Are these minirosetta applications always going to grant me such horrible credit? I feel like I am wasting my time crunching them.

---

My hosts seem to be granted about half of claim on the mini's as well.

this host doesn't seem to have anything but minis to compare to.

this host has a 5.98 task that got 117 credit granted for a 198 claim. And one that got 130 granted for a 203 claim. So that is a similarly poor ratio between mini and 5.98.

This host used to be granted credit roughly in-line with claims.

My Linux host seems to be granted credit roughly inline, or over the claim on the minis, but 5.98 granted significantly over claim (claim 55, granted 86), and 48 claim and 72 granted.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

Feet1st wrote:

My hosts seem to be granted about half of claim on the mini's as well. this host doesn't seem to have anything but minis to compare to.

My P4/HT/3GHz computer is much like your 715317 host but I'm claiming half of what you are, and my claim and grant are similar. So why are your claims so high?

So why are your claims so high?

I run with a 24hr runtime preference. So, you should see that my claim is similar to yours if you also take in to account the number of seconds of CPU time.

No, I do not run a modified or optimized client. Nor alter my benchmarks.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

So why are your claims so high?

I run with a 24hr runtime preference. So, you should see that my claim is similar to yours if you also take in to account the number of seconds of CPU time.

No, I do not run a modified or optimized client. Nor alter my benchmarks.

Not accusing you of anything. Then it must be related to the number of decoys you produce in 24 hours compared to the number I produce in 10 hours? I don't know how to do a apples-to-apples me/you task comparison, do you? Moments ago I increased by runtime from 10 to 14 hours, just so you know.

<edit> Like I said, don't know what I'm doing but I randomly grabbed one of your tasks and one of mine and you had 23 decoys (in 24 hours) and I had 20 decoys in 10 hours. Someone brighter than me needs to glean something from this.

is one with HT and one without?

---

is one with HT and one without?

Feet1st Using Boinc v5.10.45
NBIT - P4/3GHZ/HT/SP3 using Boinc version 5.10.13

Maybe certain tasks simply hit dead ends and produce fewer decoys based on the starting seed? Though it shouldn't be this way for every task. Got me, I just attach here and let the computer rip; have no clue about how the Rosetta app works.

Interesting. I've got two "identical" hosts. Both run 24x7, one runs Ralph when work is available.

Host 1 RAC is 275, this is the one that runs Ralph, P4 HT active.
Host 2 RAC is 171, P4, I just changed it to not hyperthread today.

I note that Host 1 receives credit basically inline with claims. Host 2 receives roughly half the claim. The claims per hour of CPU time are roughly identical.

Host 1 runs BOINC 5.10.20
Host 2 runs BOINC 5.10.45

hmmmmm. What BOINC version are you folks running?

The way the credit system works, if one host is consistently getting significantly less then claimed credit, there must be hosts elsewhere that are getting consistently more then claim. Has anyone spotted any hosts getting more then their claims for task crunched by mini? Perhaps duos and Quads?

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

Interesting. I've got two "identical" hosts. Both run 24x7, one runs Ralph when work is available.

Host 1 RAC is 275, this is the one that runs Ralph, P4 HT active.
Host 2 RAC is 171, P4, I just changed it to not hyperthread today.

I note that Host 1 receives credit basically inline with claims. Host 2 receives roughly half the claim. The claims per hour of CPU time are roughly identical.

Host 1 runs BOINC 5.10.20
Host 2 runs BOINC 5.10.45

hmmmmm. What BOINC version are you folks running?

The way the credit system works, if one host is consistently getting significantly less then claimed credit, there must be hosts elsewhere that are getting consistently more then claim. Has anyone spotted any hosts getting more then their claims for task crunched by mini? Perhaps duos and Quads?

perhaps you should clarify the above. I get higher than claimed, but then I am OC'd near the max that my system can take. So if you compare a intel duo core against my machine for credit and that same system is not OC'd then it would be hard to compare. look at this task for instance.
4 hour run time on a OC'd system results in Claimed credit 86.1512475967147
Granted credit 152.481392741213. this is running boinc 6.2.18 with mini 1.32

The way the credit system works, if one host is consistently getting significantly less then claimed credit, there must be hosts elsewhere that are getting consistently more then claim. Has anyone spotted any hosts getting more then their claims for task crunched by mini? Perhaps duos and Quads?

Yes I have, on my Intel(R) Pentium(R) D CPU 3.20GHz (PIV 940D,dual core, @ 3.3 GHz).
My Genuine Intel(R) CPU T2400 @ 1.83GHz gets roughly what it claims.
My 3 Intel(R) Core(TM)2 Duo CPU's E4500 @ 2.20GHz get significantly less credit.

My Pentium D is the only one that has 2MB L2 cache per core (the other cpu's only have 1MB L2 per core). I wonder if this has anything to do with it.

---

Feet1st:

<snip> I note that Host 1 receives credit basically inline with claims. Host 2 receives roughly half the claim. The claims per hour of CPU time are roughly identical.

Host 1 runs BOINC 5.10.20
Host 2 runs BOINC 5.10.45

hmmmmm. What BOINC version are you folks running?

Hmm, strange...doesn't seem to be much interest in this thread/topic. Doesn't the credit claimed factor into the running average that is ultimately used to determine what is granted? So if there are benchmarking discrepancies amongst various versions of BOINC (with respect to HT vs non-HT, single core vs multiple core, for example) these discrepancies would impact the running average adversely.

