3
D3 E3 X1
A4 B4 C4
D4 E4 Y1
A5 B5 C5
D5 E5 Z1
In this case, 5 single steps of a 2nd thread were slotted into the 5 empty spaces left by the first one. After 10 cycles the 1st thread has ran 5 interations of the loop, and the 2nd 1 iteration. A real HT system wouldn't give a preference to one thread over the other, and would have both running at euqal rates (excluding noise factors).