[QUOTE=lfm;191198]Nope, I set unreservedays=90 anyway and it didn't help. the factor job was set to start oct 1 according to the status menu

I have done several smaller bitlevels on the same exponent without any problem on the same machine even.[/QUOTE]

Thats odd. on a whim I set UnreserveDays=900 and it worked. I wonder if this machine is fooling the speed test on startup to think it is slower than it is so it thinks it has a lot more work than it does. I didn't notice before cuz I usually only get work one day in advance. Ill see if I can find any more eveidence. Thanks anyway.

[QUOTE=lfm;191277]Thats odd. on a whim I set UnreserveDays=900 and it worked. I wonder if this machine is fooling the speed test on startup to think it is slower than it is so it thinks it has a lot more work than it does. I didn't notice before cuz I usually only get work one day in advance. Ill see if I can find any more eveidence. Thanks anyway.[/QUOTE]

I've had that happen when it gets the wrong CPU type (i.e. my Q9550 at times get classified as a Xeon or even as Unknown) and then the estimated completion times almost doubled. You could look at your CPU list on the server or look at Options / CPU from your PC.

Suggestion re: timing for new end dates send...
 

I have noticed that if I ask for new end dates to be sent daily that the time of day does NOT remain consistent.

Let's say I do it manually at 8:30 AM the first day.

If it takes a couple minutes to complete sending then the next day the updates will be sent at 8:32 and each day after that creeping forward in a similar fashion.

On top of that, if for some reason an update is delayed (i.e. 70 minutes due to server congestions) then the next automatic update won't be until 9:42 the next day.

I like to be able to coordinate all my PCs to have the same update time so I can check shortly after that time and if any have not reported in I use it as a red flag to check that PC is still running.

Can I suggest one of two alternatives:
1. Let me choose the time-of-day the new-end dates are sent. It may disallow a time of busy server activity (i.e. the first couple minutes after the hour).

2. Once the first end-date transmissionis started have the client software save that time and send it each day at that same time.

thanks

I have noticed that version 25.11 takes the work line:

PRP=1,2,670600,40291

and reformats it as

PRP=1,2,670600,40291,0,0

I did not notice this with previous versions, but I may have just not been paying attention. My question is, what do the last two fields mean, the ones that are filled in with zeros?

[QUOTE=philmoore;195177]
PRP=1,2,670600,40291,0,0

what do the last two fields mean, the ones that are filled in with zeros?[/QUOTE]

PRP=k,b,n,c[,how_far_factored,tests_saved][,known_factors]

I looked at the code but don't see how it outputs the 0,0:

[code]
if (w->tests_saved > 0.0)
sprintf (buf,
"PRP=%s%.0f,%lu,%lu,%ld",
idbuf, w->k, w->b, w->n, w->c);
else
sprintf (buf,
"PRP=%s%.0f,%lu,%lu,%ld,%g,%g",
idbuf, w->k, w->b, w->n, w->c,
w->sieve_depth, w->tests_saved);
[/code]

[QUOTE=Prime95;188842]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?[/QUOTE]

Other than violating the principle of least surprise, it's not wrong. cpuid has always returned the maximum APIC address, which in the new x2APIC scheme is no longer the same as core count and thread count. I think the following gets it right, but I don't own an i7, so all I can say is it will probably compile... (This is based on section 2.1.12 of the Intel cpuid spec)

[code]
--- a/cpuid.c
+++ b/cpuid.c
@@ -424,7 +424,21 @@
/* of these situations, by comparing number of cores to number of logical */
/* processors. This test fails if the machine has multiple physical CPUs. */

