Hardware Benchmark Jest Thread for 100M exponents

[kriesel]

Originally Posted by retina

Hehe, still no one can beat my time.

Depends on whether phones are fair game, and whether an actual run timing is required. At 2.04 trillion years /332.2M= estimated 6141. years per iteration, confirmation with a completed iteration would be a while coming.

[retina]

Originally Posted by kriesel

Depends on whether phones are fair game, and whether an actual run timing is required. At 2.04 trillion years /332.2M= estimated 6141. years per iteration, confirmation with a completed iteration would be a while coming.

Pfft, estimates! No. Show me a real test timing or I won't believe ya.

[PhilF]

Originally Posted by retina

Pfft, estimates! No. Show me a real test timing or I won't believe ya.

Ok, I've got a stop watch and I'm ready to smash your record. :)

[retina]

Originally Posted by PhilF

Ok, I've got a stop watch and I'm ready to smash your record. :)

I'd love to see the timing for the first iteration.

[petrw1]

Originally Posted by PhilF

Ok, I've got a stop watch and I'm ready to smash your record. :)

True story:

Reminds me of when my mother-in-law decided to learn to type.
After lots and lots of practice she asked her son to time here for a typing test.
He said: "OK" then went into the next room and brought out a calendar.

[M344587487]

A Samsung S7 Edge has an Exynos 8890 SoC (the global version does anyway), which is an 8 core big.LITTLE ARM device. Only 6 cores are exposed to the Linux environment, but there are references to cores 6 and 7 in some of the kernels virtual filesystem. I think all 8 cores are used but a few are disabled in a weird way for load balancing. Phone performance is pretty sensitive to heat, this test had a 12cm PC fan blowing directly on the back, no additional heatsink and the phone was not dismantled. Here's the /proc/cpuinfo:

processor    : 0
Features    : fp asimd evtstrm aes pmull sha1 sha2 crc32
CPU implementer    : 0x41
CPU architecture: 8
CPU variant    : 0x0
CPU part    : 0xd03
CPU revision    : 4

processor    : 1
Features    : fp asimd evtstrm aes pmull sha1 sha2 crc32
CPU implementer    : 0x41
CPU architecture: 8
CPU variant    : 0x0
CPU part    : 0xd03
CPU revision    : 4

processor    : 2
Features    : fp asimd evtstrm aes pmull sha1 sha2 crc32
CPU implementer    : 0x41
CPU architecture: 8
CPU variant    : 0x0
CPU part    : 0xd03
CPU revision    : 4

processor    : 3
Features    : fp asimd evtstrm aes pmull sha1 sha2 crc32
CPU implementer    : 0x41
CPU architecture: 8
CPU variant    : 0x0
CPU part    : 0xd03
CPU revision    : 4

processor    : 4
Features    : fp asimd evtstrm aes pmull sha1 sha2 crc32
CPU implementer    : 0x53
CPU architecture: 8
CPU variant    : 0x1
CPU part    : 0x001
CPU revision    : 1

processor    : 5
Features    : fp asimd evtstrm aes pmull sha1 sha2 crc32
CPU implementer    : 0x53
CPU architecture: 8
CPU variant    : 0x1
CPU part    : 0x001
CPU revision    : 1

Here's simultaneous runs of FFT 18432K on cores 0,1,2,3 and 4,5 respectively. Cores 0,1,2,3:

INFO: no restart file found...starting run from scratch.
M332220523: using FFT length 18432K = 18874368 8-byte floats.
 this gives an average   17.601676676008438 bits per digit
Using complex FFT radices       288        32        32        32
[Jan 29 09:39:46] M332220523 Iter# = 10000 [ 0.00% complete] clocks = 01:51:06.104 [  0.6666 sec/iter] Res64: 1A313D709BFA6663. AvgMaxErr = 0.186969141. MaxErr = 0.250000000.
[Jan 29 11:30:47] M332220523 Iter# = 20000 [ 0.01% complete] clocks = 01:50:56.861 [  0.6657 sec/iter] Res64: 73DC7A5C8B839081. AvgMaxErr = 0.187339746. MaxErr = 0.250000000.
[Jan 29 13:21:40] M332220523 Iter# = 30000 [ 0.01% complete] clocks = 01:50:49.235 [  0.6649 sec/iter] Res64: B928CD22434EEC7C. AvgMaxErr = 0.187298437. MaxErr = 0.281250000.
[Jan 29 15:12:34] M332220523 Iter# = 40000 [ 0.01% complete] clocks = 01:50:49.755 [  0.6650 sec/iter] Res64: 307ECB47139AEB31. AvgMaxErr = 0.187432812. MaxErr = 0.250000000.
[Jan 29 17:03:33] M332220523 Iter# = 50000 [ 0.02% complete] clocks = 01:50:55.437 [  0.6655 sec/iter] Res64: 3F64ED9E01C13B1D. AvgMaxErr = 0.187490430. MaxErr = 0.250000000.

