Ok, here it is, from an ebay shop in Germany : 729 Euro + PP







Only a cable adapter in the box, no video adapters whatsoever.

The HW configuration before installing it was :

GPU 0 : GTX 780ti on PCI-E 16X slot
GPU 1 : GTX 660ti on riser 1X slot
GPU 2 : GTX 660ti on riser 1X slot
GPU 3 : GTX 660ti on riser 1X slot

The idea was to insert the GTX 1080 in the 16X slot and shift all GPUs by one logical position, letting one 660ti unused.

At first boot, the 1080 was not even recognized as such, but only as a "generic VGA adapter"..updating the drivers to the latest available ( incidentally released the same day ) fixed it.

Anyway...with this HW configuration...

GPU 0 : GTX 1080 on PCI-E 16X slot
GPU 1 : GTX 780ti on riser 1X slot
GPU 2 : GTX 660ti on riser 1X slot
GPU 3 : GTX 660ti on riser 1X slot

...only the 1080 and one 660ti were recognized by windows...tried several combination of PCI-E 1X slots, but no joy.
At this point I decided to do something more radical : time to reorganize everything and work exclusively with risers, to check if there was any possibility to use all four cards toghether. ( anyway the 1080 alone on the PCI-E 16X was running really fine..)

So, this is the result, and it works :








Please note that all the risers are 1X form factor, but the first one ( connected to the 1080) is plugged into the 16X slot.

And here is how Boinc recognize the GPUs :

7/8/2016 10:04:12 AM | | Starting BOINC client version 7.6.33 for windows_x86_64
7/8/2016 10:04:12 AM | | log flags: file_xfer, sched_ops, task
7/8/2016 10:04:12 AM | | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
7/8/2016 10:04:12 AM | | Data directory: C:\ProgramData\BOINC
7/8/2016 10:04:12 AM | | Running under account Andrea
7/8/2016 10:04:13 AM | | CUDA: NVIDIA GPU 0: GeForce GTX 1080 (driver version 368.69, CUDA version 8.0, compute capability 6.1, 4096MB, 3559MB available, 8876 GFLOPS peak)
7/8/2016 10:04:13 AM | | CUDA: NVIDIA GPU 1: GeForce GTX 780 Ti (driver version 368.69, CUDA version 8.0, compute capability 3.5, 3072MB, 2513MB available, 6022 GFLOPS peak)
7/8/2016 10:04:13 AM | | CUDA: NVIDIA GPU 2: GeForce GTX 660 Ti (driver version 368.69, CUDA version 8.0, compute capability 3.0, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 10:04:13 AM | | CUDA: NVIDIA GPU 3: GeForce GTX 660 Ti (driver version 368.69, CUDA version 8.0, compute capability 3.0, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 10:04:13 AM | | OpenCL: NVIDIA GPU 0: GeForce GTX 1080 (driver version 368.69, device version OpenCL 1.2 CUDA, 8192MB, 3559MB available, 8876 GFLOPS peak)
7/8/2016 10:04:13 AM | | OpenCL: NVIDIA GPU 1: GeForce GTX 780 Ti (driver version 368.69, device version OpenCL 1.2 CUDA, 3072MB, 2513MB available, 6022 GFLOPS peak)
7/8/2016 10:04:13 AM | | OpenCL: NVIDIA GPU 2: GeForce GTX 660 Ti (driver version 368.69, device version OpenCL 1.2 CUDA, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 10:04:13 AM | | OpenCL: NVIDIA GPU 3: GeForce GTX 660 Ti (driver version 368.69, device version OpenCL 1.2 CUDA, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 10:04:13 AM | | OpenCL: Intel GPU 0 (ignored by config): Intel(R) HD Graphics 4600 (driver version 20.19.15.4331, device version OpenCL 1.2, 1630MB, 1630MB available, 200 GFLOPS peak)
7/8/2016 10:04:13 AM | | OpenCL CPU: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 5.2.0.10094, device version OpenCL 1.2 (Build 10094))
7/8/2016 10:04:13 AM | SETI@home | Found app_info.xml; using anonymous platform

The idea is to let it run for a couple of day, and log the results with BoincTasks, to check for performance improvements, then, trying to go with 5 GPUs and finally have at least 2 task per GPUs.

