Am I reading too much into LTTs silence re F@H/BOINC?

[Hughesy26]

I used to fold regularly, and I'm still 14th in the team, but I'm starting to lose a little faith and looking for some reassurance from fellow folders.

 

I guess my big question is; Why do Linus and Luke (basically) never refer to F@H/BOINC?

 

It's like they do a bazillion benchmarks when a new cpu/gpu comes out, but they don't even acknowledge distributed computing. I would have thought even non-folders would be more interested in how a new piece of hardware does in scientific computations than say, 7-zip or the nth Cinebench test.

 

I get that the company may well decide not to fold 24/7 with all of their amazing hardware as those kind of electric costs would hurt the company. But the fact they are not even mentioned suggests, to me, Linus + co fundamentally think it's a waste of time. Unless I am missing something?  

[dany_boy]

Because most people just don't care?

[TheRandomness]

Missing the fact that you can learn everything you need to know about F@H/BOINC through the forum's FAQ and other sources (like the official webpage itself).

[malon]

I bet it is more of a time restrain like you mention and the fact that they want to be first with reviews of new cpu/ gpu releases than that they don't care. Also the fact that(i guesse) that there still would be to little intrest in those results might be a big factor.

[Imakuni]

1 hour ago, Hughesy26 said:

I guess my big question is; Why do Linus and Luke (basically) never refer to F@H/BOINC?

There are a couple of reasons for that:

1. Most people just wouldn't care. At. All. Even if a few of us do, we're the minority: spending resourcers to please us often would be a bad business move. They barely have the time to do their thing (or even properly, for that matter), let alone bother with ours.
2. It's an area that they don't know much about. You can't really talk about something you don't understand all that well, can you?
3. It's too broad of a thing. While F@H is a bit more specific, have you seen how many projects are encompassed within BOINC? It would be impossible to cover it in a meaningful way.
4. Benchmarking DC projects is a nightmare. For starters, they are always changing their software left and right, which makes long term benching not a thing. Second, there's an absolute lack of any sort of benchmark tool in the vast majority of the cases for some easy, automated way to get numbers. Third, running a single WU could take many hours, a time they don't have. And last, runtimes can vary wildly between different WUs, so you'd need to make them consistent by using a single WU repeated multiple times via disk mirrorring (as you can't get the same WU again on a different card), which is a pain. In other words, it's not for the faint of heart, it's not even for the strong of heart - it's for the crazy of heart!

I hope this gives you some insight. While I do think there are a couple of ways they could include DC projects in their videos once in a while (@Slickremember that video where you put a bunch of PCs hidden away, crunching useless data just to get dust?), I really don't expect them to actually bother.

[Hughesy26]

 I get the time restraint point with bleeding edge products, as they need to get their initial reviews out asap. However, they are always saying they are looking for ideas for material so I'm still not really sure why it doesn't come up in non time critical videos.

 

I don't really get the point about information being available elsewhere. The same could be said for the vast majority of topics LTT cover. Of course there are alternative (and comprehensive!) sources of info on folding/boinc, but the same is also true for most of the benchmarks LTT ever run.

 

...which brings me back to people not caring. If LTT valued the work the DC projects do, surely they would plug them more, in order to raise interest? These guys get to say (to a certain extent!) what is cool - with that many subs they are trend setters - and I guess I would like to see them be more evangelical about the goodness of folding. Unless of course they think it's a waste of time (huge electricty demands, supercomputers able to do the work more efficiently, or some other reason?) I which case I would want to hear that too!

[Hughesy26]

Thanks Imakuni - just read your response. I guess it was that 1 year fan config/dust collection video that got me thinking they just don't care. I take your points though and I guess I hadn't considered the nightmare in getting accurate results; which is of course paramount to their integrity

[KuJoe]

My electric bill is too high for me to even consider Folding, I'm guessing this is why a lot of people don't care about it.

[Imakuni]

1 minute ago, Hughesy26 said:

Thanks Imakuni - just read your response. I guess it was that 1 year fan config/dust collection video that got me thinking they just don't care. I take your points though and I guess I hadn't considered the nightmare in getting accurate results; which is of course paramount to their integrity

Once, I actually got my University to hook me an 8C/16t Xeon + 1 Quadro machine so that I could benchmark the usefulness of HT in a couple projects... it was then that I realized how impossible it was to benchmark without access to the actual apps. In the end, I had to abandon that little research, for the data I was getting was just very innacurate (as I described).

[dany_boy]

3 minutes ago, Hughesy26 said:

 I get the time restraint point with bleeding edge products, as they need to get their initial reviews out asap. However, they are always saying they are looking for ideas for material so I'm still not really sure why it doesn't come up in non time critical videos.

 

I don't really get the point about information being available elsewhere. The same could be said for the vast majority of topics LTT cover. Of course there are alternative (and comprehensive!) sources of info on folding/boinc, but the same is also true for most of the benchmarks LTT ever run.

