Risers is a good idea. If I was going to build a monster 4-gpu machine I would plan on using risers to get the GPUs apart from each other.

I know PSU calculators aren't perfect but here is something to at least consider:

---

XtremeSystems.org - #1 Team in GPUGrid

Heat issues always scare me, I figure now that I am running multiple machines, and crunching so often, I will eventually have to conqueror my fears of dealing with thermal paste.

Thermal paste is NOT a problem, just get some good paste, I personally like Arctic Silver, but there ARE others too. ALSO it IS possible to put too much of a blob on there, I have seen examples of 'the size of a pea' but it is just doing it several times that will tell you how much is enough and not too much. You want enough to be able to spread out over the whole cpu, but NOT out over the edges! Remember heat MUST go thru the paste and into the heatsink/fan and too thick of a layer is not good for heat transfer. Basically don't worry EVERYONE gets it wrong a few times before they sort of 'figure it out'. I have NEVER burnt up a cpu by putting too much, but when the cpu DOES run hotter than I thought it should I take it apart and redo it. I use Alcohol Prep Swabs to clean the fan and cpu, making sure it is completely dry before redoing it.

Thermal paste should be spread as thin as possible; although its designed to transfer heat, being a liquid phase it inevitably can't do it as well as the metal solids either side. So just enough to actually cover and link the CPU and heatsink. This is why some people lap both the CPU and heatsink surfaces.

You can also clean fans with water and a cotton bud, but the best fix is prevention - filters. I have a system almost a year old with filters at the front and side, it's still pristine inside, and I just need to vacuum the filters every few weeks. The other systems are not so clean ;p

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Some people may go mad about thermal paste.. but it's not that critical. Personally I like to position the heat sink and then twist it a bit forth and back before I fasten it. Doing so squeezes excess paste to the edges (it doesn't hurt there, it's just wasted) and upon several repetitions I can feel a stronger resistance to my torque, which I interpret as "sitting tightly". Never had thermal paste related problems.. so just give it a try (if you have to).

MrS

---

Scanning for our furry friends since Jan 2002

Yes, we would like to try to build a 8 gpu machine made of 690.

What are the screws to remove (3 or 4)?

gdf

Some people may go mad about thermal paste.. but it's not that critical. Personally I like to position the heat sink and then twist it a bit forth and back before I fasten it. Doing so squeezes excess paste to the edges (it doesn't hurt there, it's just wasted) and upon several repetitions I can feel a stronger resistance to my torque, which I interpret as "sitting tightly". Never had thermal paste related problems.. so just give it a try (if you have to).

MrS

I give mine a couple of little twists too before clamping it down.

GDF are making a BitCoin machine?

Definitely not!

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

What are the screws to remove (3 or 4)?

3 scrwes. Not the visible ones. (The fan blades are shrouding them)

I didn't find any picture of them, so I can make some if you like me to do.

I'm considering the hypothesis of buying a PC with 4xGTX690 water-cooled. SLI will be off. 4 GTX690 works good together? The 690 is a dual GPU, by each GPU there will be 2 WU processing together?

---

Member of Boinc Italy.

I'm considering the hypothesis of buying a PC with 4xGTX690 water-cooled. SLI will be off. 4 GTX690 works good together? The 690 is a dual GPU, by each GPU there will be 2 WU processing together?

It's very hard to get 8 GPUs working in a sigle PC. You need special BIOS, and forget about Windows.

So couldn't someone then just get 3 690's and then 1 680?

So couldn't someone then just get 3 690's and then 1 680?

More than 4 GPUs is usually problematic, especially with Windows.

I'm considering the hypothesis of buying a PC with 4xGTX690 water-cooled. SLI will be off. 4 GTX690 works good together? The 690 is a dual GPU, by each GPU there will be 2 WU processing together?

It's very hard to get 8 GPUs working in a sigle PC. You need special BIOS, and forget about Windows.

"
That`s the reason why you have "only" 3 GTX690? Is it also not a problem with 3 GTX690 and Windows?

---

Member of Boinc Italy.

