For now I am sticking with Cpu as I don't have a Gpu to use yet.

---

Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.

Would you guys run cpu's at the same temps your gpu's are running at?

Rule of thumb: each 10°C hotter halves chip life time

So a gpu running 20C hotter than idle reduces the life span to a 1/4?

In my point of view running MW on a cpu is just a waste of electricity and this crunching power could better serve other projects

I noticed you're running a few quads for MW. Is this not a waste also (or inefficient) if all the crunching is done by gpu?

It they don't then, off with their heads. :)

Seriously, I run only GPU on Collatz. Here, because I don't have the higher power, higher heat, higher cost GPU's to work with, I run only CPU.

I have two GPU projects -- Collatz and GPUGrid.

All of my other projects are CPU only. That choice here is hardware driven. About the only other project which supports GPU that I'm currently working is SETI -- and that project REALLY doesn't need me trying to do GPU work.

Einstein has been dabbling in GPU support, but so far from what I understand, their GPU application is problematic and not all that efficient.

Aqua was doing GPU work, but found their application wasn't efficient for GPU work.

Seems some folks are a bit into the divisive GPU versus CPU thing, I don't think that was Arcturus' intent. Heck, from his RAC numbers I'd guess he has one GPU for Milkyway and it doesn't run full time (an approach to avoiding heat issues for sure).

I stopped CPU-MW as soon as I noticed the ATI client back in January. In my point of view running MW on a cpu is just a waste of electricity and this crunching power could better serve other projects. Projects, where the algorithm forbids efficient use of GPUs and where we're not going to get around using CPUs for quite some time to come.

...and if someone does not have a compatible GPU and does not wish to attach to projects other than Astronomy / Physics projects?

---

Einstein is taking the cautious approach and "feeling" there way into a good application. So, they are doing part of the work on the GPU and part still on the CPU (if I am reading the discussions right). As they gain experience with the GPU code more will be migrated.

SaH's application on the GPU, unless it has changed significantly, had issues with VLAR tasks that almost took as long on the GPU as they would on the CPU with an optimized application. Though now I understand that there is a fall back so that those tasks will be run on the CPU if you have the right package saving the GPU for the tasks that can be run the most efficiently on the GPU.

Aside from SaH Beta test, the only other project that has a GPU application now would be The Lattice Project though I gave up on them because they are content with a 10% failure rate of tasks ... tasks that are up in the 90% complete area before they die ... I don't know about you ... but I am not content with an application with a failure rate like that ... and the GPU version had the same basic failure rate ...

Of course, their work is sporadic in availability ...

Still, there is slow progress ...

I have basically 4 systems running which do D.C.
I stopped running almost completely on cpu projects except for CPDN Beta and also for testing.
With this tests i check if cpu temps keeps under 55C with full load as i am doing now on a failing intel Q6600 system.
My main machine Amd X3-720 (52c full load) has 2 x XFX 4850 1 GB cards running MW or Collatz, as beyond stated about XFX claims to give lifetime warranty on their products. Both run at 690 Mhz core and mem at 1053 mhz.
top card temps Dispio 56c, memio 60c, shader 64c
low card temps Dispio 56c, memio 58c, shader 60c
so in all pretty cool for vc under load
This system is ubercooled and sounds a bit like an airplane with his 9 x 120mm fans

My Htpc Amd 4850e dual core (25c full load)is currently a bit more powerfull since it has got a XFX 4770 running also at Mw or Collatz together with his buildin hd3200.
Ofcourse the case is open since it becomes damn warm now, the motherboard is 44c and the gpu at 820mhz core is 82c mem is at 830mhz

Now back on topic i am not sure if and when these cards will fail on these constant stress
If i look back on my old machines most still work.
For those who bought my old systems (mostly build with high quality parts)
For instance my old P133 which have run many years on seti allways stressed at the fullest still works and even is now running with XP on it.

So why did another Q6600 start failing and giving weird errors, my friend who gave me this machine had run it constant on D.C. projects.
I can't say for certain why it crashes.
We assume its cause can be found in the constant running at 80 to 90c with his 4 cores under constant stress for couple of years.
But ofcourse we can't proof it since nobody has ever showed me test results from machines being tested in this manner under controlled conditions.

So if i look back i think even though we push this machines to some limits, i did not see clearly that this harm the parts.
My guess is that if something fails there has been a flaw in the manufactoring or lesser quality parts are being used, thus resulting in failure in time.

Ofcourse when you pass the limits given by factory you may risk getting failure sooner.
For instance oc my XFX 4850 cards is limited by ati with a max of 700 on the gpu and memory is being limited to 1200.
Nevertheless i know a few persons who run at much higher temps/speeds and almost never get a dying videocard or cpu. except when they go again for some oc records with the LN2 stuff and deep freeze coolsystems

PS i do game sometimes for more then 12 hours ;)

---

Its new, its relative fast... my new bicycle

Brian wrote:

...and if someone does not have a compatible GPU and does not wish to attach to projects other than Astronomy / Physics projects?

I'd have told them how beautiful the other cpu-only Astro / Physics projects are and would've gotten flamed for it by *some* people.

arcturus wrote:

Would you guys run cpu's at the same temps your gpu's are running at?

I suspect many are. But not because GPUs are magically immune to the wear out I described here, but rather because it's difficult to cool them quitely with the limited amount of space the stock coolers have.

arcturus wrote:

So a gpu running 20C hotter than idle reduces the life span to a 1/4?

Yes. In fact any semiconductor chip, be it a GPU, CPU, northbridge or whatever. That's why e.g. Intel specifies maximum (operating) temperatures for their chips. It's not that they'd suddenly break if you go past that temperature but rather that decay is so accelerated that too many of them would fail within the warrenty period (or whatever lifetime Intel is shooting for).

arcturus wrote:

I noticed you're running a few quads for MW. Is this not a waste also (or inefficient) if all the crunching is done by gpu?

They're running non-GPU projects ;)

Team_Elteor_Borislav, I don't want to be rude, but you may want to take a look at the link in the beginning of this post. Component failure is a statistical process influenced by several key parameters. We know a lot about how they affect large batches of products, but we can't use it to predict when an individual component will fail.
BTW: in my experience motherboards and power supplies fail much earlier than non-overvolted CPUs, even if they run under continous load.

MrS

---

Scanning for our furry friends since Jan 2002