 

...which brings me back to people not caring. If LTT valued the work the DC projects do, surely they would plug them more, in order to raise interest? These guys get to say (to a certain extent!) what is cool - with that many subs they are trend setters - and I guess I would like to see them be more evangelical about the goodness of folding. Unless of course they think it's a waste of time (huge electricty demands, supercomputers able to do the work more efficiently, or some other reason?) I which case I would want to hear that too!

Oh were it such a tender world were people cared about science and technology that could drive us forward. But its not. producer-consumer relationships are way more complicated than what you put forward (just study micro-economics and you'll know), meaning LTT can't simply dictate what is popular and what is not. And that leads back to the main point: most people just don't care

[Imakuni]

13 minutes ago, KuJoe said:

My electric bill is too high for me to even consider Folding, I'm guessing this is why a lot of people don't care about it.

We'd love to have you on BOINC. There are a bunch of projects which only use the CPU (WCG comes to mind here), meaning you can still contribute without having as much impact on your power bill at the end of the month.

[tobben]

I think not advertising = doesn't care, is an

unwarranted assumption. That said, i don't

know why they seem to have become so

silent about it, they(mostly Luke) have been

vocal about it in the past. 

 

As for benchmarks, as some have already

mentioned, there are frequent driver problems

and other irregularities, making any sort of

consistency between benchmarks pretty difficult.  

 

[Hughesy26]

Just noticed arstechnia included FAHbench in their 1080ti review

 

https://arstechnica.co.uk/gadgets/2017/03/nvidia-gtx-1080-ti-review/

[kangk81]

I think it's because the WUs in F@H differs in terms of processing time required. AFAIK there is no single benchmarking WU that one can run yet. I have been running my 3 GTX980 with 12 threads 5960X for about 3 months now and I still see my projected PPD fluctuate from high 900k to low 1.3mil. So in order to get any meaningful data will probably require them to dedicate and entire day or 2 to run F@H and collect the data based on what WUs were processed and how fast they were completed during the test period. Which the F@H client does not provide currently and doing it manually will tie up LTT's resources.

 

Not to mention that sometimes when GPUs get driver updates they break something in F@H.

[tobben]

@kangk81 As mentioned above though, there

is the FAHbench as one possible option for

reasonable consistency. Admittedly i haven't 

used it myself, and i don't know how it correlates

to actual folding performance.

https://folding.stanford.edu/home/download-utilities/

[kangk81]

11 hours ago, tobben said:

@kangk81 As mentioned above though, there

is the FAHbench as one possible option for

reasonable consistency. Admittedly i haven't 

used it myself, and i don't know how it correlates

to actual folding performance.

https://folding.stanford.edu/home/download-utilities/

i tried running it but Avast keeps flagging it as suspicious. Seeing that it is not actually hosted on the F@H Stanford site, I have my reservations on its integrity.

 

I might be over-sensitive on this but I think its better to play it safe.

[tobben]

9 hours ago, kangk81 said:

i tried running it but Avast keeps flagging it as suspicious. Seeing that it is not actually hosted on the F@H Stanford site, I have my reservations on its integrity.

 

I might be over-sensitive on this but I think its better to play it safe.

Github is a fairly trusted resource as far as i

am aware, nothing is 100% foolproof though. 

Might be better safe than sorry. 

 

[kangk81]

On 02/04/2017 at 0:48 AM, tobben said:

Github is a fairly trusted resource as far as i

am aware, nothing is 100% foolproof though. 

Might be better safe than sorry. 

 

I created an exception on Avast since Malwarebytes didn't flag the program(I have both running concurrently)

 

These are the results I got. Can't really make heads or tails with the data since the results were not presented in PPD like the F@H main program.

 

CPU compute on 5960X

 

GPU compute on GTX980 ref

All other settings were on default.

 

 

[Whaler_99]

Just to point out a couple of things. The LMG crew never directly supported the Compute teams. Yes, it has been mentioned on the WAN show, a few builds were featured and Luke did do that one video. His dad was the one into Boinc which was why Luke was into it a bit. But overall there has never been any official support of the teams from LMG.

 

I created the original Folding team and the primary team on Boincstats back in the day. The Compute teams have always been and will always be community driven. It's the LTT community working together under the LTT banner to support and help the various groups and projects. Together we make a difference.

 

In regards to bench marking I know Anadtech at one time, maybe they still do, added a few synthetic test scores to their reviews, but overall, as mentioned, just not enough people care and more over, it's to bloody inaccurate. A "standard" testing solution would be fine, if we all got those same WU's, work projects, whatever. In regards to Folding, there are so many variables in WU's and projects that LMG (or any other source) might show a Folding score of X, but what the hell does that mean? Especially when you plug in your card and get every value but that.  It just doesn't work well nor really mean much in the long run, unlike many of the other benchmarks they show.

 

[AlexG.Tech]

I find BOINC's (AVX) workloads at 100% Usage + 100% CPU Time is a good way to test if an overclock is stable! I use BOINC + CPU-Z's stress CPU + Intel ETU Stability testing all at once. It can find an unstable overclock FAST!