Motherboard Differentiation

http://www.digitalstormonline.com/comploadsaved.asp?id=432784

Budget: 2k but less is always great.
Primary use:
50% of the time multi-application productivity (normal workload of Photoshop, web dev software, email, multiple browsers, FTP client, etc for 8-12 apps running simultaneously on average)
30% gaming
20% encryption/decryption/security applications
GPGPU intensive applications

Special needs:
Encryption workloads demand FERMI chipset video cards right now. (OpenCL is not yet mature enough to feel comfortable going with an AMD/ATI product)

Multiple GPUs at a lower chipset implementation preferred over a single monolitic card.

I spend, on average, 12-14 hours a day in a 10x13 room with my desktop (this computer), and two laptops. Quiet is good :D

Misc:
Dont need an optical drive.
Dont need an OS (i multi-boot and would kill whatever came on the box as first thing anyway)

Questions:
Should I spend the money to go from eVGA's X58 LE to the 3X SLI? Why?

I don't want the bother or expense of "true" / "full" liquid cooling. What is my best cooling option? Do the closed loop systems perform better than the offered fans?

What does DSO's 3 year warranty cover? Is there any gotcha I need to know (like keep the box or we wont take your PC back to fix it or dont add another video card later or you void everything)?

The DSO RAM doesnt really specify a brand. What brand do you typically go with for 6GB 3-channel DDR3 kits?

Is overclocking the video cards really worth it from a performance vs risk perspective? Do you still include the cards under the 3 year warranty when you overclock them and I leave that configuration alone?

Originally posted by !ender_ [FONT style=": #000000"]check out the guide in the config section, it will answer almost all of these questions, but of course come back if you still have more

While I appreciate the pointer, I have indeed read your guides. Thank you for that effort.

Please also note that your guide on efficiency of SLI configurations is actually outdated. Most games are seeing high ratio gains when going multiple card using the GTX 460.

Further your indications on cards being a poor choice going more than one card does not apply to GPGPU configurations. In such a case, going multi-card to gain large increases in core count is very important.

Further, you might also note that Nvidia has recently introduced 3-monitor configurations which require SLI implementations for surround gaming.

Thus, my motherboard question is still relevant and still unanswered :)

In most of the others I am looking for personal views, and while the guide helps explain things like the warranty, it does not provide others' viewpoints of how the warranty has served them and gotchas they have encountered.

I hope this helps with clarification.

[By the way, the guides that have been put together are a great service to the community. Thank you for putting in the time on that.]

Edited by Clocks - 16 Aug 2010 at 2:51pm

Originally posted by Alex @Clocks, you've made some valid points and I feel we should look into your statements and update the 'Guide' accordingly. @!ender_, what do you think?

Actually he does include a few small modifications which address some of these points, they simply dont offer enough detail to address the matter completely enough :)

The point from Hard OCP is a great one, for example, in that the overhead of even a multi-GTX-460 solution will occasionally experience gameplay challenges owing to the management and synchronization overheads.

I think ender has some great views and valid points but they are somewhat clouded by the "bolt on" approach to adding these details. You may want to invest in a short rewrite of these sections to better organize the thoughts.

the Asetek is louder than the Noctua

Edited by justin.kerr - 16 Aug 2010 at 3:01pm

The Le is great for single cards, crappy for 2 and utter dog poo for 3 cards.. it has locked Pcie lane frequency, unlike the other boards that change with the amount of Pcie occupied slots.

Should I spend the money to go from eVGA's X58 LE to the 3X SLI? Why?

this was actually the originial problem which sparked the creation of the guide, people constantly asking the same questions would get less and less full and accurate answers becuase many members were getting tired of posting all of the same information over and over again. with a lot of the common questions in one place, it leaves room in someone's config thread for more in depth and productive topics, such as your GPGPU question

Edited by !ender_ - 16 Aug 2010 at 3:18pm

Originally posted by Clocks I think ender has some great views and valid points but they are somewhat clouded by the "bolt on" approach to adding these details. You may want to invest in a short rewrite of these sections to better organize the thoughts

depends on the case, most cases, with 3 GPU's you will have to run one at 16, one at 8x and one at 4x.

Thanks Dragoonseal, much appreciated. I think that makes it pretty clear where I need to go for CPU cooling and the motherboard choice.

