I've yet to have anyone point out why it won't work.

That's because credit here drives people insane.

---

But the question remains...
We wave our magic wand and get a credit system that counts flops as accurately as humanly possible...
Does credit go up? Down? Stay the same?

That depends on the integrity of the human beings who define the exchange rate.

Last time we had a flop-counting credit system in use at this project, Eric introduced a 'credit multiplier' or fiddle factor to normalise that. It's still in the metadata for every task:

<credit_rate>2.8499999</credit_rate>

(that was in a BLC file)

Eric, given a free hand and plenty of time, would make an honest effort to set that to honour the cobblestone definition. When he gives a public speech to a scientific audience (the audio from a talk he gave to a NASA seminar was released a few years ago), the figures he gives for total project throughput bear no relation to the ones made up by CreditNew.

To answer your question, the general consensus is that the current credit rates at this project are below cobblestone parity, so accurate setting of the exchange rate would raise them. But you still couldn't buy a toaster with them.

Sales blurb, that's all, sales blurb - that is the ONLY real use of such benchmarks.
You still REFUSE to understand that ANY general purpose benchmark is ONLY in INDICATOR to performance under the very specific set of conditions upon which it is based. Nobody I know ( and I know a very substantial number), who are involved in developing time-critical software rely on Linpac as it is general in its nature, not tailored to the work they are doing. I probably contributed data to the sources referred to in those two sources (and many more).

If you don't like the truth then set up a little experiment, write a program to preform a simultaneous, synchronised FFT, while tuning a three term controller and monitoring about 32 analogue inputs. You have a 3.3ms loop, and can only reliably carry internal, incomplete data forward one loop cycle. And do this on something no faster than the original IBM PC. And then compare what you see in terms of assembler instruction set usage with that in the Linpac (or indeed Whetstone) benchmarks. You will realise that the benchmark mix of assembler instructions is not representative of the operations that you actually use, and, depending on the processor selected it may be a substantially faster, or substantially slower set of instructions.

If you don't understand what I just suggested you do, then you don't understand a thing about the application of benchmarks in real-world situations.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

But the question remains...
We wave our magic wand and get a credit system that counts flops as accurately as humanly possible...

Counting FLOPs accurately??? Don't make me laugh, well not in anything approaching a sensible time. They are too dependent on the exact processor and processor instruction set combination being use for the particular task.

Does credit go up? Down? Stay the same?

That all depends on the scaling factors that are applied, and those are, as others have explained, a decision for the project admin staff to apply.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Ok Alex, I'll let you in on a wintertime experiment of mine.

Last winter this machine attained a RAC of 15K after 2 weeks while this year at the same time it's only got to 12K (still using the same apps and settings as last year, but likely more AP's have been processed so far this winter) and after 8 weeks it had hit 22K, but I doubt that it will get that far this year let alone reach an average of 23K before being shut down again.

Cheers.

@Wiggo

Contrary to popular belief I understand both your frustration and your situation.
Yes your machines are producing less credit (apples to apples) than V6.
And yes, your now lower credit (RAC) is bouncing around horribly.
These are facts.

But I also understand a few of the reasons this is happening and they have nothing to do with CreditNew.
I can quickly provide you with three, and then maybe one of the gurus or someone more articulate than myself can explain in further detail (and maybe add a couple more):

Reason #1
Our GPU apps are slower. CPUs haven't struggled as much with the change to new apps & tasks and mix of task types. This has nothing to do with CreditNew.
The GPUs have been asked to perform a task and choked.
Take a quick look at Shaggie's fantastic graphs and check out the 560TIs. Jason's used to do at least 25,000 a day (max, but fairly consistently).
What are those cards doing now?
Our knee-jerk reaction is to blame CreditNew for this. But if we want to - for example - crunch VLARS on GPUs, then we have to be adults and take the hit. Not blame it on CN.

Reason #2
AstroPulse is an outlier. It always has been.
But back in the V6 days, supply was fairly consistent (Arecibo only remember). So it wouldn't cause RAC to fluctuate.
But now Astro is being dumped in bursts. This causes RAC to fluctuate. There's nothing wrong with that.
But again, we have to be adults about it and "suck it". We can't blame this on CreditNew either.

Reason #3
Right before we changed to V7 we were all using "illegal" Intel apps.
But those apps were literally twice as fast as stock.
They contained optimizations that were Intel's intellectual property and were not allowed to be distributed openly. This - I am proud to say - we DID take like men. Hardly anybody complained.

But again. Not something we can blame on CreditNew.

