FormulaBOINC

[iwajabitw]

I'm not running any CPU tasks on my dual 980's or dual 280x's. The 280x"s doing 2 tasks are about 18-19 minutes and running cooler than when crunching Milkyway.

 

[crashtech]

I'm not running any CPU tasks on my dual 980's or dual 280x's. The 280x"s doing 2 tasks are about 18-19 minutes and running cooler than when crunching Milkyway.

I thought 27 minutes @ 3 concurrent might be low for my 280x, but it is not out of line with your results.

 

[iwajabitw]

I thought 27 minutes @ 3 concurrent might be low for my 280x, but it is not out of line with your results.

Sounds about right, I may bump to 3 tasks when I get home later.

 

[StefanR5R]

My CPU speeds:

Xeon E3 v3, 3.6 GHz, 8 threads: 09h35m (34,500 s; 20 tasks/day)

dual Xeon E5 v4, 3.2 GHz, 2x28 threads: 10h22m (37,300 s; 130 tasks/day)​

i7-6950X, 4.0 GHz, 14 threads (besides 6 GPU feeders), 7h45m (24,300 s; 50 tasks/day)​

My GPU speeds, one task at a time:

Quadro K2100M, 405 MHz, 99 % load: 6h20m (22,800 s; 3.8 tasks/day)
+ i7-4900MQ, 3.6-3.7 GHz, 1 core fully loaded

FirePro W7000, 950 MHz, 95 % load: 20m51s (1,251 s; 69 tasks/day)
+ i7-4960X, 4.5 GHz, 1 core ca. 10% loaded

GTX 1070, 2025 MHz, 86 % load-avg.: 10m44s (644 s; 134 tasks/day)
+ i7-4960X, 4.5 GHz, 1 core fully loaded

GTX 1080, 1873 MHz, 80 % load-avg.: 9m45s (585 s; 148 tasks/day)
+ i7-6950X, 4.0 GHz, 1 core fully loaded

GTX 1080Ti, 1847 MHz, 70 % load-avg.: 7m18s (438 s; 197 tasks/day)
+ i7-6950X, 4.0 GHz, 1 core fully loaded
​

GPUs with two tasks at a time:

GTX 1070, 2037 MHz, 98 % load-avg.: 19m07s (1147 s; 151 tasks/day, 12 % better)
+ i7-4960X, 4.5 GHz, 2 cores fully loaded

GTX 1080, 1873 MHz, 96 % load-avg.: 16m34s (994 s; 174 tasks/day, 17 % better)
+ i7-6950X, 4.0 GHz, 2 cores fully loaded

GTX 1080Ti, 1847 MHz, 95 % load-avg.: 11m47s (707 s; 244 tasks/day, 24 % better)
+ i7-6950X, 4.0 GHz, 2 cores fully loaded
​

GTX 1070 GPU clock differs marginally between single- and dual-task sessions due to raised fan speed for dual-tasks.

Interesting how each of my dual Xeon boxes achieves 97 % throughput of the GTX 1070, which is in the same ballpark as the ratio of their PPD at Folding@Home. CPU tasks only give 1/5 of the credits of GPU tasks. Hence PPD of a dual Xeon is less than 1/5 of that of a GTX 1070. This is in contrast to Folding@Home where the dual Xeons reach PPDs comparable to one GTX 1070.

I'm not running any CPU tasks on my dual 980's or dual 280x's. The 280x"s doing 2 tasks are about 18-19 minutes and running cooler than when crunching Milkyway.

I thought 27 minutes @ 3 concurrent might be low for my 280x, but it is not out of line with your results.

i.e. 160 tasks/day, 6 % better than GTX 1070.

Edit April 7:
Added GTX 1080 and 1080Ti, corrected Xeon/1070 comparison.
Edit April 8:
Corrected 6950X runtime.

 

[TennesseeTony]

Just made it home and I have released the little birdies from their cages. I hope the winds of the internet are favorable and that they find their way home, safe and sound, so that they may multiply.....by a factor of 3465 points each.

Wow, so you guys took first place without me? And in all three leagues too! Sweet! Way to go TeAm!

Performance note: 3 concurrent tasks on my GTX1070 (main system, Mjolnir) is too much, 100% utilization of both cards, and the system is very sluggish. Looks like 2 per card is best for the 1070's if you need to use your system, and possibly for output too, they were finishing at right about 30minutes each.

 

[yodap]

I'm afraid to ask how may Tony so I won't. Just wanted to say you guys are kicking it and it's fun to watch. Now I have to release my measly 150 or so.
Have fun!

 

[iwajabitw]

Wow Tony, Can't wait to see the boost the team is about to get!

 

[crashtech]

What's interesting to me is how well the R9 290 is doing, my guess is that Einstein has a blend of SP and DP calculations that favors the 290's 8:1 ratio.

 

[iwajabitw]

What's interesting to me is how well the R9 290 is doing, my guess is that Einstein has a blend of SP and DP calculations that favors the 290's 8:1 ratio.