Cores 4,5:

INFO: no restart file found...starting run from scratch.
M332220523: using FFT length 18432K = 18874368 8-byte floats.
 this gives an average   17.601676676008438 bits per digit
Using complex FFT radices       288        32        32        32
[Jan 29 10:42:39] M332220523 Iter# = 10000 [ 0.00% complete] clocks = 02:48:41.803 [  1.0122 sec/iter] Res64: 1A313D709BFA6663. AvgMaxErr = 0.186972266. MaxErr = 0.250000000.
[Jan 29 13:31:07] M332220523 Iter# = 20000 [ 0.01% complete] clocks = 02:48:23.214 [  1.0103 sec/iter] Res64: 73DC7A5C8B839081. AvgMaxErr = 0.187355371. MaxErr = 0.250000000.
[Jan 29 16:19:43] M332220523 Iter# = 30000 [ 0.01% complete] clocks = 02:48:32.271 [  1.0112 sec/iter] Res64: B928CD22434EEC7C. AvgMaxErr = 0.187289062. MaxErr = 0.281250000.

~7 years to complete a test on the first cluster, ~10.65 years on the second cluster, an average of a completed test every ~4.23 years. Power consumption is ~5.085W, giving an average power consumption per 100M test of ~188.4 kWh.

[kriesel]

Originally Posted by M344587487

Here's simultaneous runs of FFT 18432K on cores 0,1,2,3 and 4,5 respectively. Cores 0,1,2,3:

INFO: no restart file found...starting run from scratch.
M332220523: using FFT length 18432K = 18874368 8-byte floats.
 this gives an average   17.601676676008438 bits per digit
Using complex FFT radices       288        32        32        32
[Jan 29 09:39:46] M332220523 Iter# = 10000 [ 0.00% complete] clocks = 01:51:06.104 [  0.6666 sec/iter] Res64: 1A313D709BFA6663. AvgMaxErr = 0.186969141. MaxErr = 0.250000000.
[Jan 29 11:30:47] M332220523 Iter# = 20000 [ 0.01% complete] clocks = 01:50:56.861 [  0.6657 sec/iter] Res64: 73DC7A5C8B839081. AvgMaxErr = 0.187339746. MaxErr = 0.250000000.
[Jan 29 13:21:40] M332220523 Iter# = 30000 [ 0.01% complete] clocks = 01:50:49.235 [  0.6649 sec/iter] Res64: B928CD22434EEC7C. AvgMaxErr = 0.187298437. MaxErr = 0.281250000.
[Jan 29 15:12:34] M332220523 Iter# = 40000 [ 0.01% complete] clocks = 01:50:49.755 [  0.6650 sec/iter] Res64: 307ECB47139AEB31. AvgMaxErr = 0.187432812. MaxErr = 0.250000000.
[Jan 29 17:03:33] M332220523 Iter# = 50000 [ 0.02% complete] clocks = 01:50:55.437 [  0.6655 sec/iter] Res64: 3F64ED9E01C13B1D. AvgMaxErr = 0.187490430. MaxErr = 0.250000000.

Cores 4,5:

INFO: no restart file found...starting run from scratch.
M332220523: using FFT length 18432K = 18874368 8-byte floats.
 this gives an average   17.601676676008438 bits per digit
Using complex FFT radices       288        32        32        32
[Jan 29 10:42:39] M332220523 Iter# = 10000 [ 0.00% complete] clocks = 02:48:41.803 [  1.0122 sec/iter] Res64: 1A313D709BFA6663. AvgMaxErr = 0.186972266. MaxErr = 0.250000000.
[Jan 29 13:31:07] M332220523 Iter# = 20000 [ 0.01% complete] clocks = 02:48:23.214 [  1.0103 sec/iter] Res64: 73DC7A5C8B839081. AvgMaxErr = 0.187355371. MaxErr = 0.250000000.
[Jan 29 16:19:43] M332220523 Iter# = 30000 [ 0.01% complete] clocks = 02:48:32.271 [  1.0112 sec/iter] Res64: B928CD22434EEC7C. AvgMaxErr = 0.187289062. MaxErr = 0.281250000.

~7 years to complete a test on the first cluster, ~10.65 years on the second cluster, an average of a completed test every ~4.23 years. Power consumption is ~5.085W, giving an average power consumption per 100M test of ~188.4 kWh.

