| 01:54 | suihkulokki> | keynote at linaro connect today about fog computing.. clearfog was there first ;) |
| 02:31 | jnettlet[m]> | Yes all the rage |
| 10:54 | topi`> | back to life, risen from the dead... what have I missed here? :) |
| 10:54 | topi`> | my bodily antivirus isn't perfect, it seems... |
| 10:54 | vpeter> | topi`: errors and more errors :) |
| 11:02 | Ke> | =o) |
| 11:03 | Ke> | total bytes scrubbed: 6.11TiB with 1 errors |
| 11:03 | Ke> | one more day and it's done |
| 11:05 | Ke> | then it's time to bisect the boot regression |
| 11:06 | Ke> | and then perhaps try to find a newer firmware that can boot my mcbin |
| 11:06 | Ke> | and native compile of the boot fw |
| 11:06 | Ke> | I could perhaps try to compile the Linaro toolchain on aarch64 |
| 11:09 | Ke> | x86 dependency is kind of nasty, as I only have one x86 system and it's more than 5 years old |
| 11:21 | jnettlet[m]> | topi`: I pushed all sorts of u-boot updates for MX6 and a38x |
| 11:22 | Ke> | does that include a8040 as side effect? |
| 11:38 | jnettlet[m]> | not yet. but some of the patches would be related |
| 12:02 | topi`> | now when we start to get those 8- and 16-core 8040 chips, then the real fun starts :) |
| 12:03 | topi`> | I started to run a Zencash securenode on my mcbin, and it beats the crap out of intel Atom servers with the same amount of cores |
| 12:03 | topi`> | the perf is very decent, partly thanks to the good memory subsystem of the Marvell chip |
| 12:04 | topi`> | I'd say it competes neck-to-neck with a Core i3 |
| 12:04 | topi`> | (zencash calculates absolutely huge polynomials as part of its shielded transaction creation) |
| 12:04 | topi`> | (this is totally unrelated to crypto mining) |
| 12:07 | Ke> | would be nice to know, how are the spectre-relations on A72/a8040 |
| 12:08 | Ke> | can you run processes independently on different cores |
| 12:08 | topi`> | since nobody runs services on aarch64, I don't think any real exploits exist, yet |
| 12:08 | Ke> | I guess I am the only one who worries about this, yes |
| 12:09 | Ke> | also both linux and ATF have some sort of mitigations |
| 12:09 | topi`> | this will change when Qualcomm Falkors enter the machine rooms |
| 12:09 | Ke> | well there are also many others, but quite a few of them are in order cores |
| 12:10 | Ke> | like thunderX and Synquacer |
| 12:10 | topi`> | yes, very niche |
| 12:11 | topi`> | the Falkor will be the first one which will just chew through any workload you throw at it, whether single or multi threaded |
| 12:11 | topi`> | obviously it's going to be weaker at some workloads against a Xeon, but on others, it will gain an edge |
| 12:11 | Ke> | well the thing is, that even open source software still has poor support for aarch64, as you can see with the linaro toolchains |
| 12:12 | Ke> | we run hundreds of precompiled toolchains at work and I think we have only have one that is not x86-hosted |
| 12:13 | Ke> | I don't think qualcomm will change that bottomline |
| 12:14 | Ke> | bigger companies running huge homogenous data centers can obviously afford to optimize and adapt for aarch64 |
| 12:56 | topi`> | getting the source to compile shouldn't be a problem, except when the autoconf macro files are outdated |
| 12:56 | topi`> | aarch64 and x86_64 implement quite similar machines, i.e. word size and byteorder. The only real difference is the memory model |
| 12:57 | topi`> | x86_64 implements a strong memory model |
| 12:57 | topi`> | but this usually only has implications at the lowest level |
| 12:58 | topi`> | soon SR will ship mostly aarch64 boards, now that imx8 is out :) |
| 12:58 | topi`> | or is it? 12 weeks lead time I hear |
| 13:07 | jnettlet[m]> | it is available for pre-order |
| 13:07 | jnettlet[m]> | but we will continue to sell the MX6 |
| 13:51 | rabeeh> | topi`: i.mx6 will live forever :) |
| 13:54 | topi`> | yes, I hope so, for the sake of our customer's :) |
| 13:55 | rabeeh> | initially the consumer version was 2022; but now they extended even that |
| 13:55 | topi`> | but why do you have such long turnarounds at product line? I thought everything was ready for the main batch, but then a 2 week delay? |
| 13:55 | rabeeh> | industrial and automotive parts are 15y |
| 13:55 | topi`> | it worked differently while I was at Nokia. We would go into the line, drive the first few boards through the machine, and then we would check that all components are in place and software actually works |
| 13:56 | topi`> | and then, green light and the machine starts in a few hours |
| 22:08 | Ke> | hmm. perhaps upstream u-boot would work, there is defconfig |