Prime95 version 25.11

[Prime95]

Originally Posted by patrik

Just to be clear: But v. 25.9 does not have the line

if (num_logical_processors <= cores) CPU_HYPERTHREADS = 1;

where CPU_HYPERTHREADS then is changed from 2 to 1.

Ah, you have an early 25.9 build. That line was added for an i7 with hyperthreading disabled in the BIOS. CPUID reports the chip can run 8 threads on 4 cores but the OS says there are only 4 logical processors. The logical assumption is that hyperthreading has been disabled in the BIOS.

[joblack]

Did anyone try to use 8 hyperthreads with the Core i7 ... is prime95 faster than 4 'real ones'

[Prime95]

Originally Posted by joblack

Did anyone try to use 8 hyperthreads with the Core i7 ... is prime95 faster than 4 'real ones'

There were a couple of threads on that. I don't remember the exact results, but as I recall multithreading (running 4 workers each using 2 threads) was worth plus-or-minus a few percent.

[jinydu]

I'm trying to install this on Linux; but it isn't working. After unzipping and then untarring, a bunch of files show up as expected. But typing ./mprime then gives the error messages:

./mprime: /lib/i686/libpthread.so.0: version `GLIBC_2.3.2' not found (required by ./mprime)
./mprime: /lib/i686/libc.so.6: version `GLIBC_2.3.4' not found (required by ./mprime)

[Paulie]

The link in item 0 for the Mac GUI version should be Prime95-MacOSX-2511.zip

[lfm]

Originally Posted by jinydu

I'm trying to install this on Linux; but it isn't working. After unzipping and then untarring, a bunch of files show up as expected. But typing ./mprime then gives the error messages:

./mprime: /lib/i686/libpthread.so.0: version `GLIBC_2.3.2' not found (required by ./mprime)
./mprime: /lib/i686/libc.so.6: version `GLIBC_2.3.4' not found (required by ./mprime)

Those are system libraries. you probably have older versions on your system. You should consider upgrading your Linux distro.

Alternatively try to find a static linked version of mprime. I'm not sure if there is one currently offered.

[petrw1]

Originally Posted by Prime95

There were a couple of threads on that. I don't remember the exact results, but as I recall multithreading (running 4 workers each using 2 threads) was worth plus-or-minus a few percent.

Is this what you were thinking of??

http://www.mersenneforum.org/showthread.php?t=11025

Specifically the results seen by Phantomas with AffinityScramble?

[nucleon]

I'll add my machine in the 'me too' basket. I have a corei7 with 6GiB ram.

Yep I have the incorrect detection of core/HT. 25.11 detects 8 core, 1x hyperthread. 25.09 detects correctly 4core, 2x hyperthread.

Setting the following in local.txt fixed things as a workaround:
NumCPUs=4
CpuNumHyperthreads=2
CpuSpeed=2699

Of course CpuSpeed is different depending which corei7 you have.

-- Craig

[patrik]

Originally Posted by Prime95

Ah, you have an early 25.9 build. That line was added for an i7 with hyperthreading disabled in the BIOS. CPUID reports the chip can run 8 threads on 4 cores but the OS says there are only 4 logical processors. The logical assumption is that hyperthreading has been disabled in the BIOS.

I have now investigated the CPUID values of my Core i7 a little further. I have learnt a little assembly language and have written two small programs that directly call CPUID and return some value in the exit code. It seems that CPUID.01H returns 16 in EBX[Bits 16-23], and CPUID.04H (and with ECX=0) returns 7 in ECA[Bits 26-33] (meaning 8 cores). These values are the same whether or not I disable hyper-threading in the BIOS.

/proc/cpuinfo behaves as expected and shows 8 logical processors with HT enabled and 4 when disabled.

::::::::::::::
readcores.asm
::::::::::::::
section .text

   global _start

   _start:
      mov     eax, 04h            ; EAX argument to CPUID
      mov     ecx, 0              ; ECX argument to CPUID
      cpuid                       ; Perform the CPUID
      mov     ebx, eax            ; Copy EAX to EBX
      shr     ebx, 26             ; Get bits 26-33
      mov     eax, 1              ; exit() system call
      int   80h                   ; Return the results as exit value
::::::::::::::
readhts.asm
::::::::::::::
section .text

   global _start

   _start:
      mov     eax, 01h            ; EAX argument to CPUID
      cpuid                       ; Perform the CPUID
      shr     ebx, 16             ; Get bits 16-23
      mov     eax, 1              ; exit() system call
      int   80h                   ; Return the results as exit value

Note: Perhaps I should add that the programs were compiled (assembled?), linked and run with

nasm -f elf64 readhts.asm
ld -o readhts readhts.o
./readhts
echo $?

[Prime95]

Originally Posted by patrik

CPUID.01H returns 16 in EBX[Bits 16-23], and CPUID.04H (and with ECX=0) returns 7 in ECA[Bits 26-33] (meaning 8 cores).

Which is dead wrong according to my reading of the Intel CPUID documentation. I checked the i7 errata and nothing alarming is there.

Did I read the Intel docs incorrectly? Anyone with ideas for a workaround?

[monst]

Hello,

I need to stop Prime95 for an hour or so at a time, a few times a week, on one of my machines. I've been digging through the documentation and can't seem to find the settings to do this. (I know I've seen this before somewhere.)

Can someone please point me in the right direction?

Thanks,

-- Rich