Too fast, retina still holds the title for slowest M332m iteration at 3.5 seconds. Hold my beer, no, brrr, hot chocolate, while I dust off a 486? A Quadro 2000 is way too fast in CUDALucas:

Starting M332220523 fft length = 18432K
|   Date     Time    |   Test Num     Iter        Residue        |    FFT   Error     ms/It     Time  |       ETA      Done   |
|  Jan 30  16:23:18  | M332220523       200  0x184d420359b5d1e5  | 18432K  0.31250 150.2123   30.04s  | 577:14:06:20   0.00%  |
|  Jan 30  16:23:48  | M332220523       400  0x263bb77002b531e0  | 18432K  0.28320 150.1744   30.03s  | 577:12:20:57   0.00%  |
|  Jan 30  16:24:18  | M332220523       600  0x8d1d9013c95ef654  | 18432K  0.30273 150.1736   30.03s  | 577:11:43:58   0.00%  |
|  Jan 30  16:24:48  | M332220523       800  0x85773229c143f973  | 18432K  0.26563 150.1654   30.03s  | 577:11:13:53   0.00%  |

one core of an i3-M370:

[Jan 30 16:39:11] Iteration: 100 / 332220523 [0.00%], ms/iter: 589.645, ETA: 2267d 06:27
[Jan 30 16:41:24] Iteration: 200 / 332220523 [0.00%], ms/iter: 596.800, ETA: 2294d 18:45

Darn, still too fast, and yet 6.3 years to complete in prime95.
Hmm, here's a contender for a lowly minion:

[Jan 30 16:48] Worker starting
[Jan 30 16:48] Starting primality test of M332220523 using FFT length 18M, Pass1=2K, Pass2=9K, clm=4
[Jan 30 16:50] Iteration: 10 / 332220523 [0.00%], ms/iter: 5805.210, ETA: 22321d 20:54
[Jan 30 16:52] Iteration: 20 / 332220523 [0.00%], ms/iter: 14515.949, ETA: 55815d 22:12

MPC Transport T3000 nominally 2.1Ghz Pentium M, clocks in at ~800Mhz in prime95 29.4b7 win32 for some reason, estimated 153. years to complete.

Seriously though, I wonder how the phone results at ~ 5.3 100Mdigit tests per MW-hour compare to other choices. Here that is about $22.50 each in electrical cost.

[ewmayer]

Originally Posted by M344587487

~7 years to complete a test on the first cluster, ~10.65 years on the second cluster, an average of a completed test every ~4.23 years. Power consumption is ~5.085W, giving an average power consumption per 100M test of ~188.4 kWh.

How does that total-power figure compare to Prime95 on recent Intel? Sounds like it's in the same ballpark based on various Watts-under-load figures I've seen bandied about hereabouts.

So let's see ... now you just need to fill your basement, garage or spare room with ~1500 such discard-due-to-cracked-screen-dead-battery-or-upgrade-to-latest-greatest phones in order to be cranking out one 100Mdigit LL-test result per day, on average. Figure a 10-phone module with 5mm spacing between phones for airflow and cooled by a single 125-150mm case fan occupies a roughly 6"-per-side cube. You need 150 such modules, so stacked on a horizontal surface we're talking about the area of a ping-pong table. How much does a 5-10kW solar-array install cost these days? Curtis Cooper needs to watch his back! :)

[kriesel]

Originally Posted by ewmayer

How does that total-power figure compare to Prime95 on recent Intel? Sounds like it's in the same ballpark based on various Watts-under-load figures I've seen bandied about hereabouts.

So let's see ... now you just need to fill your basement, garage or spare room with ~1500 such discard-due-to-cracked-screen-dead-battery-or-upgrade-to-latest-greatest phones in order to be cranking out one 100Mdigit LL-test result per day, on average. Figure a 10-phone module with 5mm spacing between phones for airflow and cooled by a single 125-150mm case fan occupies a roughly 6"-per-side cube. You need 150 such modules, so stacked on a horizontal surface we're talking about the area of a ping-pong table. How much does a 5-10kW solar-array install cost these days? Curtis Cooper needs to watch his back! :)

Buy, root, & install one phone a day, and assuming no failures, it takes just over one presidential term to get up to speed. Then start replacing with newer faster model phones one per day for the next 4 year cycle. A good layout and labeling for locating failed units could be very useful. Could get interesting with ~8KW waste heat (5+W each, plus power conversion inefficiency, distribution losses, 150 fans running)

[kriesel]

Originally Posted by ewmayer

How much does a 5-10kW solar-array install cost these days?

6kw system $13k net after federal subsidy.

https://news.energysage.com/how-much...st-in-the-u-s/

[retina]

Originally Posted by kriesel

... it takes just over one presidential term to get up to speed.

There are more than 100 countries in the world. And many don't have presidents. Or if you are talking about a company, then there are millions of them, and if they have a president then the term will be of an unknown value. [/rant]