[3] NVIDIA GeForce GTX 690 (2047MB) driver: 310.33

This is actually one GTX 690 and another GPU (GTX670 or something).
Boinc reports each individual GPU. A GTX690 has two GPU's so Boinc reports two GTX690's! It's doesn't report the card count, just the GPU count, and it has to call it something, so GTX690 it is. Boinc also reports the first (or biggest/most powerful GPU) and then the number of GPU's. So if you have a GTX680 in there, it will report it as another GTX690.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

[3] NVIDIA GeForce GTX 690 (2047MB) driver: 310.33

That`s means also 1X690 and another GTX right?

The maximal number of 690 is also 2?

Another question...with one 690 the PC will crunch 2 WU at the same time?

What`s the better solutions between 2x690 and 4x680?

---

Member of Boinc Italy.

[3] NVIDIA GeForce GTX 690 (2047MB) driver: 310.33

That`s means also 1X690 and another GTX right?

Exactly. This host has one GTX 690 (reported by BOINC manager as two GTX 690s) and a GTX 670.

The maximal number of 690 is also 2?

Yes.

Another question...with one 690 the PC will crunch 2 WU at the same time?

Yes.

What`s the better solutions between 2x690 and 4x680?

It's hard to tell. Both configuration have pros and cons.
2x690:
Pros: the motherboard have to have only two PCIe x16 slots, less PCIe power cables required (four 8-pin)
Cons: less overclockability, larger heat dissipation into the case (considering the factory made cooler)
4x680:
Pros: higher overclockability, the heat dissipated only through the rear grille (considering the factory made cooler)
Cons: the motherboard have to have four PCIe x16 slots, and the PSU have to have four 8-pin and four 6-pin PCIe power cables
I would choose the 2x690, with lager than factory made cooler (or water cooler), and a motherboard with 4 PCIe x16 slots (in this kind of MB the GPUs are 1 slot farther apart, allowing better airflow)

Thank you very much Zoltan :-)

---

Member of Boinc Italy.

[3] NVIDIA GeForce GTX 690 (2047MB) driver: 310.33

This is actually one GTX 690 and another GPU (GTX670 or something).
Boinc reports each individual GPU. A GTX690 has two GPU's so Boinc reports two GTX690's! It's doesn't report the card count, just the GPU count, and it has to call it something, so GTX690 it is. Boinc also reports the first (or biggest/most powerful GPU) and then the number of GPU's. So if you have a GTX680 in there, it will report it as another GTX690.

[6] NVIDIA GeForce GTX 690 (2048MB) driver: 301.42...what`s this? 6 GPU...

---

Member of Boinc Italy.

[6] NVIDIA GeForce GTX 690 (2048MB) driver: 301.42...what`s this? 6 GPU...

The really interesting part is that this host is running Windows 7 Ultimate x64.
I guess that this host has UEFI BIOS. I'm sure that besides crunching there is no use of a 3rd dual GPU card in a PC, because you can't connect the 3rd card with an SLI cable to the other two, since dual GPU cards have only a single SLI connector.

Not new to GPUGrid but recently completed my new rig. It includes 2 GTX690 (intended for GPUGrid) and one GT640 (intended for SETI@HOME) I have one problem with the GPU utilization, hope to get advice from the masters ;)

BOINC recognizes all GPU's in my system:

