Strategic Double Clicking

[ATH]

Originally Posted by storm5510

It seems that Titan Black's were tailored for something different and not the obvious, gaming. In theory, as you say, the time to test this particular exponent would be much less than 30 hours. How much less, I wonder?

It takes roughly 22h for a 46M, so I do not understand how a 1080 can do it in 30h, but that's great.

[storm5510]

Originally Posted by ATH

It takes roughly 22h for a 46M, so I do not understand how a 1080 can do it in 30h, but that's great.

There's a snip below from GPU-Z while CuLu was running. Perhaps you will see something in it that stands out.

[ATH]

Your result fits well with the 1080 entry in this GPU chart:
http://www.mersenne.ca/cudalucas.php

If you look at the 1080 at 45M and mouse over the 63.1 number it says 2.2 ms/iteration and 27.5 hours for 45M, so it must be that figure 250 GFLOPS FP64 speed that is wrong.

[storm5510]

Originally Posted by ATH

Your result fits well with the 1080 entry in this GPU chart:
http://www.mersenne.ca/cudalucas.php

If you look at the 1080 at 45M and mouse over the 63.1 number it says 2.2 ms/iteration and 27.5 hours for 45M, so it must be that figure 250 GFLOPS FP64 speed that is wrong.

Interesting. I saw one thing in the GTX 1080 line that does not match what I have. TDP. It says 180W. Mine runs at 210W. I have a little gadget that I can plug into a wall outlet and plug whatever into it. It jumps up 210W when I load the 1080.

[Mark Rose]

Originally Posted by storm5510

Interesting. I saw one thing in the GTX 1080 line that does not match what I have. TDP. It says 180W. Mine runs at 210W. I have a little gadget that I can plug into a wall outlet and plug whatever into it. It jumps up 210W when I load the 1080.

Don't forget power supply inefficiencies. Yours looks to be about 86% efficient at that load level, assuming a 180W draw from the card.

[storm5510]

Originally Posted by Mark Rose

Don't forget power supply inefficiencies. Yours looks to be about 86% efficient at that load level, assuming a 180W draw from the card.

I did some searching on the web. This particular model falls into the 200 to 205 range on wattage. I don't know why. A third fan, perhaps. This PSU is bronze rated. There are better ones, I know. I've had no issues with it.

At the moment, it's loaded about as heavy as I ever allow. Prime95 and mfaktc running at the same time. The exhaust air is only slightly warm.

[S485122]

Originally Posted by LaurV

... the first test was done by Chuck Norris curtisc, who doesn't make mistakes
...

Sorry to "necropost" but it permits me to come back on the subject of this thread. Computers in the Curtisc account do have errors, I suppose some have bad hardware and then since they are student lab computers a lot of things can happen. I have seen quite a few bad results coming from that account, the curtisc computer grn209--00l, for instance, has 17 bad and 17 good results in the 20M-30M range.

Jacob

[ATH]

8 more from this thread above 80M that needs a triple check, if anyone wants to do triple checks at that range. Otherwise we'll just leave it for the future:


https://mersenne.org/M80151979
https://mersenne.org/M80490829
https://mersenne.org/M80740679
https://mersenne.org/M80793413
https://mersenne.org/M81245897
https://mersenne.org/M81390061
https://mersenne.org/M81543971
https://mersenne.org/M81611749

[Madpoo]

Originally Posted by S485122

Sorry to "necropost" but it permits me to come back on the subject of this thread. Computers in the Curtisc account do have errors, I suppose some have bad hardware and then since they are student lab computers a lot of things can happen. I have seen quite a few bad results coming from that account, the curtisc computer grn209--00l, for instance, has 17 bad and 17 good results in the 20M-30M range....

curtisc definitely has some bad results in there, but considering the total # of exponents he turns in it's not too bad by percent. Right now he has 34,725 good results and 626 bad. 1.77% of the bad+good total.

Of course he has a large amount of unknown results... 157,520. So it's hard to say if that error rate would hold true once those get double-checked.

[nofaith628]

Haven't seen this post updated for a while. Pinging Madpoo, any double checks that need churning?

[kriesel]

Hi,

As a low priority subproject, I've been filling the double-check gaps in the work distribution map at https://www.mersenne.org/primenet/. That is, where there is a range of 1 million in exponent value (which I'll call a bin) where there's a double check assignment available but none completed, I run one. The theory is that it might possibly smoke out a reliability problem at higher fft length than the current wavefront, in time to address it, since most first checks will have been done by prime95/mprime, and the double checks I run are CUDALucas 2.06 beta. I've been doing them in ascending order, since 84M in mid 2017, so that now all bins up to 111 million have at least 1 matching doublecheck indicated. (See https://www.mersenne.org/report_ll/?...exfactor=1&B1=) 112 is under way and 114 queued. Meanwhile the wavefront of new first LL assignments is ~81M-88M.

The following bins below 200M have no first-time LL test reported, so no double check assignment available. It would be good if some volunteers collectively ran first-time checks on at least one exponent in each of these million-range bins so that double checks could become available in them. In the list below, the bins are abbreviated, for example, "115" means the exponent range 115,000,000-115,999,999. Note these will definitely each be more time consuming than usual double checks (time is proportional to a bit higher than exponent squared; M112m is a little over 8 days on a GTX1080.)
115
135
136
140
142
156
161
163
166
169
171
172
173
174
175
181
182
183
184
185
186
187
190
192
193
197
198
199
(there are more above 199M, of course, which can wait.)