I'll post results ( and doubts ) here folks...thanks for your help so far !!

A.

I would go with 3 tasks /GPU

Your 660 should handle it, and you 1080 will LOVE it.

I would go with 3 tasks /GPU
Your 660 should handle it, and you 1080 will LOVE it.

Although AMDave kindly linked to
http://setiathome.berkeley.edu/forum_thread.php?id=79271&postid=1772388 which sould have clarified the concept to me, it did not. I'm getting old ( 51 )..

So, I have 4 cores ( 8 with HT ) and the four GPUs you know...what would you suggest ?

Running
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.5</cpu_usage>
at the moment

The idea, would be to use :

<gpu_usage>0.33</gpu_usage>
<cpu_usage>0.34</cpu_usage>

so that as BilBg said :

3 GPU tasks * 0.34 = 1.02 = (truncated to) 1 = BOINC will run one less CPU task (will "free a CPU core")

so that with four GPUs, it should reserve ( free) 4 HT cores.

Right ?

With 4 real cores running 4 cards, that is tight.

But yes, I would shut down 1 core for each CPU.

@Brent Norman

Thanx Brent !! and...what about 5 Gpus ? :-D

Ok, update:

Guppi everywhere !! And here are the numbers for them ( blc2_2bit_guppi...) :

GTX 1080 OC : 27 min / WU
GTX 780ti OC : 33 min / WU
GTX 660ti OC : 38 min / WU


I now have added the 5th GPUs : another 660ti...running then 5 GPU tasks and 3 CPU tasks.





And this is the power consumption ( @ 220V/50Hz )



7/8/2016 1:41:01 PM | | Starting BOINC client version 7.6.33 for windows_x86_64
7/8/2016 1:41:01 PM | | log flags: file_xfer, sched_ops, task
7/8/2016 1:41:01 PM | | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
7/8/2016 1:41:01 PM | | Data directory: C:\ProgramData\BOINC
7/8/2016 1:41:01 PM | | Running under account Andrea
7/8/2016 1:41:02 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 1080 (driver version 368.69, CUDA version 8.0, compute capability 6.1, 4096MB, 3559MB available, 8876 GFLOPS peak)
7/8/2016 1:41:02 PM | | CUDA: NVIDIA GPU 1: GeForce GTX 660 Ti (driver version 368.69, CUDA version 8.0, compute capability 3.0, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 1:41:02 PM | | CUDA: NVIDIA GPU 2: GeForce GTX 780 Ti (driver version 368.69, CUDA version 8.0, compute capability 3.5, 3072MB, 2513MB available, 6022 GFLOPS peak)
7/8/2016 1:41:02 PM | | CUDA: NVIDIA GPU 3: GeForce GTX 660 Ti (driver version 368.69, CUDA version 8.0, compute capability 3.0, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 1:41:02 PM | | CUDA: NVIDIA GPU 4: GeForce GTX 660 Ti (driver version 368.69, CUDA version 8.0, compute capability 3.0, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 1:41:02 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 1080 (driver version 368.69, device version OpenCL 1.2 CUDA, 8192MB, 3559MB available, 8876 GFLOPS peak)
7/8/2016 1:41:02 PM | | OpenCL: NVIDIA GPU 1: GeForce GTX 660 Ti (driver version 368.69, device version OpenCL 1.2 CUDA, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 1:41:02 PM | | OpenCL: NVIDIA GPU 2: GeForce GTX 780 Ti (driver version 368.69, device version OpenCL 1.2 CUDA, 3072MB, 2513MB available, 6022 GFLOPS peak)
7/8/2016 1:41:02 PM | | OpenCL: NVIDIA GPU 3: GeForce GTX 660 Ti (driver version 368.69, device version OpenCL 1.2 CUDA, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 1:41:02 PM | | OpenCL: NVIDIA GPU 4: GeForce GTX 660 Ti (driver version 368.69, device version OpenCL 1.2 CUDA, 2048MB, 1683MB available, 2810 GFLOPS peak)
7/8/2016 1:41:02 PM | | OpenCL: Intel GPU 0 (ignored by config): Intel(R) HD Graphics 4600 (driver version 20.19.15.4331, device version OpenCL 1.2, 1630MB, 1630MB available, 200 GFLOPS peak)
7/8/2016 1:41:02 PM | | OpenCL CPU: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 5.2.0.10094, device version OpenCL 1.2 (Build 10094))
7/8/2016 1:41:02 PM | SETI@home | Found app_info.xml; using anonymous platform