The way I figured it after reading the Einstein forums, is that its a SP project, but if you have a DP card, the task doesn't have to check back with the CPU as much because a calculation gets processed on the GPU since it has DP. So your 290 should be faster because of it being better at SP. Now, I know there is a lot of technical stuff that explains all that but that's what I took away from researching it.

 

[uallas5]

Just checked on the machine with the K200's and they're doing units in 8 HOURS each not 8 mins...OOOPS

Oh well, they're slow but its still points for the TeAm!! At least the 280X's are making a decent contribution.

 

[crashtech]

The way I figured it after reading the Einstein forums, is that its a SP project, but if you have a DP card, the task doesn't have to check back with the CPU as much because a calculation gets processed on the GPU since it has DP. So your 290 should be faster because of it being better at SP. Now, I know there is a lot of technical stuff that explains all that but that's what I took away from researching it.

I'll have to look again, but my R9 Fury (16:1) is not doing as well as the 290, that made me think the DP part was pretty significant.

 

[TennesseeTony]

Only Milkyway remains the one true DP project it would seem. Other's benefit from a high DP output, but it seems the fabled R9-280X is now only REALLY good at MW, and mining currency, fair to mid-range in other projects.

I doubt it's due to my bunker (possible though), but the TeAm has jumped 4M points. And we still lead all three leagues, but the #2 in our league likewise had a big jump (3M maybe?).

 

[iwajabitw]

Only Milkyway remains the one true DP project it would seem. Other's benefit from a high DP output, but it seems the fabled R9-280X is now only REALLY good at MW, and mining currency, fair to mid-range in other projects.

I doubt it's due to my bunker (possible though), but the TeAm has jumped 4M points. And we still lead all three leagues, but the #2 in our league likewise had a big jump (3M maybe?).

Watching there thread at Overclock.net, the posted an "Oh Snap" after you uploaded, LOL!

 

[iwajabitw]

I'm glad the Formula website is getting our results posted faster than Free-DC. With Einstein it seems like results take forever to update vs other projects.

 

[Ken g6]

Watching there thread at Overclock.net, the posted an "Oh Snap" after you uploaded, LOL!

Got a link for the rest of us?

 

[GLeeM]

I doubt it's due to my bunker (possible though), but the TeAm has jumped 4M points.

I'm in this Sprint.
That might be mostly you. I dumped a little before you and my score only changed by about 440,000.

 

[iwajabitw]

Got a link for the rest of us?

Hang on, finally got home.

Page 11

http://www.overclock.net/t/1620003/formula-boinc/100#post_25991856

 

[Orange Kid]

Got a link for the rest of us?

Link

EDIT: Guess I was a little slow in finding it

 

[iwajabitw]

I'm guessing I had about a million points bunkered. Free-DC has me stuck at 9.6mill, and that was before the sprint started. Einstein web account says 10.9mill, so I had somewhere in between bunkered.

 

[StefanR5R]

dual Xeon E5 v4, 3.2 GHz, 2x28 threads: 10h22m (37,300 s; 130 tasks/day)
GTX 1070, 2025 MHz, 86 % load-avg.: 10m44s (644 s; 134 tasks/day)
[...]
Interesting how each of my dual Xeon boxes achieves 97 % throughput of the GTX 1070, which is in the same ballpark as the ratio of their PPD at Folding@Home.

Err, nope. The E@H CPU tasks earn 693 points/task, whereas the GPU tasks earn 3465 points/task. Looks like they packed 5 jobs into one GPU WU.

 

[iwajabitw]

Free-DC finally updated.

http://stats.free-dc.org/stats.php?page=team&proj=eah&team=12

 

[Finrond]

Watching there thread at Overclock.net, the posted an "Oh Snap" after you uploaded, LOL!

Hey that was me!

Just wanted to drop in and wish you all well on this sprint, you are truly worthy competitors.

 

[TennesseeTony]

I was expecting more points today....but I checked my pending tasks, and found some of the missing points. I previously had ~450 pending, that has now nearly doubled to 880, (which is good for ~3M points total, ~1.5M from the bunker).

This particular GPU race has occurred at a good time for me, as I've been acquiring 1080's off ebay, to replace my other Nvidia cards with. But I also still have those other cards at the moment. So I've got four 1080's, three 1070's, and three 280X's, and no CPU in this race.

Thank goodness the weather has gone nuts and it's winter again here, so I can excuse myself for creating some otherwise unneeded heat. Thankfully I've already sold off the real heaters, the 980Ti's.

 

[crashtech]

All I have to add at this point is that it became clear that the G1620 Celeron in one of my little GTX 1060 Folding boxen was seriously hindering GPU performance in Einstein, though it had not been an impediment to Folding previous to this sprint. But I happened to have an i3-2100 originally meant for another box that, once installed, served to bring this little 1060 in line with its brethren.

Link to my current observations.

 

[TennesseeTony]

Uh oh.

Pretty amazing to see where we stand though, overall. Look at all those sleeping giants we are outperforming.