allquixotic

unlikely as it is

somehow I'm actually here

blog.xkcd.com/2019/08/26/how-to-send-a-file

just gotta move my game servers and I'm done with old box

| ra      | RUNNING | 51.81.21.6 (eth0)            |                                               | PERSISTENT | 0         |
|         |         | 172.18.0.1 (br-31153ee2c440) |                                               |            |           |
|         |         | 172.17.0.1 (docker0)         |                                               |            |           |
+---------+---------+------------------------------+-----------------------------------------------+------------+-----------+

new:

+---------+---------+------------------------+------+------------+-----------+
| ra      | STOPPED |                        |      | PERSISTENT | 1         |
+---------+---------+------------------------+------+------------+-----------+

old:

roota.cc is back.

"brief"

Moving Synapse onto new server, brief outage of roota.cc

apple.slashdot.org/story/19/07/09/1453239/…

!!info

for one thing, I have to reach back to the left behind my left hand to reach the ctrl key, but the Cmd key on a Mac is right there, easily accessible by thumb

also maybe it's me but Cmd+Left / Cmd+Right / Cmd+Tab / Cmd+C / Cmd+V are easier, more convenient shortcuts than Home/End/Alt-Tab/Ctrl+C/Ctrl+V

@rahuldottech why do you think low key travel is bad for programming / lots of typing? I think I type just as fast, if not faster, on a modern MBP keyboard than a nice mechanical keyboard (and I have some really nice ones)

sort of like how data transfer between cores inside a given CCX isn't ever really a bottleneck because they all share the same caches and such

these are general problems with multithreaded workloads in general, but on Intel and non-Infinity Fabric AMD CPUs, all cores shared the same L3 cache, and data transfer between cores was not really a factor in performance

and you end up wasting computation on speculative execution when it turns out that the shared data needs to be updated

and the frequency of shared data update basically dictates how frequently the infinity fabric has to sync data across the CPUs' L3 caches

the problem is, even if the OS knows it's a NUMA chip, the performance of different workloads will range from: horribly unoptimized (with lots of threads updating and reading the same data), to decent (with some shared data and some non-shared), to amazing (with each thread not sharing any data with any other thread)

that would be a bad idea imo because the performance definitely does not resemble a UMA chip

either that or it's exposed to the system as a UMA chip, which seems unlikely given the chiplet architecture

I'm guessing LTT didn't test "Game Mode" if that's still an option for Ryzen 3

that's for Zen+ threadripper and not Ryzen 3, but the concept is probably the same

so they've obviously been thinking about this; "Ryzen Master" is their proprietary Windows software that lets you tweak your CPU

> AMD carves the Threadripper 2990WX into four NUMA domains that cannot be altered. As such, the processor does not have a local memory toggle for its Game Mode feature. Instead, the processor simply flips into "1/4" mode, which disables all but one die and effectively creates an 8C/16T CPU. Ryzen Master also has "1/2" and "Off" options that expose 16 cores and 32 threads, or 32 cores and 64 threads.

tomshardware.com/reviews/…

there might be some sort of compiler hint about thread data locality that would give the C runtime some sort of idea of whether it should run the threads in the same CCX or try to spread them out... if there isn't, there should be

if you have 12 threads constantly accessing the same data (and updating it in a very tight loop), you might just want to put it all on one CCX

if you have 12 threads running on unrelated data, distributing them across all 12 cores (3 CCXes) is completely fine and you'll saturate your CPUs without using much infinity fabric at all

yes, it depends entirely on the data access pattern of the threads

you just have to know which cores correspond to which CCXes on your CPU

@djsmiley2k tech-recipes.com/rx/37272/…

it's literally faster to run Battlefield V on 4 cores instead of 8 or 12 for Ryzen 3

@bwDraco Linus Tech Tips shows that if you restrict Battlefield V to running on the threads of a single CCX (instead of all of them) it can beat the 9900K, but otherwise Intel crushes AMD on that particular game, so clearly infinity fabric "2" is still a bottleneck if you have threads banging on the same data across multiple CCXes

it's sad when China is more pro-nuclear (and soon to be greener) than the rest of the world, and will be adopting European Power Reactors before Europe (although at least UK is building them, good on them)

If they start permitting and site selection for some EPRs this year, and ramp up solar and wind in the near term, their EPRs will be ready to go by the time the solar panels and wind turbines need replacement. But cynically they'll probably rely on increased coal burning for their AC needs :/

news.slashdot.org/story/19/07/04/1410205/…

I'd love to know which algorithms are now using "salt and pepper" - that made me giggle

well that's a new one! :D

> salt and pepper

I guess I don't understand the context

> metamorphically

????

> a acceptable

the WSL bit is interesting, but TBH I use code-server to run VS Code in a browser instead

6/24/19 Internet outage (BGP leak) explanation