I will post the numbers. ;-)

Ciao

Andrea

Interesting setup. I'd feel more comfortable without those nylon cable ties supporting the cards. Nylon DOES get brittle when exposed to heat and UV over time. No UV exposure I assume since they are indoors. Could have some disastrous results if the the cable ties broke and the card swung down. I would feel better with metal hangers. Don't know if this is some temporary setup or is going to be the permanent installation.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Interesting setup. I'd feel more comfortable without those nylon cable ties supporting the cards. Nylon DOES get brittle when exposed to heat and UV over time. No UV exposure I assume since they are indoors. Could have some disastrous results if the the cable ties broke and the card swung down. I would feel better with metal hangers. Don't know if this is some temporary setup or is going to be the permanent installation.

Well..the nylon ties are "indefinitely temporary".. :-) I agree with you tho, they don't last forever... I doubled almost all of them, plus I added, when possible, a third tie on the center of the GPUs ( 660ti ).
For sure this is something to deal with again soon..

A.

Here's a couple pics of my setup, similar thought, little different approach:





Actually works pretty well.

---

Here's a couple pics of my setup, similar thought, little different approach:
Actually works pretty well.

Very neat Al, indeed, mine's still a mess compared... :-D
What is your HW configuration ? CPU, GPUs, MB...etc etc..

A.

Hmm, let see, dual Xeon X5690's, 48 gig of fairly fast RAM, dual 460(?) gig SSD HD's in a RAID 1 config (overkill, I know...) 3 GTX 950 models and one 750Ti, the MB is a EVGA Classified SR-2. Overclocked a little bit on both the CPU and GPU's. Dual Noctua CPU coolers and fans, and an EVGA Platinum 1000 P2 power supply. I think that covers it pretty much. Oh, and a Al brand GPU rack, can't forget that one!!

---

Question for GazzaVR,

Are you running that 1080 on a PCIe 1x riser in a 16 bit slot?

It sure looks like it to me from the pictures.

Oh, and a Al brand GPU rack, can't forget that one!!

hmmm...looks more like a "DeWALT & Woods GPU rack"...the totally customize-able series of course!!! ;-}

Man, you guys are giving me too many ideas!
There's just one issue...
Has anyone done this type of open-air setup with long-haired cats or dogs around?
What would make more noise when intertwined: the fans or the pet? :-O
I'm guessing it also depends if electricity is added to the blood stream!
{my bad}

---

I would think cats deciding they want the heat from the GPU might create a problem when they lie down on top of them....

2 words ... Electric Fencer :P

Oh, and a Al brand GPU rack, can't forget that one!!

hmmm...looks more like a "DeWALT & Woods GPU rack"...the totally customize-able series of course!!! ;-}

Oh, no, that was after the IP aquisition. They did rebrand it top DeWALT & Woods after I sold them the rights to it. As now seen on TV! lol

As to the Fur, well, that's a tough one. Any way to not let them into that room? Or, you could surround the outside of the shelves with furnace filters and have a box fan blowing out for negative pressure... or the electric fence idea was mentioned... :-O

One slightly less lethal thing you might consider to keep them away is what I purchased for as a gift for a friend of mine who's room mates cats seemed to like to visit upon his kitchen counters and table after using the litter box.. YUM!! Not!