Originally posted by justin.kerr 16x 8x 4x unless you have a 8 slot case.. like I said above. 5%-10% less performance with a GTX 480 going from 16x to 8x, depending on game, resolution ect... too many variables like I said.

Except it's not, it's only 1%-5%, with just 2% on average. Already linked to benchmarks here.

It is not possible for an X58 chipset based product to host more than 36 lanes.




Edited by Clocks - 16 Aug 2010 at 5:46pm

they can have more but it would involve one or more JMicronor or nf200 to do so

Edited by Raif - 16 Aug 2010 at 6:18pm

my bad just trying to help

Uh. X58 is not a socket.

X58 is a primary chipset and integrated controller for the LGA1366 socketted processors which includes access to the graphical controllers. No matter what other chipsets you have augmenting the capabilities of the system, the data still comes back through the X58 chipset.

The x58 chipset maxes out at 36 lanes of PCI-E connectivity. E.g. You will never have more than 36 "lanes" of bandwidth available in aggregate, EVEN IF you introduce some kind of gimmick to try and make it appear that you are getting additional x16 lanes, ultimately that data is still going through a bottleneck of 36 channels to the system chipset.

And, I would say if I am wrong, help me see why. Show me a board that exceeds this system limit.

Edited by Clocks - 16 Aug 2010 at 6:40pm

might want to look at the Le board, the top slot is not in the same location as your Micro... like I said, build at least on LE build before you spew.

If not sure if the SR-2 can count. That beast is in a whooole different league. lol. It's not even an X58 chipset. However it does have 1366 sockets. Well, you did show a board that exceeded the limit, but's it's not X58, and i think that's what Clocks was asking for.

Originally posted by justin.kerr might want to look at the Le board, the top slot is not in the same location as your Micro... like I said, build at least on LE build before you spew.

Hey there you go, some actual info. See how easy that was? You could of saved yourself like three posts of running around the bush being pompous if you would of just explained from the start that the LE has an extra dead space where the first slot is so everything is shifted down a slot. You'll have to excuse me if I don't have one of every motherboard on hand to personally scrutinize and catch little details like that.

Thank you for correcting me.
  First off, the Classified SR-2 is an Intel 5520 chipset, not a x58 chipset. But you already knew this, so you shouldn't of even brought it up in the first place.

Secondly, the Classified SR-2, or any other Intel 5520 chipset board, still only has 36 PCIe 2.0 lanes.

http://www.intel.com/Assets/PDF/datasheet/321328.pdf



The Classified SR-2 has two Nvidia NF200 SLI bridge chips, this means they advertise their x8/x8/x8/x8 as x16/x16/x16/x16, but it is not, it's just x8 lanes with the extra GPU to GPU SLI bridges. And here's the kicker, NF200 SLI bridge chips only work for Nvidia video cards and nothing else, they do nothing for ATI cards, they nothing for any other add on cards. All NF200 chips do is add latency because all PCIe cards have an extra chip to send data through, x8 lanes on the x58 are not a bottleneck in the least so all the chips manage to do is reduce PCIe speeds.

And what if you use more than 4 PCIe cards? They didn't document it anywhere of course, but this reviewer contacted EVGA to find out, and turns out then all slots run at x8. And remember, NF200 chips, so this is marketing code for all slots run at x4.

And you know what really sucks? I can't even tell if the SR-2 has a single x16 lane at all. The SR-2 spec sheet (page 33) says their "x16" lanes have a bandwidth of up to 4GB/s or 8GB/s concurrent. That's PCIe 2.0 x8 speed ladies and gentlemen, a single PCIe 2.0 lane is 500MB/s or 1GB/s concurrent, a x16 lane is 8GB/s or 16GB/s concurrent. So either that spec sheet is just worded badly or all its lanes are potentially locked into a x8 or lower configuration. Bleh.

So what have we learned? Marketing is pure evil,The EVGA x58 LE is still awesome but don't run triple GPUs on it unless you have a 8 slot or larger case, and calling people names doesn't make you any less wrong.

Originally posted by sirsiddius If anyone cares Hardocp recently did an x16 x8 test. It should still be on their frontpage. The results are actually quite intresting.

Heh, sure is.







Look like within a 2% natural variance to you? Looks like it to me.