Why should something with much better performance (architecture cache etc), pumping out more work lose out to this average/median thing though and never ever get a true reflection of its performance, always be losing a percentage of credits? That was'nt what it was like here when people first joined. It does'nt exactly encourage people to invest in the best and produce more work for Rosetta.

I was actually looking forward to a fair credit system, believe it or not and said so in a few threads before the new system was described.

---

So, if the new credit system awarded say... 20% more credit then it does now, to all contributors, are you (Whl) suggesting that would be better?

---

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/

No. You are just being silly now. What should be reflected is the true performance of a machine with a benchmark within the application(s) with encryption. I know the application is always changing, but that part of the code within the application(s) need never change unless new cpu's etc appear. More work I know, but the benefits to Rosetta would be huge.

---

Wasn't trying to be silly. Your point seemed to be that when people see their RAC dropping, that it doesn't encourage them to go the extra distance to help the project. If credits were increased 20%, then your RAC would probably be increasing instead, and therefore give you the incentive you suggest.

I guess the point I'm missing will be answered with the question:
What about the new credit system does not reflect "true performance"?

Or, perhaps you misunderstand the new credit system. You said:

Why should something with much better performance (architecture cache etc), pumping out more work lose out to this average/median thing?

Noone is losing out (unless you are comparing to the old system, which would have meant you should have liked the idea of using 20% more credits). If you are actually pumping out more work, then you are getting more credits.

For example, if you overclock a CPU and now it runs at 3ghz, the work you produce will be on par with a 3ghz machine, and you will get the credit of a 3ghz machine. Regardless of report specs which might indicate it's only a 2.6 ghz machine.

But if the CPU is overclocked, and your floating point calculations are still occuring at the same speed as the non-overclocked 2.6ghz system, well, since most of what Rosetta does is floating point calculations, your only going to be crunching at a rate comparable to the 2.6ghz system. And will be awarded credits that reflect that.

---

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/

Now you are being silly Feet1st. ;-)

Have you ever seen a modern CPU (not with co-processor) which has been overclocked and its FPU does'nt increase somewhat ?

No matter what you say, top end machines are being dragged down with this system.

Maybe it is you who does'nt fully undertand what is going on ? ;-)

---

No. You are just being silly now. What should be reflected is the true performance of a machine with a benchmark within the application(s) with encryption. I know the application is always changing, but that part of the code within the application(s) need never change unless new cpu's etc appear. More work I know, but the benefits to Rosetta would be huge.

The project has chosen to grant for work done, so if a benchmark is to be introduced, it has to reflect the algorithm of the application (currently rosetta 5.25), not the architecture of the CPU/memory/board config. And a new one has to be implemented in a new app. The theoretical power of a machine for other tasks, like rendering, is irrelevant for Rosetta. So if SSE or Altivec isn't used in the app, it won't get any more credit for the simple presence of these possibilities.

You won't get more points by the FIA in Formula 1 racing, because your car can climb 45� steep hills, it's irrelevant for the "work done", i.e. fast flat rounds.

So why not go the whole hog and optimise the apps as well then ?

---

What should be reflected is the true performance of a machine with a benchmark within the application(s) with encryption. I know the application is always changing, but that part of the code within the application(s) need never change unless new cpu's etc appear. More work I know, but the benefits to Rosetta would be huge.

Whl, the new system does exactly what I think you mean but without a benchmark. The problem with a benchmark is that it can never be perfectly aligned with the work being done which has been covered in detail elsewhere. The new system awards credit for work done so credit is aligned to work done. Ignore the benchmark that is part of BOINC.

The new credit system DOES award more credit for more work done. If you have a faster CPU, more CPUs, more cores, a more efficient CPU, more cache, more RAM etc, you will get more credit!

---

No matter what you say, top end machines are being dragged down with this system.

That's absolutely possible.
If the "top end" stuff is something the project cannot use, those machines will be "dragged down" with the new system. As I stated often enough: irrelevant possibilities don't matter, work done matters.

Edit:
It would of course be very fine, if those possibilities could be made usefull, i.e. that the application could be optimised for those instruction sets.
But validity is more important than speed, and if vectorisation gives wrong results, it's not useful. But I can't say anything about this except that it's possible that the use is impossible.

No matter what you say, top end machines are being dragged down with this system.

They aren't! All machines get the credit they deserve! Their RAC may be dropping, but only if it was artificially high under the old system, and so the RAC will fall to the level the computer deserves, based entirely on its turn-out of results. Overclocked (stable) computers will have a higher throughput and so receive more credit/higher RAC.

---

So why not go the whole hog and optimise the apps as well then ?

'Optimised' isn't a discrete term - you can only optimise code in comparision to another version of itself. The Rosetta code may be very well optimised already - it's been discussed many times that turning on SSE support in the compiler have made no difference to performance as the code isn't suitable, at least as it currently stands.

There seem to be a few people who assume it's a case of flicking a switch and the code will run faster. If it were, they'd do that!

---

I do understand all that Danny. I know high end machines will get more credit than less powerfull machines, but they are still being dragged down by a percentage.

I have to go meet someone for now.

---

I do understand all that Danny. I know high end machines will get more credit than less powerfull machines, but they are still being dragged down by a percentage.

They're not being dragged down by anything though ;). If a computer returns WUs twice as fast as another, it will get twice the credit. There's no averaging of performance. There is averaging of credit per decoy, but faster computers get more credits by returning more decoys, not by getting more credits per decoy.

HTH
Danny

---

I do understand all that Danny. I know high end machines will get more credit than less powerfull machines, but they are still being dragged down by a percentage.

Can you tell me how this should be possible?

Can you tell me how this should be possible?

Yes.

---

Can you tell me how this should be possible?

Yes.

So how?

So why not go the whole hog and optimise the apps as well then ?

'Optimised' isn't a discrete term - you can only optimise code in comparision to another version of itself. The Rosetta code may be very well optimised already - it's been discussed many times that turning on SSE support in the compiler have made no difference to performance as the code isn't suitable, at least as it currently stands.

There seem to be a few people who assume it's a case of flicking a switch and the code will run faster. If it were, they'd do that!

Missed this somehow.

No offence Danny, but this place is looking more and more like UD every day. People have asked for ages for things to be improved with the UD agent and have always been told it was'nt possible. WCG seem to be able to do it with the same agent though.

---

Can you tell me how this should be possible?

Yes.

So how?

Average/mean/median makes for an average bland credit system.

Off to get that sitter person and bring them here now.

---

Can you tell me how this should be possible?

Yes.

So how?

Average/mean/median makes for an average bland credit system.
Off to get that sitter person and bring them here now.

It's average per decoy.
Why should a decoy on a dual core get more credits then the same decoy on a P3?

By reading your posts, WHL, it seems to me you are calling for a credit system, which does not award credit according to the actual work done, but according to the potential work that a specific processor could achieve. This idea was first presented by Hymay here iirc, but it has some shortcomings, with the most severe one that the potential work is never to be measured accurately whereas the actual work done can be measured reasonalbe accurate.