Posts by Partygott

1) Message boards : Number crunching : less credits for doing more work? (Message 1099)
Posted 27 Jul 2017 by Partygott

Hi. Thx for apply.

Ok, if credit new is used here in an untouched way, than this explains all.
Because credit new punishes high performers while it gives bonus to low performers. And it is vulnerable for turbo clocks and overclocking.
Because of that, most projects use their own credit system or they use a rebalanced system based on credit new.

This means a Pentium 1 @ 100 MHz gets much more credits per wu than a 7700k @ 5 GHz. Credit new says that the Pentium needs much more effort than the 7700k for crunching one wu and the Pentium owner should be animated to be part of Boinc.
But nowadays...

Let's get more practical...
I took a host with a Ryzen 1700 and compared it to a host with a Ryzen 1800X.
Both are crunching 24/7. Both crunch 16 wu simultaneously. Both use the same OS. Both use the same optimized app.
The 1800X runs stock, the 1700 is fixed @ 3,8 GHz allcore, which is faster than the 1800X allcore turbo but slower than the 1800X single core turbo.

benchmarks (rounded):
1800X >> 4,5k / 16k (min. 4,0 GHz)
1700 >> 4,2k / 15k (3,8 GHz)

average time per wu (rounded):
1800X >> 4200s (max. 3,6 GHz)
1700 >> 3900s (3,8 GHz)

RAC (rounded):
1800X: 13k
1700: 10k

Ok, the 1700 does more wu per time but it gets less credits for them.
Why is this? It's because credit new "thinks" that the 1800X invests more effort in crunching. It has higher benchmark results and needs longer to complete 1 wu, so it does more work and so it must get more credits for it.
But the truth is that the 1700 does more work - but credit new does not know.

2nd example:
Use any kind of pc.
Crunch 100 wu with see2 app and after that, do it again with the optimized see2 app.
What you will see? You will see that you will get less credits per wu with the optimized app. Because you are faster but your benchmarks didn't change.
So in the mind of credit new, you invested less effort.
You will only get a higher RAC if your optimized app can more than compensate the minus in credits with the saved time per wu. In the worst case, you are 10% faster but get 10% less credits per wu, so your RAC will not change.
(This is why there were also "optimized clients" when the first optimized apps for SETI were released. They optimized the app AND the internal benchmark app.)

3rd example:
Use any kind of pc.
Crunch 100 wu @ stock clocks. Then do some serious overclocking, do the benchmarks, set it all back to stock and again crunch 100 wu.
Guess what will happen. You will get more credits per wu than before. Credit new "thinks" you did more effort because your benches are higher, but you never did.

Sure, credit new also compares several results from the same wu to estimate if a result seems to be plausible and the differences are less or more, but the tendency is clear and was proven in countless tests.
This is why projects use rebalanced credit systems.
(Don't get me wrong...it's nice to see if a project works reliable!)

2) Message boards : Number crunching : less credits for doing more work? (Message 1095)
Posted 22 Jul 2017 by Partygott

Hi.

I did some comparison between my cpu and other cpus from the same line (same specs, only different clocks). And I found a host which has slightly higher benchmark numbers but needs about 400sec longer per wu than my host.
This should be because of the different singlecore and allcore turbo clocks.

But why does this other host gets around 1-3 credits more per wu than mine, which results in several thousand credits difference in rac?
For me this means that I get less credits for doing more work.
(Sorry for not understandig how exactly the credits are calculated on this project).

greets

3) Message boards : Number crunching : Optimization (Message 1077)
Posted 29 May 2017 by Partygott

I tested 3x the same 16 WUs on a Ryzen 1700 @ 3,8 GHz with SMT on.

avx: 28:30 - 28:48
sse: 28:56 - 29:14
fma: 29:43 - 30:01
standard: ~35min.

i5-3570 (3,2 GHz) standard: ~28min.

So there is a difference of ~18% per WU on the Ryzen, what is pretty nice.
But I wonder why the Ryzen isn't faster than the old i5, even with the optimized app and 600 MHz more clock speed. I saw this also on other Projects.
Seems to be that most of the apps from all projects are optimized for Intel.