| 13:43 | topi`_> | wohoo! bigger DIMMs work just fine with the macchiatoBIN |
| 13:43 | topi`_> | [ 0.000000] Memory: 16130796K/16513152K available (8331K kernel code, 544K rwdata, 2948K rodata, 336K init, 242K bss, 365972K reserved, 16384K cma) |
| 13:43 | topi`_> | that is 16 gigabytes |
| 13:46 | topi`_> | also installed a small (noisy) fan. Now the temp from thermal_zones 0,1,2 is below 40 |
| 13:47 | vpeter> | topi`_: You have device on a work table? |
| 13:47 | topi`_> | yes |
| 13:47 | topi`_> | in less than 25C ambient |
| 13:48 | vpeter> | Why not put it somewhere else? |
| 13:48 | topi`_> | in the fridge? |
| 13:48 | topi`_> | hah |
| 13:49 | vpeter> | Some other room I meant :) |
| 13:49 | topi`_> | I'll try my noise cancelling headphones |
| 13:49 | topi`_> | 1529 root 20 0 332056 33172 2224 S 400.0 0.2 0:26.14 pbzip2 |
| 13:49 | topi`_> | now I'm trying to see how high it goes with increasing cpu load |
| 13:51 | topi`_> | OK it went from 40C -> 48C |
| 13:51 | topi`_> | not a big rise |
| 13:51 | topi`_> | now just waiting for the marvell 8080, or whatever the octa-core part will be called ;) |
| 14:06 | topi`_> | now trying out 2000mhz clocks for the cores, let's see how if affects the temps |
| 14:07 | topi`_> | how is the heatsink connected to the 8040? Is it glued somehow? |
| 14:10 | topi`_> | surprising: the idle temp has shifted from 40C to 42C |
| 14:10 | topi`_> | I expected more |
| 14:33 | topi`_> | 54C when compiling linux kernel with make -j6 |
| 14:34 | topi`_> | so it does warm up under heavy stress |
| 14:39 | topi`_> | it took 11 mins wall clock time to compile imx_v6_v7 defconfig from linus's 4.12 tree on the Marvell 8040 |
| 14:40 | topi`_> | that's seriously impressive, about the same time than on my 2016 Macbook Pro |
| 14:40 | topi`_> | I expected gcc to take a severe hit from the meager 512K of L2 cache |
| 14:59 | topi`_> | the same compilation on my next fastest ARM system, the Tegra K1 board (Cortex A-15 at 2.33 GHz) took 13 minutes |
| 15:00 | topi`_> | but that chip has 2 MB of L2 cache |
| 15:00 | topi`_> | and an advantage in clock speed |
| 15:06 | vpeter> | topi`_: how do you measure this time? With time command? |
| 15:12 | topi`_> | yes |
| 15:12 | vpeter> | and this 11 seconds in user time right? |
| 15:13 | topi`_> | btw the reason why I bought that Tegra K1 board was that it was much *faster* compiling the kernels than my Core 2 duo macbook (at that time) |
| 15:13 | topi`_> | 11 minutes, not seconds |
| 15:13 | topi`_> | user time was 43 mins |
| 15:14 | vpeter> | On old i5-2500: |
| 15:14 | vpeter> | real 5m31.898s |
| 15:14 | vpeter> | user 19m21.676s |
| 15:14 | vpeter> | sys 1m1.300s |
| 15:14 | topi`_> | ouch, my results are not comparable, the marvell board compiled with the gcc-6 crosscompiler (arm-gnueabihf) whereas the Tegra was compiling with a native gcc-4.8.4 |
| 15:15 | vpeter> | time is always comparable :) And in my case it is better :) |
| 15:17 | topi`_> | I guess that i5 has 4 cores? |
| 15:17 | vpeter> | yes. |
| 15:17 | topi`_> | my MBP has the low-end Skylake i5 and it has only 2 cores |
| 15:18 | topi`_> | hence about 10 mins... |
| 15:18 | vpeter> | Yes. |
| 15:18 | vpeter> | What exactly is the reason you build on marvell board? |
| 15:19 | topi`_> | if I run an ages old flops20.c benchmark, the cortex-a72 is way ahead of the overclocked Tegra K1 |
| 15:19 | topi`_> | well, why not? I can run arm64 code on the board |
| 15:19 | topi`_> | I admit it is a bit expensive for some geeky experiments |
| 15:20 | topi`_> | MFLOPS(4) = 2147.8886 |
| 15:20 | topi`_> | whereas the K1 got only 1145 |
| 15:21 | topi`_> | ...and 4383 MFLOPS on my skylake i5 @ 2.0ghz |
| 15:22 | topi`_> | this is no big deal since the 8040 has 4 cores vs. 2 in the core i5 |
| 15:23 | topi`_> | I think we can safely assume that Marvell's chips are cheaper than Core i5's |
| 15:25 | topi`_> | anyways, that's comparing apples and oranges, since the 8040 is a 10Gbps/SATA beast |
| 15:25 | topi`_> | the aggregate bandwidth of all those interfaces must be humongous |
| 15:25 | vpeter> | But to compile kernel it takes long :) |
| 15:25 | vpeter> | longer |
| 15:26 | topi`_> | no big deal |
| 15:26 | topi`_> | still faster than *most* arm systems |
| 15:27 | topi`_> | I like to avoid Intel's systems where possible. Who knows what's the next exploit in the IME engine or other shit |
| 15:27 | topi`_> | I have 0 confidence in Intel |
| 15:28 | vpeter> | :) |