Let's look at how absurd the current situation is. Note that when My 1060 runs against a faster GPU it scores much fewer points than when run against a 'slower' GPU. The 1060 should be scoring the same credit in the following tasks;

   Task     Computer          Sent                  Time reported                 Status         Run-time  CPU-time  Credit                    Application
6672672779  6796479  27 May 2018, 7:03:26 UTC  27 May 2018, 13:47:28 UTC  Completed and validated  202.49   195.21   61.78  SETI@home v8 Anonymous platform (NVIDIA GPU)
6672672780  7475713  27 May 2018, 7:03:27 UTC  27 May 2018, 13:31:39 UTC  Completed and validated   67.26    67.16   61.78  SETI@home v8 Anonymous platform (NVIDIA GPU)
http://setiathome.berkeley.edu/workunit.php?wuid=2989824826

6672672962  6796479  27 May 2018, 7:03:26 UTC  27 May 2018, 13:57:48 UTC  Completed and validated  202.51   194.26   45.51  SETI@home v8 Anonymous platform (NVIDIA GPU)
6672672963  7475713  27 May 2018, 7:03:27 UTC  27 May 2018, 13:41:56 UTC  Completed and validated   38.55    38.73   45.51  SETI@home v8 Anonymous platform (NVIDIA GPU)
http://setiathome.berkeley.edu/workunit.php?wuid=2989824853

6672672728  6796479  27 May 2018, 7:03:26 UTC  27 May 2018, 13:47:28 UTC  Completed and validated  204.48   197.21   61.96  SETI@home v8 Anonymous platform (NVIDIA GPU)
6672672729  7475713  27 May 2018, 7:03:27 UTC  27 May 2018, 13:36:48 UTC  Completed and validated   67.62    67.35   61.96  SETI@home v8 Anonymous platform (NVIDIA GPU)
http://setiathome.berkeley.edu/workunit.php?wuid=2989824870

6672672726  6796479  27 May 2018, 7:03:26 UTC  27 May 2018, 14:08:07 UTC  Completed and validated  205.43   197.56   44.72  SETI@home v8 Anonymous platform (NVIDIA GPU)
6672672727  7475713  27 May 2018, 7:03:27 UTC  27 May 2018, 13:41:56 UTC  Completed and validated   38.39    38.76   44.72  SETI@home v8 Anonymous platform (NVIDIA GPU)
http://setiathome.berkeley.edu/workunit.php?wuid=2989824864

There are many more where those came from.

Alex - which part of Wiggo's statement "still using the same apps and settings as last year" are you unable to understand?
He is running THE SAME computer
He is running the SAME applications
And yet YOU say he's changed application, or GPU or something else.

The ONLY thing that has changed in the last twelve moths are the effective (miss)calibration constants on the servers used to calculate the credit awarded for each task - something almost everyone else understands to have happened, and many complain about.

Let's consider you "reasons", one at a time.

Reason #1:
GPUs vs CPUs - Let's be generous to you.... GPUs do choke, or at least used to. But only when faced with a particular subset of tasks, and only then a particular subset of GPUs were affected. With the latest applications this situation appears to have been resolved.

Reason #2:
AstroPulse is NOT an "outlier", it just has a different set of internal measurement points, and the servers just aren't capable of dealing with such a situation correctly.

Reason #3:
As for "illegal applications" - You are an absolute comedian, and obviously do not understand what you are saying. Trying to keep a very complex legal situation simple:
(Long?) After the stock, and some of the optimised applications were compiles Intel changed its conditions of use, which conflicted with the GPL license conditions (under which both SETI & BOINC were released). After much legal debate between the GPL community (of which BOINC and SETI are members) and Intel a period of grace was granted to give the GPL community time to re-compile their applications using other compilers (of non-Intel origin). There was also agreement that "existing" applications could continue to be distributed (in the exact state they were on the "action date"), but any new or modified applications either had to pay a substantial license fee, or face legal action, or go through the BA of recompiling using other (GPL compliant) compilers. SETI & BOINC chose the last of those options as they knew there would be changes to both applications and they didn't want to line either Intel's, or their lawyers', pocket.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Mr. Smith, there is nothing I have ever said that you agree with.
Almost every single reply of yours has been "Alex, bla-bla-bla YOU ARE WRONG!"
Half those times you haven't understood what I'm saying and then go on to say the same thing I'm saying.
The other half, you again haven't understood what I'm saying and go off on a tangent.

More math and fewer bla-bla-bla personal attacks please.
Because one of these days I'm going to wake up on the wrong side of bed.

