getting all cores to work on Win7
 

Hi,

I recently bought an 8-core machine so that I could do 8 LL tests at the same time.

I'm running prime95 on Win7 64-bit.

The trouble I'm having is that I can't seem to get 8 cores working at the same time.

There's an option on Test->Worker Windows to change the number of CPU's to use but I'm not sure this is the thing that needs changing and besides, the option is not active (it's fixed).

In the past, when I've ran prime 95 on a machine with x cores, prime95 automatically gets x assignments.

Can someone help me get all 8 cores working?

What is your CPU? Is it 8 physical cores, or are you including hyper threads? These are just preliminary questions. I've never seen the "Number of worker windows" option locked down, so I'm not sure what that could signify.

You might try assigning specific affinities for each core. At the least, this might give other indications of what is going on.

AMDFX9370 Black Edition 4.4GHz

How do I go about assigning different affinities to each core then please?

stop prime95
test -> worker window
set option 'CPUs to use (multithreading)' to 1 and click 'ok' (needed!)
again test -> worker window
set option 'Number of worker windows to run' to 8 and click 'ok'
continue prime95

now it should work if you have 8 assignemts for 8 workers in your worktodo-file.

edit:
hm, ok, if the option 'CPUs to use (multithreading)' is fixed then its another problem.
Or try to change it manually in 'local.txt' in your p95-folder.
stop prime95
change option 'WorkerThreads=' to 8
change option 'ThreadsPerTest=' to 1 (I hope it is the correct order...)
save changes and restart prime95
I hope this helps. Maybe the relative new FX9370 BE is not detected correct by prime95, but not sure, sry

[QUOTE=MatWur-S530113;378138]stop prime95
test -> worker window
set option 'CPUs to use (multithreading)' to 1 and click 'ok' (needed!)
again test -> worker window
set option 'Number of worker windows to run' to 8 and click 'ok'
continue prime95

now it should work if you have 8 assignemts for 8 workers in your worktodo-file.[/QUOTE]

At 4.4GHz with 8 cores running this system will be hot.

You may run into some memory (or other) bottlenecks if you try to run 8 LLs at once. You might find better performance by having each test run on two or more cores.

If you haven't already, you should run a benchmark in Prime95 (Options > Benchmark); one of the parts of it shows what total throughput (in iterations per second) you get with different combinations of workers per core. This should help you choose the best setup.

[QUOTE=wildrabbitt;378129]How do I go about assigning different affinities to each core then please?[/QUOTE]

From "undoc.txt":
[QUOTE]The program makes its best guess at how the OS maps hyperthreaded logical CPU
numbers to physical CPUs. It also assigns workers and helper threads
to CPUs for optimal speed. However, bugs, new architectures, or situations we
haven't considered may make different affinity settings desirable. In
local.txt set
[CODE] AffinityScramble2=string[/CODE]
Where the characters in "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz()"
represent 64 logical CPU numbers. For example, let's say you have a system
with 8 logical cores with 4 workers each using a helper thread. Also, assume
your system has logical CPUs 0 & 4 on the same physical CPU core, 1 & 5, etc.
If the program is properly determining which logical CPUs share the same physical
CPU, then the program internally generates an affinity scramble string of "04152637".
The program's default policy is to assign the worker and helper threads to the same
physical CPU. If the program is not properly determining which logical CPUs share the
same physical CPU, or you think a different affinity policy would result in better
performance, then set AffinityScramble2 accordingly. Let's say you think
running the helper threads on a different physical core would be better, then
you might set AffinityScramble2=02134657 to test out your theory.[/QUOTE]

[QUOTE=R.D. Silverman;378141]At 4.4GHz with 8 cores running this system will be hot.[/QUOTE]

That's for sure! 220W TDP has the processor. I hope it is a liquide-cooled one ;)
And Core Temp 1.0 RC6 (actual version, free available) is very helpful for all CPUs running prime95. GPU-user should use GPU-Z 0.7.8 (actual version, free available), too.

Thanks a million guys. It's great to know there's lots of knowledgeable people on this site.

I managed to get all 8 cores working :) :)

I'm a bit worried about the CPU overheating.

I've got a Noctua HN-D14 cooler on it.

What's your reckoning - is it going to break?

/* EDIT */

What I did was add a variable called WorkerThreads to the list in local.txt because there wasn't one there.
I set it to 8, and lo and behold, all 8 cores have a job BUT, the expected completion dates are Nov 2016.

Since I can do 4 similarly sized Mersenne number tests on a slower machine in 3 months I'm wondering why these ones are set to take so long.

It will be hot (and possibly not 8-fold fast) for yet another reason:
[QUOTE="many places, e.g. http://us.hardware.info/reviews/5135/2/"]

The Vishera chip the FX-9590 and FX-9370 are based on was launched back in October 2012 (review). That chip is built around four Piledriver modules, the second generation of AMD's Bulldozer architecture. Each Piledriver module contains two integer CPU cores and [B]one[/B] floating point core. It's an 8-core CPU in a way, but [B]for floating point computations there are four[/B].[/QUOTE]

Hopefully, you already knew that.

Most of the LL (prime95) computation is floating point, so the scaling factor will be just around 4. (It is possible that 8 threads will be slower than 4: there are still just 4 FPUs but twice as much memory access. You have to test all comnfigurations, and pick the best.)

Thanks.

I changed WorkerThreads to 4 in local.txt

The program does run 4 seperate tests but when

I go to to the worker windows box and click OK, I get the message

You have allocated more threads than CPU's available.

...which explains why the expected completion dates for these 4 are Sep 2015.

I guess things aren't as simple as adding a variable to a config file.

AMD's naming these chips 8-core (and other similar non-Opteron-non-Phenom versions) was a subject of quite a bit of scorn. Some could even say that this was deceptive advertising. But then again, they are 8-[I]integer[/I]-core chips, so some algorithms may benefit. Maybe even a special LL implementation. Maybe, George or Ernst could comment.