- if (max_cpuid_value >= 1) {
+
+/* Determine if we should use leaf 11 topology enumeration. */
+/* Otherwise, use leaf 1 + leaf 4 */
+ if (max_cpuid_value >= 11) {
+ reg.ECX = 0;
+ Cpuid (11, &reg);
+ }
+
+ if (max_cpuid_value >= 11 && reg.EBX != 0) {
+ CPU_HYPERTHREADS = reg.EBX & 0xFFFF;
+ reg.ECX = 1;
+ Cpuid(11, &reg);
+ unsigned int cores = reg.EBX & 0xFFFF;
+ if (num_logical_processors <= cores) CPU_HYPERTHREADS = 1;
+ } else if (max_cpuid_value >= 1) {
Cpuid (1, &reg);
if ((reg.EDX >> 28) & 0x1) {
CPU_HYPERTHREADS = (reg.EBX >> 16) & 0xFF;
[/code]

[QUOTE=nfortino;195188]Other than violating the principle of least surprise, it's not wrong. cpuid has always returned the maximum APIC address, which in the new x2APIC scheme is no longer the same as core count and thread count. I think the following gets it right,[/QUOTE]

I'll give it a try in the next release. Thanks.

Hello,
could you clear me the point of memory statistics for Prime95x64 25.11 for Win W7x64. It consumes a lot more memory according e.g. nightly statistics (peak and current) in Win taskmanager with torture tests in manual memory settings.
Testings done in safe mode with 90% of ram consumption 8-4096KB FFTs. And i'm afraid that this manual memory limit that's set as option is violated and this leads to possible swappings(
Any formulaes for calculation the complete memory consumption for this kind of manual settings tests?
Any ideas?

[QUOTE=asmfan;196031]
could you clear me the point of memory statistics for Prime95x64 25.11 for Win W7x64. It consumes a lot more memory according e.g. nightly statistics (peak and current) in Win taskmanager with torture tests in manual memory settings.
Testings done in safe mode with 90% of ram consumption 8-4096KB FFTs. And i'm afraid that this manual memory limit that's set as option is violated and this leads to possible swappings(
Any formulaes for calculation the complete memory consumption for this kind of manual settings tests?
Any ideas?[/QUOTE]

I don't own Windows 7, so I cannot comment on whether prime95 torture test allocates/uses more memory than in earler Windows versions.

You are correct that if too much memory is used swapping will occur thus reducing the effectiveness of a torture test. My only suggestion is to be very safe in your memory settings - say 60% of RAM. This will still produce a very good torturing of your system.

Actually is there a strict formulae for memory allocations? Or how strict the manual memory limit is regarded? I think it doesnt matter which OS but process statistics matters. Everyone can look at process memory statistics in taskmanager (ctrl+shift+esc on Win and select proper columns) with its peak memory level and exceeeds manual limits a lot. Can you check it yourself after nightly torture run with manual memory limits?
I'm trying to get max testing for MCH and also memory as well as CPU thus use max free mem available but doubt why peak exceeds limits leading i guess to swaps.
Also can Prime95 output some performance statistics like operations per second? then it could catch swappings if OpS (Flops) drops. Is it hard to count operations number for template equation with variable params in formula?

[QUOTE=asmfan;196151]Actually is there a strict formulae for memory allocations? Or how strict the manual memory limit is regarded? I think it doesnt matter which OS but process statistics matters. Everyone can look at process memory statistics in taskmanager (ctrl+shift+esc on Win and select proper columns) with its peak memory level and exceeeds manual limits a lot. Can you check it yourself after nightly torture run with manual memory limits?
I'm trying to get max testing for MCH and also memory as well as CPU thus use max free mem available but doubt why peak exceeds limits leading i guess to swaps.
Also can Prime95 output some performance statistics like operations per second? then it could catch swappings if OpS (Flops) drops. Is it hard to count operations number for template equation with variable params in formula?[/QUOTE]

The memory limit in prime95 only refers to the main data structures in P-1 steps of factoring really. Other steps and other memory uses are separate and basically fixed depending on the size of the exponent(s) assigned.

You aren't going to be assigning it accurate to the last byte in any case so just leave lots of extra room. There is such a thing as over-tuning and it sounds like you are there to me.