25/12/2012 03:21:49 p.m. |  | NVIDIA GPU 0: GeForce GTX 690 (driver version 310.70, CUDA version 5.0, compute capability 3.0, 2048MB, 8382371MB available, 3132 GFLOPS peak)
25/12/2012 03:21:49 p.m. |  | NVIDIA GPU 1: GeForce GT 640 (driver version 310.70, CUDA version 5.0, compute capability 3.0, 1024MB, 836MB available, 692 GFLOPS peak)
25/12/2012 03:21:49 p.m. |  | NVIDIA GPU 2: GeForce GTX 690 (driver version 310.70, CUDA version 5.0, compute capability 3.0, 2048MB, 1955MB available, 3132 GFLOPS peak)
25/12/2012 03:21:49 p.m. |  | NVIDIA GPU 3: GeForce GTX 690 (driver version 310.70, CUDA version 5.0, compute capability 3.0, 2048MB, 1955MB available, 3132 GFLOPS peak)
25/12/2012 03:21:49 p.m. |  | NVIDIA GPU 4: GeForce GTX 690 (driver version 310.70, CUDA version 5.0, compute capability 3.0, 2048MB, 1955MB available, 3132 GFLOPS peak)
25/12/2012 03:21:49 p.m. |  | OpenCL: NVIDIA GPU 0: GeForce GTX 690 (driver version 310.70, device version OpenCL 1.1 CUDA, 2048MB, 8382371MB available)
25/12/2012 03:21:49 p.m. |  | OpenCL: NVIDIA GPU 1: GeForce GT 640 (driver version 310.70, device version OpenCL 1.1 CUDA, 1024MB, 836MB available)
25/12/2012 03:21:49 p.m. |  | OpenCL: NVIDIA GPU 2: GeForce GTX 690 (driver version 310.70, device version OpenCL 1.1 CUDA, 2048MB, 1955MB available)
25/12/2012 03:21:49 p.m. |  | OpenCL: NVIDIA GPU 3: GeForce GTX 690 (driver version 310.70, device version OpenCL 1.1 CUDA, 2048MB, 1955MB available)
25/12/2012 03:21:49 p.m. |  | OpenCL: NVIDIA GPU 4: GeForce GTX 690 (driver version 310.70, device version OpenCL 1.1 CUDA, 2048MB, 1955MB available)
25/12/2012 03:21:49 p.m. |  | NVIDIA library reports 5 GPUs
25/12/2012 03:21:49 p.m. |  | No ATI library found.

I have correctly (I think) configured the cc_config file to use the 690 in GPUGrid and the 640 in SETI, this is my config file:

<cc_config>

	<log_flags>
       
		<coproc_debug>1</coproc_debug>

		<sched_op_debug>1</sched_op_debug>

	</log_flags>

	<options>

		<use_all_gpus>1</use_all_gpus>

		<ncpus>0</ncpus>

		<report_results_immediately>1</report_results_immediately>

		<exclude_gpu>
			<url>http://www.gpugrid.net/</url>
				[<device_num>1</device_num>]
		</exclude_gpu>

		<exclude_gpu>
			<url>http://setiathome.berkeley.edu/</url>
				[<device_num>0</device_num>]
		</exclude_gpu>
		<exclude_gpu>
			<url>http://setiathome.berkeley.edu/</url>
				[<device_num>2</device_num>]
		</exclude_gpu>
		<exclude_gpu>
			<url>http://setiathome.berkeley.edu/</url>
				[<device_num>3</device_num>]
		</exclude_gpu>
		<exclude_gpu>
			<url>http://setiathome.berkeley.edu/</url>
				[<device_num>4</device_num>]
		</exclude_gpu>

		<max_file_xfers_per_project>4</max_file_xfers_per_project>

	</options>

</cc_config>

and this is the response I get from BOINC:

25/12/2012 03:21:49 p.m. |  | Config: report completed tasks immediately
25/12/2012 03:21:49 p.m. |  | Config: use all coprocessors
25/12/2012 03:21:49 p.m. | GPUGRID | Config: excluded GPU.  Type: all.  App: all.  Device: 1
25/12/2012 03:21:49 p.m. | SETI@home | Config: excluded GPU.  Type: all.  App: all.  Device: 0
25/12/2012 03:21:49 p.m. | SETI@home | Config: excluded GPU.  Type: all.  App: all.  Device: 2
25/12/2012 03:21:49 p.m. | SETI@home | Config: excluded GPU.  Type: all.  App: all.  Device: 3
25/12/2012 03:21:49 p.m. | SETI@home | Config: excluded GPU.  Type: all.  App: all.  Device: 4

The GT640 is is crunching WUs from SETI, as expected. But my problem is that I only have 2 WUs for GPUGrid. And the curious thing is that one WU is one one card (say the card on PCIe slot 1) and the other is on another card (say the one on PCIe slot 2). This leaves 2 GPU cores idle, one on each card.

I can confirm this by using MSI afterburner, following temps are on the cards:

GPU1 (GTX690) 41 C
GPU2 (GTX690) 82 C
GPU3 (GTX690) 81 C
GPU2 (GTX690) 35 C
GPU2 (GT 640) 43 C

and by manually feeling the exhaust air temps on the cards: the first 690 is expelling hot air only one the "inner" card (inside the case), while the other 690 is expelling hot air on the "outer" (away from the system grille).

What am I doing wrong here?

Thanks for your help!