overclocking and prime95 test
 

hello
playing recently with one PC, trying to squeeze it to the max but keep it stable and found something strange
quick description - quad core xeon e5450 (3.0ghz 1333 FSB) on ga p35c-ds3r with 8gb ddr3 1600 and win7 64bit

initially single thread was running on one core only, so when I run prime95 with 4 threads - 4 corres at 100%, if i run prime95 with 1 thread, a random core run at 100% and other 3 at 0-1%
so overclocked to 1520 FSB was working pretty stable

but later (dont know how and why, problably win updates or I did something while playng with settings) single thread begin to switch quickly between the cores, not only prime95 but also other single thread processes
so when I run prime95 with 4 threads - again stable for a long time, but when run with 1 thread, it begin swithcing quickly between the cores, cpu load is 25% average total, and each core load vary a lot, then fail in less than a minute, so I had to reduce FSB to 1400 to keep it stable

so .... can anyone explain what and why is happening? why before was only 1 core fully loaded, and later it spread the load and fail more quickly
thanks

Affinity is a strange animal on windows 64. What you experience, switching from a core to the other when you run in less cores than max, it is a normal process. Your computer is quite clever and it will "protect itself" (read: avoid getting the CPU too hot) by a number of tricks, including switching to a cooler core when one gets too hot, or throttling when all get too hot. This extends the lifetime of your CPU. The dark side of it is that switching cores take time, therefore lowering a bit the performance.

Related to why you experienced different behavior in the past, there might be many reasons about this, my money are on thermal stuff. Your computer was more able to cool itself in the past, than it is now, and this can have plenty of causes. You either moved the computer to a warmer room, or the summer came to you, or you increased the clock, or some dust clogged into your fan meantime, decreasing its cooling performance, or [URL="http://www.hoax-slayer.com/snake-in-computer.shtml"]some snake[/URL] found a warmer place to live...

You can "lock" some task (worker) to some core, using the "set affinity" option in task manager. Open the task manager, go to "processes" tab, right-click any process (i.e. Prime95) and use "set affinity" to "lock" it so some specific cores. Look around in this forum and the web for best practices in setting the affinity. Also, resd the "undoc.txt" file which come with Prime95 distribution.

You better don't play with the affinities, if you don't know what you are doing. You may gain a little bit of performance more, but decrease the lifespan of your CPU. The first step when playing with this kind of things is to install a temperature-monitoring program and closely monitor the temperatures of the cores. I use RealTemp and AsusProbeII, but there are also better tools there. Sometime the best way is to let it default (switching cores when less workers).

Of course, the maximum performance for P95 is reached when you have a number of workers equal to the number of you [U]physical[/U] (i.e. non Hyper-threaded) cores, and each worker uses a physical core. If you don't have any other reasons to run fewer workers, than you better use 4 workers, each single-threaded.

The temperatures are a good thing to keep in mind. I was having "stability" issues that got fixed when I fixed my thermal problem.

Even a safe voltage can overheat a cpu with a bad cooler, so Prime95 is not telling you the whole story.

Do you know what temps the CPU was reaching?

If you want a single worker (thread) to run on a particular core, go to[INDENT]Test>Worker Windows
[/INDENT]Here you can say how many workers to run, and make choices about where they run. In the third field, change "All Workers" to the number of the worker. If you only have one in the top field (Number of workers to run), then the third field grays out on Worker #1. At this point you can choose what core it will run on in the "CPU Affinity" drop down.

With multiple workers repeat this process for each worker.

thanks for all answers
I forgot to mention - my approach was for overclocking, so I put the low power version of CPU, E5450 is only 80W design(equiv X5450 is 120W)
then a huge copper heatsink and huge fan, the maximum temperature (with 4 cores long time at 100% load) is only 62 degrees, this is internal CPU temperature
core voltage is set to manual 1.25V @ 1400FSB and 1.281V @ 1520 FSB
online calculator tell me that the maximum CPU power with 1.281V @ 1520 FSB should be 104W

now same behaviour with switching cores is without overclocking, where temperature always less than 60 degrees with 100% load, so definitely not temperature.....
my thought was same for switching - it may take resource for switching too quickly and may loose performance...

also I did manual set affinity to 1 core, and then running only 1 thread, only 1 core at 100%, it is working same stable as before at 1520FSB

so 4 threads on 4 cores work stable at 1520 FSB
1 thread on 1 core work stable at 1520 FSB
1 thread switched between cores work stable at 1400 FSB

that mean something is going wrong while switching between the cores

[QUOTE=nik;372047]core voltage is set to manual 1.25V @ 1400FSB and 1.281V @ 1520 FSB
online calculator tell me that the maximum CPU power with 1.281V @ 1520 FSB should be 104W

[/QUOTE]

Erm... the front side bus at 1400 MHz?

I dont actually understand the question
motherboard frequency setting is 350 for example, which is 4 times for CPU 4*350 = 1400 FSB

update - replaced power supply with a quality 750W - no change

Did you buy the heatsink yourself or was it the stock one? What are your temperatures?

it was sitting arround from old gaming PC
it is copper with heatpipes
cpu internal temperature never get more than 62 degrees

Well, I wasn't heavily into PC's until the core i series came out. Before then, they were just mystery black boxes.

From what I understood, though, the Front Side Bus (FSB) [I]is[/I] the motherboard (north bridge, specifically) frequency which gets multiplied by the CPU multiplier. Maybe AMD still does that to this day; I'm not familiar with overclocking in the red camp.

A base frequency of 350 MHz sounds a touch high to me but that's just a gut feeling. I could be completely wrong.

[QUOTE=TheMawn;372130]I could be completely wrong.[/QUOTE]
No, you are not wrong. Or at least, not completely :razz: Classical [URL="http://en.wikipedia.org/wiki/Front-side_bus#Related_component_speeds"]FSB architecture[/URL], with north and south bridges, is not used anymore in today's computers. The maximum FSB clock ever used was 533MHz, and that was only for a single type of Pentiums (3.2GHz with multiplier 6). The usual FSB speed at the end of its life (before being [URL="http://en.wikipedia.org/wiki/HyperTransport#Front-side_bus_replacement"]changed to HyperTransport[/URL] and/or [URL="http://en.wikipedia.org/wiki/Direct_Media_Interface"]DMI[/URL], about 10 years ago) was around 400MHz. In the AMD circles, they use to call for example "1400FSB", and that is where the confusion is from. The 1400 FSB T-bird Athlon, for example, is an 133 FSB (or double pumped 266 FSB) with the right internal multipliers.