wonder why?
19 May 2012 0:06:33 UTC
Topic 196337
(moderation:
I just started up my new computer Windows 7, 3.6 and 8 core versus the older one which is also Windows 7, but 3.2, and 6 core. My Einstein units take longer on the more powerful computer by about 40 minutes each. I have a GTX 580 and the cuda units crunch much faster than on the older one with a GTX 460. That I understand, but I am puzzled by the cores running slower. Does anyone have an answer?
wonder why?
)
Assuming you have bulldozer. Don't know about this project, but if they do use floating path math, you bulldozer shares those resources within the 4 modules, so each "core" gets 1/2 of a modules FPU (floating point unit).
Your Phenom 2 however had 6 independent cores, each with its own FPU.
As a side note, if you're running on all 8 cores+GPU, you should tell BOINC to only use 87.5% of your processors. This will give you GPU more CPU time when it needs it. Should speed others up as well.
Cheers
Sorry, but could you repeat
)
Sorry, but could you repeat it in simpler terms for the novice here, please?
NOT TECHNICAL AT ALL AT THIS
)
NOT TECHNICAL AT ALL AT THIS STUFF
I'll try. Not the most technical myself. To put it as simple as I can, floating point can handle, let's say more decimals for the sake of argument than integer math can. integer math may be like 1+1=2 whereas floating point can do 100/9. FOR EXAMPLE (may be wrong it my description, but I'm right below about bulldozer design)
FPU is can handle more numbers than integer I THINK.
A bulldozer is made of 8 cores in 4 modules. Each module contains 2 cores. Now each bulldozer core contains its own integer (resources), so if this project used integer math, you would be running 8 integer cores (as far as your CPUs resources go).
In bulldozer, there is only 1 FPU PER MODULE. Not per core. So, each core must share the FPU resources of the module.
With your Phenom 6 core, each core had its own FPU. It was not having to share its resources with another core. Thereby, making the tasks run faster.
Not an engineer (obviously lol), so this is the best I can do. Hope this helps.
Thank you Spot, it does help
)
Thank you Spot, it does help and it also made me realize that I should have taken more science courses in college.
So, in a nutshell, these 8 cores run slower than the 6 cores in the other machine, even though it has a higher Gig processor.
I have heard that you can make a computer run faster by overclocking it. Is that a safe thing to do?
Thank you.
BE CAREFUL IF YOU WISH TO DO
)
BE CAREFUL IF YOU WISH TO DO THIS, YOU HAVE BEEN WARNED
As long as you have aftermarket cooling than yes. If you're using your stock cooler, it depends.
Bulldozers tend to be a pretty warm chip, so if you wish to check your temps, I would recommend downloading Core Temp first. See what your temps are. overclock.net says max AMD temps for bulldozer are around 55-60C, so you want to be around 45-50C, since we use these 24/7/365.
OVERCLOCKING IS SERIOUS BUSINESS TAKE TIME TO LEARN AS MUCH AS YOU CAN, AND GO SLOWLY!!!!!
You should be able to bring it to 4.0GHz no problem, as long as you have decent cooling.
Here is a rather basic guide which I think could suit your needs http://apcmag.com/building-a-pc-amd-bulldozer-overclocking-guide.htm
Don't worry about stress testing, as this is what grid computing (BOINC) is for. With max voltage being 1.4 (vCore), you'll want to stay as under this as possible. The higher the voltage, the shorter the lifespan of the chip. Same applies for temps. The lower the better.
IF YOU TAKE THE TIME TO LEARN HOW TO OC PROPERLY, YOU WILL LEARN A VERY VALUABLE NEW SKILL. SEARCH INTERNET FORUMS, AND TAKE YOUR TIME. YOU WILL BE SURPRISED AT HOW MUCH YOU CAN LEARN.
AGAIN, OC AT YOUR OWN RISK
You will need to pay attention to the multiplier, vcore, and temperature. These will be the most important aspects of overclocking your CPU
RE: Thank you Spot, it does
)
Besides what 5pot said, and that all the overclocking stuff could be an interesting learning, the truth is that your gaining in terms of RAC will not be impressive.
I used "RAC" not only thinking in the credits itself, but also as a meassure of your contribution to the science of the projects...
One thing to take in account is that while your phenom take less time to do the work it does at most 6 Wus in that time, while the new CPU will do 8, that's a 33% more WUs, so if the crunching times are not longer than a 33% your will be doing more work anyway...
Think of Bull(crap)dozer "8
)
Think of Bull(crap)dozer "8 core" as being a 4 core with poorly designed hardware level hyperthreading and the most bizarrely useless cache design imaginable.
Foul- unnecessary roughness.
)
Foul- unnecessary roughness.
Assuming that aspects of this
)
Assuming that aspects of this discussion refer to hyperthreading then see this thread for an interesting discussion. ( Alas I hadn't gotten around to any further Ix as I had alluded to there ). Also beware any effect ( thinking of GW analysis especially ) of variation between the work units themselves - they may not all be created 'equal'.
@Miklos : hyperthreading ( in simple terms ) refers to a facility whereby one physical core is 'two-faced' or represents two virtual cores. Think of it like a bank teller that serves two queues. Note that this is a different level concept than 'threads' or 'processes' ( which are operating system abstractions based on varieties of 'task switching' ) as virtual cores have a significant underlying hardware implementation. There is still only one physical core per two virtual cores. All these concepts - cores, threads, processes - refer to degrees of implementation/switching/scheduling of instruction queues ( the lists of instructions that the CPU should perform ) and associated data areas.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Think of
)
Deary me, you obviously have no idea of processor design!
Is "hyperthreading" 'crap' because it fools you into believing you have 2 CPUs and yet only delivers 1 CPU and 0.3 of a further CPU?...
The Bulldozer design is rather clever and nicely high performance. The limiting factor is the utterly CRAP software and the uneducated Marketing steered unrealistic user expectations.
Happy?
Now please look again at what can be done and at what price.
If you are wanting ultimate performance, then your best bet is to go for an "uber" GPU, and then whatever CPU is adequate to merely feed that GPU.
On that count, my research suggests that AMD CPU + nVidia GPU are best for whatever price point you might want.
Otherwise, please keep with the original goals of Boinc and acquire whatever computer best fits your needs and then donate only spare CPU/GPU cycles to the furtherance of science.
Happy cool crunchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)