I got him a length of approx 3 foot wide carpet runner, with those pointy little things on the back that hold it to the carpet, and he cut them to size and put them on all of his counters and kitchen table, points up obviously. Last I heard, their gross roaming was a thing of the past, he even verified it by dusting the counters with a little flour after taking them off after a week or 2, and no tracks! Yay! Hopefully it's permanent. Might be something to consider?

---

Greets All

Guys they are some serious cruching rigs there, but excellent work on getting them fired up and running.

Oh the ideas floating around in my head, only if I had a an insulated shed/garage lol.

Another question is, running costs, do you have really cheap electricity in your part of the world. Not so cheap here where I am (Western Australia).

Okay, time to duck out and change over the latest SOG app (3484) and give that a run on my two 760's and 1 770.

Happy Crunching

Hmm, let see, dual Xeon X5690's, 48 gig of fairly fast RAM, dual 460(?) gig SSD HD's in a RAID 1 config (overkill, I know...)

Wow...Al …seems pretty out of budget for my Pocket !!
If I'm not wrong, each Xeon is worth about 1500 euros...😲😲
Hope you' e got some deal for them....

Question for GazzaVR,
Are you running that 1080 on a PCIe 1x riser in a 16 bit slot?
It sure looks like it to me from the pictures.

Sorry for the late reply Brent...I'm outta town as usual..
Yes you've got it right: the riser for the 1080 is plugged onto the 16 bit slot.
Saw it working in a video on YT, and it seems to work pretty well here too...sadly, if I plug the 1080 directly on it, the other gpus are not recognized... Or at least partially ( one out of three )

Another question is, running costs, do you have really cheap electricity in your part of the world. Not so cheap here where I am (Western Australia).

Have no idea...And I'm afraid I will find out next bill 😖😖😖
However, for my GF house, the MAX power you can use is 3.3 KW and the cost is divided into three main timeslots.
Anyway, the average cost for an Italian family is 0.163 €/kWh ( 0.18 $/kWh or 0.238 AUD/kWh)

Naa, they were like $250 or so each. That's why it's fun putting together "old" hardware, these things were new I believe 3-4 years ago, and businesses are dumping them like crazy. There's a thread called Building a 32 thread xeon system doesn't need to cost a lot, and this system is in there, as well as one I built that is slightly more current, as it uses two V2 Xeon procs, and it has 24 cores per proc. To be able to afford it, it had to be the slowest one they made, and even then it wasn't exactly cheap. Nothing remotely close to what you mentioned, but almost double the X5690s.

I'm doing it as an experiment, to see how well something with that many cores grinds out WU's, and I just recently put in my new GTX 1080 FTW into it as well. The RAC had sort of stabilized around 32k with just the procs, and is just passing 54k and still climbing, will be interesting to see how well it produces. I am running 4 tasks on it, and assigned 4 cores, 1 per task. Other than a little glitch I had last night, it's been pretty much flawless since I put it together over a month ago.

I am also playing with a couple boards that I bought for cheap, that were mentioned in the thread as well, they are motherboards out of a HP Z800. Dual proc, can handle the X5690s no prob, aren't really expensive, but you have to do a little fiddling around to get the power to work right (buy an adapter off of ebay is the easiest, about $5) and if you are planning on putting it into a standard case, that might present an issue as well. It can be done, but the selection is limited. Right now I am modding the CPU mounting brackets to allow a standard CPU cooler to fit, because as it is, only a (very expensive) HP heatsink fits in it without modification. I followed the trailblazing lead of this gentleman, he did most of the homework for the rest of us Z800 hackers. I see power consumption mentioned as a concern, I am fortunate to have installed a 10kw solar array on my house a few years back, so my summer electric bills aren't _too_ bad, winter are worse due to frequent snow cover and Really short days. The joys of living at 45 degrees North lat.

---