So, my girlfriend and her business partner's four year old laptops are dying.

I finally convinced them to buy desktop machines rather than laptops, since neither of them ever take any of their laptops home (or anywhere) with them. They both have tablets.

Great, so I source from a local supplier a generic desktop build based on generic parts. Easy and inexpensive to upgrade, fix and expand. Total cost for the hardware: less than 1/5th the cost of a comparable laptop.

The unhappy me part? The price for the Operating System (Winblows) is about 2/3rds the price of the hardware!

Oh well... Perhaps in a few years AutoCAD will be available for Linux....

[QUOTE=airsquirrels;426564]... my injured Titan[/QUOTE]

This is turning into a bad trend. I just had a second Titan, this time on water, melt down under similar circumstances. These cards were rock solid doing TF work for months but it appears LL work is causing the MOSFET/Coils dedicated to powering the RAM to overheat... Unfortunately this card is in the middle of a block of 4 on water so I had to resort to disabling that PCIe slot in bios and unplugging the power cables.

[QUOTE=chalsall;426573]So, my girlfriend and her business partner's four year old laptops are dying.

I finally convinced them to buy desktop machines rather than laptops, since neither of them ever take any of their laptops home (or anywhere) with them. They both have tablets.

Great, so I source from a local supplier a generic desktop build based on generic parts. Easy and inexpensive to upgrade, fix and expand. Total cost for the hardware: less than 1/5th the cost of a comparable laptop.

The unhappy me part? The price for the Operating System (Winblows) is about 2/3rds the price of the hardware!

Oh well... Perhaps in a few years AutoCAD will be available for Linux....[/QUOTE]

Try WINE?

[QUOTE=airsquirrels;426577]These cards were rock solid doing TF work for months but it appears LL work is causing the MOSFET/Coils dedicated to powering the RAM to overheat...[/QUOTE]

Oh shit! Abort!

LL work involves a lot more distant RAM access. That is what you are seeing in the power and thermals.

[QUOTE=ixfd64;426579]Try WINE?[/QUOTE]

Nope. Not an option.

I am not going to expose myself to all the support calls because WINE didn't perfectly support AutoCAD's software virtualization.

[QUOTE=airsquirrels;426577]This is turning into a bad trend. I just had a second Titan, this time on water, melt down under similar circumstances. These cards were rock solid doing TF work for months but it appears LL work is causing the MOSFET/Coils dedicated to powering the RAM to overheat... Unfortunately this card is in the middle of a block of 4 on water so I had to resort to disabling that PCIe slot in bios and unplugging the power cables.[/QUOTE]

This definitely discourages me from trying LL on GPU.

Running LL on GPU is a folly reserved for rare instances of validating the new Mp's.
/... let the eggstorm begin! .../

Seriously, -- there are [B]so many[/B] other things that GPUs are exceptionally good for. Why waste them on things that they do at best mediocrely (barely on par with a small bundle of CPU cores) and overheat in the process?

[QUOTE=Batalov;426602]Running LL on GPU is a folly reserved for rare instances of validating the new Mp's.
/... let the eggstorm begin! .../

Seriously, -- there are [B]so many[/B] other things that GPUs are exceptionally good for. Why waste them on things that they do at best mediocrely (barely on par with a small bundle of CPU cores) and overheat in the process?[/QUOTE]
I have found GPU LL much more troublesome than TF on the same card. I like to see if I can get CuLu to run on a card, as a hardware test, as much as anything.

EDIT: Since I lack real CPU power (except in consumption), an occasional DC on a GTX 580 is an interesting diversion.

[QUOTE=Chuck;426599]This definitely discourages me from trying LL on GPU.[/QUOTE]

I wonder if there's some sort of design flaw with Nvidia cards. It shouldn't be possible to physically damage hardware simply by running a program.

[QUOTE=ixfd64;426607]I wonder if there's some sort of design flaw with Nvidia cards. It shouldn't be possible to physically damage hardware simply by running a program.[/QUOTE]

LaurV described this in the previous posts - there is such a flaw, specifically in EVGAs air cooler, they do not adequately dissipate heat from the power phases for the memory chips. It isn't something guaranteed to happen just from running LL, but card orientation, airflow, etc. will all affect the likelihood of blowing the component.

LL testing thrashes a GPUs memory far more than TF and much more than typical occasional gaming. Never giving the card a breather means those MOSFETs are hot, think > 110*C. Properly bonding them to the cooler or water block would prevent the issue, but the real issue is inadequate power phases for the amount of memory chips on the card.

In general GPUs are much better at TF, but Titans have a pretty low crossover point due to their massive F64 power.

[QUOTE=airsquirrels;426564]I would certainly be willing to help obtain and forward that host board to you, along with my injured Titan.[/QUOTE]
I already asked Xyzzy and got an offer for help from him. Finding the mobo is the hardest part. If you find one, we can trade. I will pay around $500 for the board and around $50 for transportation. Adding a i7-5960x CPU to it could help too, for which I will also pay the price, this I can buy locally, it just takes a long time. The prices for the last are almost the same here. The target mobo is [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16813182968"]this one[/URL], and when I said "about" for the prices, I meant \(\pm\)20% for the mobo and \(\pm\)100%for shipping. I take care personally about customs (another +10% from the [U]declared[/U] (insured) value (which can be half, for example). Memories I can buy locally. The last resort is to buy a "rampage 5 extreme" locally, for the same price, but half of the features (and only 3x8 1x16 PCIe instead of 4x16 - the speed difference for cudaLucas is about 11-12% for example, like 5.8ms/iter against 6.5ms/iter, for some big FFT, this is the maximum, if higher or lower, like for the current LL front, the speed is around 3ms/iter and the difference is only 3-5%) and no usb 3.1, but that is the last resort only.