Chouette, on reprend !

Thank you for your work! Nice to see the project up and running again!

I would like to give some feedback on what i think a lot of users would like to see:

- Switching back to the fixed 480 credit system
- ARM64 Application
- CUDA Application Update to support newer GPUs

Thanks and happy crunching

Spectrum

Actually the project is now using the most resent Credit System, provided by Boinc.
There are many reasons why project is switching to it. In contrary to the old one, the new system is more stable, more accurate, more fair and more bulletproof to cheating.
Here https://boinc.berkeley.edu/trac/wiki/CreditNew you can find some details if you are interesting.

Cheers,
Georgi

---

“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson

Actually the project is now using the most resent Credit System, provided by Boinc.
There are many reasons why project is switching to it. In contrary to the old one, the new system is more stable, more accurate, more fair and more bulletproof to cheating.
Here https://boinc.berkeley.edu/trac/wiki/CreditNew you can find some details if you are interesting.

Cheers,
Georgi

The Credit New system and it's ease for cheating is the reason LHC@Home was dropped by Gridcoin with a significant loss of computing resources for the project. Your previous system, with fixed credit for validated work was far "more fair and bulletproof to cheating". The one and only reason any project uses Credit New is because it's the default.
All you have to do to abuse Credit New is keep creating new clients. You don't need new machines, just wipe the old client and start fresh. It's trivial.

The Credit New system and it's ease for cheating is the reason LHC@Home was dropped by Gridcoin with a significant loss of computing resources for the project. Your previous system, with fixed credit for validated work was far "more fair and bulletproof to cheating". The one and only reason any project uses Credit New is because it's the default.
All you have to do to abuse Credit New is keep creating new clients. You don't need new machines, just wipe the old client and start fresh. It's trivial.

I fail to see how creating a new client will benefit anyone who intends to cheat. none of the factors in the credit new system take into account host age

there are however some cases of inconsistent credit award depending on the validating systems.

example: https://asteroidsathome.net/boinc/workunit.php?wuid=149859374

with the new app, it seems if the two devices are both Ampere cards, they get about 50x the normal credit. but if either of these devices validates against a CPU or older GPU it gets the normal credit amount.

why should credit award change this much based on who happens to be your wingman?

this is another argument to just go back to the static credit scheme. then all tasks are awarded the same no matter what device crunches it. faster systems will earn more credit based on rate of work completed and not by the luck of the draw with who your wingman is. this is undoubtedly a more fair system.

---

there are however some cases of inconsistent credit award depending on the validating systems.

example: https://asteroidsathome.net/boinc/workunit.php?wuid=149859374

with the new app, it seems if the two devices are both Ampere cards, they get about 50x the normal credit. but if either of these devices validates against a CPU or older GPU it gets the normal credit amount.

why should credit award change this much based on who happens to be your wingman?

this is another argument to just go back to the static credit scheme. then all tasks are awarded the same no matter what device crunches it. faster systems will earn more credit based on rate of work completed and not by the luck of the draw with who your wingman is. this is undoubtedly a more fair system.

I don't understand why someone gets 50 times the credit for minimum working hours, when others are given credit according to the length of working hours.

there are however some cases of inconsistent credit award depending on the validating systems.

example: https://asteroidsathome.net/boinc/workunit.php?wuid=149859374

with the new app, it seems if the two devices are both Ampere cards, they get about 50x the normal credit. but if either of these devices validates against a CPU or older GPU it gets the normal credit amount.

why should credit award change this much based on who happens to be your wingman?

this is another argument to just go back to the static credit scheme. then all tasks are awarded the same no matter what device crunches it. faster systems will earn more credit based on rate of work completed and not by the luck of the draw with who your wingman is. this is undoubtedly a more fair system.

I don't understand why someone gets 50 times the credit for minimum working hours, when others are given credit according to the length of working hours.

it's not just based on "working hours" it's based on flops too. faster devices will complete tasks faster so it needs to be scaled accordingly. but there's obviously some problem happening here with the flops on fast GPUs like Ampere. the reported flops is way higher than previous gens.

it's fair that a GPU would get the SAME credit as the CPU. since both devices did the same amount of work. the GPU just did it faster. but what's not fair is the untended bonus that the CN system applies when two very fast systems validate against each other, presumably the same amount of work that a CPU does (since all the tasks here seem similar in size with some variance) but with lower reward.

static reward fixes these issues.

---

it's fair that a GPU would get the SAME credit as the CPU. since both devices did the same amount of work. the GPU just did it faster. but what's not fair is the untended bonus that the CN system applies when two very fast systems validate against each other, presumably the same amount of work that a CPU does (since all the tasks here seem similar in size with some variance) but with lower reward.

static reward fixes these issues.[/quote]

I agree with the GPU and CPU getting the same job, even if the GPU can do it faster. I just wanted to point that out
that for the same unit and it doesn't matter how quickly someone makes it, someone collects 70 credits and the other over 6000 credits. The old way rewards were fairer.

it's not just based on "working hours" it's based on flops too. faster devices will complete tasks faster so it needs to be scaled accordingly. but there's obviously some problem happening here with the flops on fast GPUs like Ampere. the reported flops is way higher than previous gens.

it's fair that a GPU would get the SAME credit as the CPU. since both devices did the same amount of work. the GPU just did it faster. but what's not fair is the untended bonus that the CN system applies when two very fast systems validate against each other, presumably the same amount of work that a CPU does (since all the tasks here seem similar in size with some variance) but with lower reward.

static reward fixes these issues.

No need for anything fancy. A task is a certain size and gets a certain amount of credit whenever it's completed and however long that takes. Just like me paying you $20 to mow my lawn. I don't care if you do it fast with a ride on mower, or slow with a push mower, you get paid the same amount when you complete it.

You do know that they are using 'credit new' now and are set on that at least for the short term...right? 'Credit New' is pushed by the Boinc Developers and other senior level tech people and they seem to be the people the Project is listening too right now.

I am excited that Asteroids@home is back, but I am only limited to CPU tasks since another project will retire a GPU app in the next few weeks.

I like the new look too!!! I some how double posted, so I thought I change this one.

By the way, if you edit your post to two spaces and nothing else, it will vanish. Weird way to do things, but it seems to work on all project forums.

Another Boinc thing.

---

“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson

Another Boinc thing.

You missed out the phrases "cobbled together" and "badly designed". I spend more time making Boinc behave than maintaining my 10 computers. The other day I had a computer download 150 days of work instead of 1. Apparently because "it hadn't learned how long the tasks took yet", even though it was actually overestimating the time, not under.

I'm not or maybe I'm missing a lot. But that is intentionally. Our team is facing Boinc related troubles almost every day. Sometimes I'm able to patch the code behind, sometimes not. It is highly time-consuming process, and the thing is that I don't have that needed spare time on dally basis. Not to mention that it is a never-ending learning curve in reverse engineering stile. Aside all bad taste, the Boinc project is one powerful platform, and we must admit that. There are better one out there for sure. Boinc is not perfect, and it might suffer from the an enormous in size Bad Practice Sickness for years already (having 234 separate branches in their repository, where most of which are there just to serve different teams requirements, while at the same time the most important 'master' branch is full of bugs), but it is what it is. And I'll do my best to keep our forked version at its best possible condition knowing that I'm facing a "vast unknown".

Edit:
Just to be clear, when I'm saing that I'm patching the code, I mean only the web part. Not the core. The core is and will alwais be the one provided by Boinc.

---

“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson