1)
Message boards :
Number crunching :
Very low credit for Zen 4 mobile CPUs?
(Message 2171)
Posted 9 Jan 2024 by WuJJ Post: This post is no longer relevant after I switched to anonymous platform. Thanks. |
2)
Message boards :
Number crunching :
Very low credit for Zen 4 mobile CPUs?
(Message 2170)
Posted 7 Jan 2024 by WuJJ Post: I just signed up for the project recently and has been playing with different machines I have. I happen to have both desktop and mobile Zen 4 CPUs (7950X and 7840HS). Given they have the same micro-architecture, I generally expect similar performance scaled by frequency. At worst a bit lower due to smaller cache on the mobile CPU. I get such expected result from a few other BOINC projects. However, on sidock, the mobile CPU is getting around 1/10 to 1/8 of credit for a similar runtime. This seems to be incredibly low for a modern CPU. 7950X: https://www.sidock.si/sidock/results.php?hostid=54061&offset=0&show_names=0&state=4&appid= 7840HS: https://www.sidock.si/sidock/results.php?hostid=54054&offset=0&show_names=0&state=4&appid= I'm curious why and if this indicates the mobile CPU are not suitable for the workload and I'd be better off moving them to other projects. Thanks. PS: The mobile CPU is a mini-PC, not laptop. I've verified it's running at expected frequency, not thermal throttled. |
3)
Message boards :
News :
CmDock source code
(Message 2169)
Posted 6 Jan 2024 by WuJJ Post: I'm inclined to try building my own binary, but I realized even my newly signed up hosts get tasks of quorum 1. That's probably not a great idea in general when the reliability and correctness of a host is unproven. I remember BOINC has some option to turn on additional validation until a host is proven to generate reliable results. Would I get additional validation if I switch to anonymous platform? The last thing I want is accidentally generating bogus results because of some compiler difference. Or would `-Dtests=true` be enough confidence that my compiled binaries are correct? When I tried to compile on Ubuntu 23.10, I've already had to make one change. $ git diff diff --git a/src/lib/Rbt.cxx b/src/lib/Rbt.cxx index e953312..7cf5bb2 100644 --- a/src/lib/Rbt.cxx +++ b/src/lib/Rbt.cxx @@ -14,6 +14,7 @@ #include <algorithm> //For sort #include <climits> //For PATH_MAX +#include <cstdint> //For uint64_t #include <cstdlib> //For getenv #include <ctime> //For time functions #include <dirent.h> //For directory handling @@ -581,4 +582,4 @@ bool Rbt::endsWith (std::string const &fullString, std::string const &ending) { } else { return false; } -} \ No newline at end of file +} Also thanks to @kotenok2000. I haven't got to the anonymous platform part yet, but your comment will likely make that far easier for me. :-D |
©2024 SiDock@home Team