Formula Boinc Sprints 2018

[ao_ika_red]

And... It's amicable numbers!!!

 

[petrusbroder]

Joined ...

 

[crashtech]

Grrrr....

I won't put up with much more of these shenanigans by FB.

 

[TennesseeTony]

Good golly! Check out this user from [H]: https://sech.me/boinc/Amicable/hosts_user.php?userid=7913

 

[StefanR5R]

Relative performance per Watt:
E5-2690v4 : 1080Ti = 1 : 2.7
E5-2696v4 : 1080Ti = 1 : 2.1

 

[Ken g6]

Good golly! Check out this user from [H]: https://sech.me/boinc/Amicable/hosts_user.php?userid=7913

Looks like a miner. 8 cards on mid-range chips aren't going to get much PCIe bandwidth.

 

[lane42]

I don't know much about amicable, but his w.u. times look pretty good on his 1070Ti's for 8 on a rig.

 

[IEC]

Good golly! Check out this user from [H]: https://sech.me/boinc/Amicable/hosts_user.php?userid=7913

Impressive. Most impressive.

 

[Skillz]

He's working on opening up those lanes on those GPUs with new systems. He just doesn't know (well no one does) what the best, cheapest approach would be for getting full lanes on 8 GPUs per rig.

Maybe even 4 GPUs per rig?

Any thoughts?

 

[TennesseeTony]

I tried an ASUS workstation board once, 7 16x slots (single spaced) with a multiplexer chip on-board, and a variety of shielded extension cables (which all caused errors with more than one GPU in use.) Apparently only 3M makes extension cables good enough for multiple GPUs, but they're over $100 each for a ribbon cable.

Then I tried rigid risers, which did work until the MOBO died.

 

[[H]Coleslaw]

I'm guessing math projects probably don't need a lot of CPU time in general. Most of the work can probably be done straight off the GPU and thus needing less data lane bandwidth. Just my guess since that is kinda what cryptos are really doing most of the time.

 

[StefanR5R]

Within a series of 8-threaded tasks which generally take 2 hours on the particular host, I had one which took 16 hours. Possible explanation:

A small portion of WUs can't be parallelized to many threads. The one that you aborted had command line /from 2*5*7*11*389*31249 /to 2*5*7*11*389*31271 /task_size 1938395193 which means it could run only 4 threads in parallel (31249, 31253, 31259, 31267 as the last prime number in factorization). But such WUs are rare.

https://sech.me/boinc/Amicable/forum_thread.php?id=80

 

[ao_ika_red]

This Amicable project looks favouring Nvidia over AMD as there's quite a gap of performance between crashtech's 7970 and Ken's 1060 3gb.
And according to this list, mine is only a bit better than 4 year old 750ti.

 

[TennesseeTony]

Yep. I'm pulling my pair of R9-280Xs from the race, as they together output almost the work of a single 1080 but at 2.5x the power input.

 

[StefanR5R]

I am noticing that UK BOINC Team is active in the sprints too, with varying results.

 

[StefanR5R]

Good golly! Check out this user from [H]: https://sech.me/boinc/Amicable/hosts_user.php?userid=7913

Looks like a miner. 8 cards on mid-range chips aren't going to get much PCIe bandwidth.

I don't know much about amicable, but his w.u. times look pretty good on his 1070Ti's for 8 on a rig.

Amicable's bus utilization is below the 250 MB/s of single lane PCIe v1. CPU utilization by the GPU feeder thread is low too (at least in Linux opencl_nvidia). A GPU based mining rig with weak bus and weak CPU should indeed do well at Amicable.

 

[Orange Kid]

Day one of Amicable Numbers find us ambling along in 2nd place overall (first place for Non Profits).
SETI.USA seems to be trying to make a bit of a move on us, time will tell. The tight races are for 5~9 at the moment.
Keep counting as fast as you can, it could be tight at the end.

1 Gridcoin .................................25 ..............165,046,222
2 TeAm AnandTech .................18 ...............38,050,253
3 SETI.USA ...............................15 ................25,423,804
4 Czech National Team .............12 ................17,664,724
5 Rechenkraft.net .......................10 ................10,616,608
6 L'Alliance Francophone ............8 ..................10,117,566
7 UK BOINC Team ......................6 ...................9,974,006
8 Overclock.net ...........................4 ....................9,372,421
9 Boinc@Taiwan .........................2 ....................8,422,190
10 OcUK - Overclockers UK .......1 .....................6,555,909

 

[Skillz]

2nd place in league 1.
3rd place overall.

 

[StefanR5R]

Well; by a hair.

 

[Skillz]

A ticks hair, right?

 

[TennesseeTony]

[H] is crushing it in League Two, if y'all haven't noticed. Kinda glad now that we're NOT still in League Two.

 

[StefanR5R]

Number of active users at Amicable today, according to stats.free-dc's 16:41 BST data,
League 1:

TeAm ........ 18
SUSA ......... 12
CNT ........... 28
UBT ............. 9
RKN ........... 13
LAF ............ 29
OCN ............ 8
P3DN ........ 12​

League 2:

[H] ............. 19
XS ................ 7
EVGA ......... 12
MK ............... 5
TSBT ............ 2​

League 3:

Pirates ........ 4
Litomyšl ..... 4​

Background noise:

GRC ......... 152​

Edit, almost forgot the UK in League 1.

 

[lane42]

Thanks for the stats O.K.

 

[StefanR5R]

SETI.USA seems to be trying to make a bit of a move on us, time will tell.

Their user Rainer (https://sech.me/boinc/Amicable/show_user.php?userid=8) joined RKN 3 times yesterday (https://sech.me/boinc/Amicable/team_display.php?teamid=69). Whatever he wants to accomplish by that. Perhaps it's the next best thing to bunkering, which is harder at Amicable than at some other projects. Edit, or he wants to keep LAF behind RKN.

Spoiler: RKN team page

 

[[H]Coleslaw]

StefanR5R, I don't know him nor their plans, but I have seen teams lend their support like that numerous times across several events. With FB it works even better because not only can you maintain higher points by staying ahead, you can also keep the other team further down the list by propping up another team in the efforts. People manipulated the WCG X-mas challenge by bouncing around teams too. Since WCG's system is flawed with how they report total team contributions at the very end of the day rather than hour by hour. So, one can move their results to numerous teams throughout the day as long as they move back to the team they really want to support at the end of the day where all of the days worth of results will be tallied. Certainly some strategy involved.