Perhaps it is about time you looked and read what others have said about the current situation with CreditNew before you make threats to me. Then you will see how consistently wrong in your assumptions and assertions you are.

If you cannot, or will not, (I don't mind which it is) understand the maths that underlie NewCredit then don't argue with those who do understand what is going on with it these days - which I will admit have moved on considerably in impact during the years since you were last active, and suffered the "Zimbabwe Bank blast" from Richard, for doing exactly the same as you are doing today. - ignoring facts for your version of the truth.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Yeah something was certainly lost there in translation as nothing has changed from this winter to last (except for those VLAR's going to GPU's), but it will be interesting to see what the results are for that rig a year on.

Cheers.

Good evening Mr Phelps.

For the past decade it seems that a most promising & worthwhile project has been, year on year, been providing diminishing returns.
Your mission should you accept is to return the Boinc Project to worthwhile aims.

We suggest that you ignore the CreditNew debacle within the project & just focus on un-borking Boinc itself.

If successful, your secondary mission is to "persuade" the project leaders from their continued misguided attempts to "force" all projects to use their "specific" template.

Should you fail, we will disavow your existence.

This post will self destruct in 5 seconds...

and suffered the "Zimbabwe Bank blast" from Richard

Aaaaaaaaaaaaaaaaahahahahaha!!! I'm crying :')
You reeeally need to learn how to read. LOL.
Seriously. Worst example you could have come up with.
(And unlike some people I know, Richard is actually capable of learning mid-conflict)

Without going back to check, here's what happened:
I was trying to convince anybody who would listen that credit is a flopcounter.
Then I wanted to say that it's cute we have our OCDs about credit bouncing around but that's not why the credit system needs fixing ASAP...

Alex: Credit is a flopcounter
Richard: No, no, no. They used to be. Not anymore.
Alex: Credits are Cobblestones, and Cobblestones are flops. Ergo flopcounter. (And this is true to this day)

At this point I thought a lightbulb would go off and Richard would say "Holy crap, we are STILL declaring flops!".
(Lightbulb doesn't go off but Richard more than makes up for it later)

Instead I get "The Bank of Zimbabwe speech". It's a beautiful speech and you guys should read it. The essence of the speech is spot on.
However it was a tangent to the point I was trying to make. And I would have let "the tangent" slide except...
Richard made a mistake. A mistake that was critical to the point I was trying to make. So unfortunately I had to call him out on it.

Richard: Credit is useless. CreditNew/Boinc is issuing credit that has lost all value and meaning. A credit here is NOT the same as a credit there. It's like we're pretending a US dollar is the same as a Zimbabwean dollar!!!
Alex: Yes I know. Except CreditNew isn't a bank. The projects have their own banks. And credit is still being declared in flops.

(And somewhere here a lightbulb goes off and Richard gets 3 steps ahead of the conversation and straight to the point I was trying to make)

Richard: OMG we are still declaring flops!!! That means projects can be accused of FRAUD!
Alex: (Smiles sardonically, proud that finally one of the gurus understands that one of these days we are going to get caught with our pants down)

Ahaha! Rob got me so worked up I just went from memory.
BIG mistake!
I totally mixed up two different threads above!

Funny thing is, later on I'm saying "fraud is a bit harsh".

I have no idea what I was thinking. Can't figure it out.
Maybe trying to instinctively defend SETI or something.

Rob can you please just put me on ignore?
Save us both the drama...

Alex on a more upbeat note, your beautiful daughter must be about two, how are you dealing with the terrible twos?

---

Note that when My 1060 runs against a faster GPU it scores much fewer points than when run against a 'slower' GPU.

I've been wanting to reply to this all day, but got completely side-tracked. Now I don't have any energy left.

This is a really good catch TBar.
Appears to be 100% CreditNew as well.
Wow.

Alex on a more upbeat note, your beautiful daughter must be about two, how are you dealing with the terrible twos?

She's the star of the show :) She's doing great so... so are we! Thank you so much for asking :)

Just got to 22 months on the 26th. And yeah, I think we are slowly entering 'terrible twos" territory... she's become a bit more demanding.
But I can't recognize the signs, she's my first :) I have no experience with "terrible twos".

What are you guys using now that PhotoBucket isn't really an option?
Might be time to spam the café with pics :)

Richard: OMG we are still declaring flops!!! That means projects can be accused of FRAUD!
Alex: (Smiles sardonically, proud that finally one of the gurus understands that one of these days we are going to get caught with our pants down)

Actually, no.

