[QUOTE=TheJudger;443782][CODE]mfaktc v0.21 (64bit built)
[...]
CUDA version info
binary compiled for CUDA 8.0
CUDA runtime version 8.0
CUDA driver version 8.0

CUDA device info
name GeForce GTX 1080
compute capability 6.1
[...]

Selftest statistics
number of tests 26192
successfull tests 26192

kernel | success | fail
-------------------+---------+-------
UNKNOWN kernel | 0 | 0
71bit_mul24 | 2586 | 0
75bit_mul32 | 2682 | 0
95bit_mul32 | 2867 | 0
barrett76_mul32 | 1096 | 0
barrett77_mul32 | 1114 | 0
barrett79_mul32 | 1153 | 0
barrett87_mul32 | 1066 | 0
barrett88_mul32 | 1069 | 0
barrett92_mul32 | 1084 | 0
75bit_mul32_gs | 2420 | 0
95bit_mul32_gs | 2597 | 0
barrett76_mul32_gs | 1079 | 0
barrett77_mul32_gs | 1096 | 0
barrett79_mul32_gs | 1130 | 0
barrett87_mul32_gs | 1044 | 0
barrett88_mul32_gs | 1047 | 0
barrett92_mul32_gs | 1062 | 0

selftest PASSED!
[/CODE]

Reference (aka Founders Edition) 1080:
[CODE]Starting trial factoring M66362159 from 2^73 to 2^74 (28.83 GHz-days)
k_min = 71160531149400
k_max = 142321062305090
Using GPU kernel "barrett76_mul32_gs"
Date Time | class Pct | time ETA | [B][COLOR="Red"]GHz-d/day[/COLOR][/B] Sieve Wait
Sep 29 13:31 | 0 0.1% | 2.345 37m29s | [B][COLOR="Red"]1106.37[/COLOR][/B] 82485 n.a.%
Sep 29 13:31 | 4 0.2% | 2.345 37m27s | [B][COLOR="Red"]1106.37[/COLOR][/B] 82485 n.a.%
Sep 29 13:31 | 9 0.3% | 2.340 37m19s | [B][COLOR="Red"]1108.73[/COLOR][/B] 82485 n.a.%
[/CODE]

I guess we've a new performance and performance per watt leader.

Oliver[/QUOTE]

Just wait till we see Pascal Titan X numbers...

Some benchmarks:

CUDA 8.0, Linux, mfaktc 0.21, M66362159 from 2[SUP]73[/SUP] to 2[SUP]74[/SUP]

reference GTX 980 Ti: ~780 GHz-d/day, ~240W (~3,25GHz equivalent per watt)
reference GTX 1070: ~800 GHz-d/day, ~130W (~6,15GHz equivalent per watt)
reference GTX 1080: ~1090 GHz-d/day, ~180W (~6,05GHz equivalent per watt)
reference TITAN X (Pascal): ~1410 GHz-d/day, ~250W (~5,64GHz equivalent per watt)
[LIST][*]all GPUs are in an environment where GPU temperature is kept below the temperature target[*]GTX 1080 and TITAN X (Pascal) are limited by TDP while the other two are below their limit[*]power measurement is done with 'nvidia-smi'[*]compared to earlier post this time performance numbers are long term instead of quick benchmark.[/LIST]
Not a real surprise: Pascal is much more energy efficient than Maxwell.

[U]Your mileage my vary, take 10 cards of same spec and you measure 10 slightly different results.[/U]

Oliver

[QUOTE=airsquirrels;443795]Just wait till we see Pascal Titan X numbers...[/QUOTE]

12 minutes :razz:

[QUOTE=TheJudger;443797]12 minutes :razz:[/QUOTE]

So with that crazy high TF number vs. LL, does this mean we should be raising the TF cutoff?

[QUOTE=TheJudger;443796]reference GTX 980 Ti: ~780 GHz-d/day, ~240W (~3,25GHz equivalent per watt)
reference GTX 1070: ~800 GHz-d/day, ~130W (~6,15GHz equivalent per watt)
reference GTX 1080: ~1090 GHz-d/day, ~180W (~6,05GHz equivalent per watt)
reference TITAN X (Pascal): ~1410 GHz-d/day, ~250W (~5,64GHz equivalent per watt)[/QUOTE]Great to hear.
Now that people can run GTX 10x0 properly, I would really appreciate a bunch of new benchmark data so I can update my performance chart :smile:
[url]http://www.mersenne.ca/mfaktc.php#benchmark[/url]

The numbers quoted by Oliver above are higher than I expect on my chart, but then I don't really have any good benchmark data to work from, just a rough guess.

This is an open call for everyone with GTX 9xx or 10xx (or newer) to submit new mfaktc benchmarks, thanks!
(CUDAlucas benchmarks for the same cards are also in demand, see the top of [url]http://www.mersenne.ca/cudalucas.php[/url])

[QUOTE=TheJudger;443796]reference GTX 1070: ~800 GHz-d/day, ~130W (~6,15GHz equivalent per watt)
reference GTX 1080: ~1090 GHz-d/day, ~180W (~6,05GHz equivalent per watt)[/QUOTE]

Is the reference GTX 1060 comparable in terms of GHz per watt? Or is it less efficient?

According to the the performance figures I have (perhaps slightly off from actual but the scaling should be similar):
GTX 1080 = 5.079 GHz-days per day per watt (914GHd/d, 180W)
GTX 1070 = 4.284 GHz-days per day per watt (643GHd/d, 150W)
GTX 1060 = 3.569 GHz-days per day per watt (428GHd/d, 120W)

You can see this yourself at [url]http://www.mersenne.ca/mfaktc.php[/url] (put "GTX 1" in the filter/search box to limit to those GPUs)

I am not sure that those numbers are completely correct. My 750Ti never reaches its full TDP with mfaktc. It only reaches 70% before it is limited by operating voltage according to gpu-z. mmff only makes it reach 50% TDP.
I would imagine that other cards will have similar issues especially the Maxwells.

[QUOTE=airsquirrels;443803]So with that crazy high TF number vs. LL, does this mean we should be raising the TF cutoff?[/QUOTE]

Just in case you missed it: 12 minutes was the difference between your post "waiting for TITAN X (Pascal) numbers" and my post. :smile:

[QUOTE=henryzz;443850]I am not sure that those numbers are completely correct. My 750Ti never reaches its full TDP with mfaktc. It only reaches 70% before it is limited by operating voltage according to gpu-z. mmff only makes it reach 50% TDP.
I would imagine that other cards will have similar issues especially the Maxwells.[/QUOTE]

Exactly. But this is not limited to Maxwell cards. IIRC most cc 2.1 didn't hit their TDP while running mfaktc. As written above the reference 1070 I've used didn't reach the TDP while the 1080 did.

[QUOTE=TheJudger;443856]Just in case you missed it: 12 minutes was the difference between your post "waiting for TITAN X (Pascal) numbers" and my post. :smile:[/QUOTE]

Caught it. Glad to see we're finally factoring using a recent CUDA version