voidcomp wrote:


The processing speed will slow down, once the L3 cache is not enough and the CPU needs to load data from RAM too often. So expect a slowdown when this happens. On the 1700 with only 1 MB L3 per thread it happens sooner, than on the 3600 with ~2.8 MB L3 per thread.

On a Ryzen 1700 with more than 3 or 4 MIP units at once there definitely starts a noticable slowdown. Which means, if one interpolates with the amount of L3 cache, that a slowdown with more than 6-8 units on a Ryzen 3600 might occur.

MIP at the moment is the only project I know with this issue. MCM, FAH2, Zika, SCC do/did not have this issue, they seem to use much lower amounts of cache. ARP I'm not sure, I have not tested running many units of it at once.

*edit: it is possible, that ARP is also cache-hungry, at least here is something that points into that direction: https://www.worldcommunitygrid.org/forums/wcg/viewpostinthread?post=617626

Therefore, I would advice to run at max 3 units of MIP together with any other projects you like on a Ryzen 1700, at least that's what I do.

AVX2 perf doubled... or at least they can do an instruction in 1 cycle rather than 2 previously. also ~5-12% IPC increase. and a slight bump in clock speeds, too. i should know. my 180 watt TDP 1950X is weeping over the 105 watt 3900X's power savings.

https://www.techpowerup.com/review/amd-ryzen-5-3600/6.html
https://www.techpowerup.com/review/amd-ryzen-5-3600/5.html <-may not beat the 1700X but it's neck and neck
https://www.techpowerup.com/review/amd-ryzen-5-3600/7.html
https://www.techpowerup.com/review/amd-ryzen-5-3600/9.html

you can probably extrapolate with these:
https://www.anandtech.com/show/14605/the-and-...-review-raising-the-bar/8
https://www.anandtech.com/show/14605/the-and-...-review-raising-the-bar/9

there's enough of a mix of benches that i'm sure it's safe to say that performance will translate here, too.

>I would advice to run at max 3 units of MIP together with any other projects you like on a Ryzen 1700, at least that's what I do.

How is this accomplished? Is there some project limiter tweak which sets an upper limit of concurrent instances of a particular project running?

There's also a setting in your device profiles

if you have multiple machines with different CPU then use what Apis has posted. But also set a limit on MIP.
ATM, when you couple MIP with another sub project says MCM, because MIP will probably finish faster than most ... boinc tend to grab more MIP than others. That sometime cause your machine is full with just MIP (if set Unlimited) and other threads will sit idle.
For ex: you know you don't have any machine that can run more than 12 MIP
Set 25 in the device profile
then use app_config.xml for each rig
1700 app_config of 3 MIP + 13 MCM

<app_config>
<report_results_immediately>1</report_results_immediately>
<app>
<name>mip1</name>
<max_concurrent>3</max_concurrent>
</app>
<app>
<name>mcm1</name>
<max_concurrent>13</max_concurrent>
</app>
</app_config>

If you have very deep pockets, get the new 64 core beast!
https://www.theverge.com/circuitbreaker/2019/...-128-threads-release-date