The projects are declaring credit.
BOINC central is declaring flops.

There, on the BOINC home page.

24-hour average: 18.366 PetaFLOPS.
Active: 161,917 volunteers, 848,356 computers.

Those are Zimbabwean flops, because the projects are declaring Zimbabwean credits.

BOINC, the organisation that provides the software that we all use, used to be funded by the NSF as a computer science experiment. It was under development. That phase of its life came to an end about three years ago, and the funding came to an end. We've been running on fumes ever since. I don't know whether the NSF has a programme to fund mature infrastructure, and if it does, whether BOINC applied to it. Instead, David Anderson (personally, but in conjunction with other universities - not Berkeley) applied for replacement funding to the NSF, and was successful. The application is online, and has been since 1 June 2017.

The summary version

Volunteer computing (VC) uses consumer devices such as PCs and smartphones for scientific
computing. Currently 500,000 devices participate, with 2.3 million CPU cores and 290,000 GPUs. VC
supplies 44 PetaFLOPS of computing power, and it could provide ExaFLOPS.

The full proposal

500,000 devices are actively participating in VC projects, 70% via BOINC. These devices have about 2.3
million CPU cores and 290,000 GPUs, and collectively provide an average throughput of 93 PetaFLOPS.
The computers are primarily modern, high-end machines: they average 16.5 CPU GigaFLOPS and 11.4
GB of RAM. 73% have a GPU capable of general-purpose computing with OpenCL or CUDA.

So, how many PetaFLOPS of scientific research is BOINC providing? 18.366, 44, 93? If we were counting them, we'd get the same answer every time?

That goes far beyond SETI, which is why I haven't explicitly called it out here - although I've been worrying about it for almost a year. Those flops are not counted, and we all know projects where they are vastly inflated (I participate in one myself). The documentation talks in language implying scientific veracity, and it's obviously fooled you, but...

Alex - you've been told , from the "horse's mouth", that you have misinterpreted what Richard said. You regularly (deliberately?) misquote or misunderstand others that I have to doubt that you actually recognise the truth even when it hits you in the face.
I hope that one day you accept that BOINC, the core of the SETI system has a defective credit system, and there is basically no way that the concept can work for projects that adopt the "adaptive" model (one of four available for all BOINC projects to select from).




For information the four credit granting models within BOINC are:
Pre-assigned - can come in two flavours, a crude and simple "every validated task will get x credits" or a more complex "estimate the effort in advance and thus the credit due on validation". Both these are independent of the user and thus rigidly consistent in terms of credit awarded:

Post-assigned - Similar to above, but relies on a property of the user, thus may be "corruptible" if the user doesn't do what is expected (e.g. uses a highly parallel application instead of a single thread one):

Runtime credit - This really sounds so simple, but is so fraught with issues. If the project only (directly) supports CPU applications it may work OK, but, as has been identified by David Anderson that there are issues when working in a heterogeneous world of processors:

Adaptive credit - Sounds great in theory, but has many pitfalls. In this model there is constant "re-calibration" of each client and processor. It is this model that is deployed by SETI, and we've been discussing the pitfalls off and on almost since its first deployment on SETI all those years back....

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

I could have sworn that somebody in this thread asked about Eric's speech to NASA, where he mentioned the total collective throughput of SETI@Home in TFlops, but I can't find it now. So I can't reply directly to whoever it was, but I looked up the references anyway. (Edit - it was a PM. They've just reminded me. The answer's here, anyway.)

Friday March 21 [2014] SETI talks to be webcast live
Search for Life Beyond the Solar System Conference

The recording of Eric's talk is still accessible - I listened to it again this morning. Scroll down to the end for the link - Eric got the graveyard slot, after lunch on Friday for a 5-day conference.

Our discussions moved, as they should, from news to number crunching. I reported Eric's figures in message 1493304, where I said:

If you listen to Eric's talk (link in first post - at about 8 minutes in), he is very careful to describe his 237 TFLOP claim as "actual" computing power, and to distinguish it from

... theoretical computing power, which is what a lot of people report in their distributed projects.

Given the nature of the talk he was giving, on the record, and the sort of people he would expect to be in the audience, I suspect that those were very, very, carefully chosen words.

I'd also like to re-draw attention to my message 1493516 a little later in the same thread.

Edit 2: the closest I can get to the CN claim for that date, via https://web.archive.org/web/20140327064934/https://boincstats.com/en/stats/0/project/detail/, is 676.278 TeraFLOPS for 27 March 2014 - that was another question asked in the PM.