hermann on Tue, 27 Jun 2023 23:50:59 +0200


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: gp-2.15.3 compiled on Ubuntu on AMD 7600X very slow compared to i7 (many page faults), Win11 gp-2.15.3 on 7600X better than i7. How to fix AMD Ubuntu?


On 2023-06-27 09:10, Bill Allombert wrote:
There used to be (it was called MACHINES, see commit
6cf8508275cbfd566682b0fc66f1a631e6b29b9d) but nowadays this bench only measure
the time it takes to read the test files.

This seems to be fast ...

It is ! It is the same time as on the Mac M1.

Cool, best M1 is "only" rank 63 with value 3871 in Passmark "Single Threaded Performace" list of 3,100+ CPUs:
https://www.cpubenchmark.net/singleThread.html

I managed to get my new 619$ only PC to run with 5.3GHz, although I had to reset Bios by shortening "Clear CMOS Jumper" and not knowing why default settings now show this. I added the details sofar:
https://github.com/Hermann-SW/7600X#details-of-pc

Now this PC is on rank 18 of the 3,100+ CPU single core list with value 4222.

"make bench" became much faster as well:
hermann@7600x:~/Downloads/pari-2.15.3$ make bench
Making bench in Olinux-x86_64
make[1]: Entering directory '/home/hermann/Downloads/pari-2.15.3/Olinux-x86_64'
* Testing objets           gp-sta..TIME=    0   gp-dyn..TIME=    1
* Testing analyz           gp-sta..TIME=    4   gp-dyn..TIME=    5
* Testing number           gp-sta..TIME=    5   gp-dyn..TIME=    5
* Testing polyser          gp-sta..TIME=    0   gp-dyn..TIME=    2
* Testing linear           gp-sta..TIME=    0   gp-dyn..TIME=    0
* Testing elliptic         gp-sta..TIME=    6   gp-dyn..TIME=    4
* Testing sumiter          gp-sta..TIME=    2   gp-dyn..TIME=    0
* Testing graph            gp-sta..TIME=    1   gp-dyn..TIME=    0
* Testing program          gp-sta..TIME=    1   gp-dyn..TIME=    1
* Testing trans            gp-sta..TIME=    4   gp-dyn..TIME=    1
* Testing nfields          gp-sta..TIME=   18   gp-dyn..TIME=   19
+++ Total bench for gp-sta is 26
+++ Total bench for gp-dyn is 22
make[1]: Leaving directory '/home/hermann/Downloads/pari-2.15.3/Olinux-x86_64'
hermann@7600x:~/Downloads/pari-2.15.3$


And most importantly, 7600X CPU determines "sqrt(-1) (mod p)" for 100,355-digit prime in 75.7% runtime of i7-11850H CPU (matching single threaded list numbers for both CPUs well: 3152 / 4222 = 74.7%).


Now I feel to have the right PC for playing with PARI/GP, number theory and integer factorization ... ;-)


Regards,

Hermann.