So if there are benchmarking discrepancies amongst various versions of BOINC (with respect to HT vs non-HT, single core vs multiple core, for example) these discrepancies would impact the running average adversely.

Correct, although perhaps the effect is not "adverse" for some machines. So it doesn't explain why so many are reporting low credit, but it could cause higher or lower claims. So for those of you comparing your claim to the credit granted, to be an indication of a problem, it is possible the benchmarks changed. Any time you change the benchmarks, the similarity they have to running actual Rosetta work will be altered, and probably vary by CPU type as well. Does anyone know the specifics of a benchmark change?

Perhaps the workload of new benchmarks is not as close to the actual Rosetta work as it used to be. For example, historically, the benchmark has never measured the CPU's ability to use cache memory, because the benchmark program was to small. But the reality of running Rosetta is that it will run much better on a machine with an L2 cache of a good size. And yet, many machines with large L2 cache also have higher clock speeds, and so benchmark higher as well.

---

Rosetta Moderator: Mod.Sense

So for those of you comparing your claim to the credit granted, to be an indication of a problem, it is possible the benchmarks changed. Any time you change the benchmarks, the similarity they have to running actual Rosetta work will be altered, and probably vary by CPU type as well. Does anyone know the specifics of a benchmark change?

I think benchmarking has changed as versions have changed. But not everyone has upgraded BOINC version. I and many others are still running 5.10.x versions, while many have upgraded to the 6.x versions. So the mass migration upgrade to 6.x may be impacting the credit/claim issue? And the BOINC upgrade would somewhat coincide with Rosetta migrating to mini-rosetta as the main application. Someone like Ageless/Jord might be able to answer the benchmarking question.

Could someone with technical savvy review this BOINC forum thread and see if it is relevant to this topic?
Benchmarks lower with 6.2.14

Could someone with technical savvy review this BOINC forum thread and see if it is relevant to this topic?
Benchmarks lower with 6.2.14

Hmmm, they are talking about other version of the client, it is for Linux and it is a 64 bit client. I don't think it is related to the subject of this thread.

Disclaimer:
I don't claim to be technically savvy. :)

Definitly a Linux thread.
My Disclaimer: I at least know what Ubuntu is. Even if it sounds like some sort of Austrailian bird.

OK, so what do you make of this? I changed this host to no longer use the hyperthreading. It's claim to granted ratio has improved markedly. I have another host with more L2 cache, and it sometimes gets more credit granted then claimed.

I am starting to conclude that the mini application is significantly more dependant on large L2 cache to run well.

And so I would predict that the hosts that consistently get more credit then claimed are those with newer CPUs that have larger L2 cache... OR, are not running multiple Rosetta WUs at the same time.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

Yep, found a host with much better credit numbers. It has 4 CPUs, and 4MB L2 cache and pulls a RAC of 1567. Here is another, 4 CPUs, 6MB L2 cache, RAC 1525.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

OK, so what do you make of this? I changed this host to no longer use the hyperthreading. It's claim to granted ratio has improved markedly.

I see your benchmarks now are 1597/3043 while my virtually identical machine with HT on has 1311/2395. Do you know what your benchmarks were before HT was turned off? Would a higher integer benchmark improve your claim?

Also you and I have 1 MB L2 cache though mine is reported as half that because HT is on. Still I get fairly equal claims/grants vs your claims/grants. With all your new theories about L2 cache how do you explain the differences between your claims/grants and mine, even though we have virtually identical machines? We both have Windows XP though yours is professional, I use Boinc 5.10.13 while you use 5.10.45...I don't know, just fishing and not willing to accept the L2 cache theory yet, not that my opinion matters.

I was just trying to compare my machine with itself. You can see that since I turned HT off, I'm getting much better credit. Roughly double what was granted before... but then my 24hrs of CPU time now only results in one completed task too. So it is interesting that my claims are still roughly 200 credits per 24hrs, just as they were previously, but I'm now granted 150 or more credits which is roughly double what I was getting with HT on.

Sorry, no I didn't note my benchmarks prior to switching off HT.

Yes, I've heard of a few cases where cache size is not reported properly. I don't know all the details, so it makes comparison difficult.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

I was just trying to compare my machine with itself. You can see that since I turned HT off, I'm getting much better credit. Roughly double what was granted before... but then my 24hrs of CPU time now only results in one completed task too. So it is interesting that my claims are still roughly 200 credits per 24hrs, just as they were previously, but I'm now granted 150 or more credits which is roughly double what I was getting with HT on.

Sorry, no I didn't note my benchmarks prior to switching off HT.

Yes, I've heard of a few cases where cache size is not reported properly. I don't know all the details, so it makes comparison difficult.

You're a long-time Rosetta cruncher so I don't presume to know more than you by any stretch. However, referring to your little experiment of comparing your machine to itself... HT is supposed to increase your throughput by maybe 20% over non-HT (who knows). Logically it seems to me you have a single core and L2 cache that is shared by two threads when HT is active so each thread can only produce roughly half (ignoring the 20%) as many models in a 24-hour period that a single task with HT off can produce in the same period. Does that make sense? The bottom line is how many models do you produce in 24 hours with HT on versus how many with HT off? That determines what you are granted per day.

With respect to claimed credit that number just averages into all the other claims provided by other participants to determine what someone else is granted for their completed tasks. From that perspective you cannot reap the benefit directly from what you claim.

Anyway I thought you were trying to compare mini-Rosetta to the elder v5.98?