Overkil

If quad-core procs still were Socket 940, perhaps. Most likely that board wants 4x 8xx series Opterons on it.
 
I can outdo that.. though i dont think quad cores would be available on current sockets, but even with dual cores, u still get 16.

Motherboard
Daughterboard

EDIT: And yes, the OP's link is for 800 series opterons MB
 
Or, you could try this related board: http://www.tyan.com/products/html/thunderk8qe.html

Here's a pic of the thing in action, with some article to go a long with it: http://www.theinquirer.net/?article=30302

Hopefully now that dual-cores are becoming commonplace we'll see games that can actually take advantage of a lot of threads. Then again, they'll probably just optimizes for two threads because most boxes aren't multi-socket. If we ever *do* see a game that can handle 16 threads, however, this thing starts looking mighty nice.
 
Doorbasher said:
If we ever *do* see a game that can handle 16 threads, however, this thing starts looking mighty nice.
Click to expand...
It looks great already; there are other interesting applications for computers aside from games.
 
Stang Man said:
PCI-X is the way to go in a server board
Click to expand...
I'm pretty sure he was talking about PCIe x16 slots for SLi. They make RAID controllers in x4 now though, I think.
 
*ringring*
Reality calling.

That board is utter shit.

What's wrong with it? Let's see, the chipset suffers chronic overheat. The SCSI controller suffers chronic overheat. You cannot fully open the DIMM sockets because they're too close physically - now THERE'S a real smart move. The traces are too long on the SCSI controller and it has trouble maintaining U320 as a result. SATA controller had too many problems to list. PCI-X B can't go over 100MHz stable. The list just goes on and on and on. I had to do workarounds, because that board used to be $2K before the case with the N+1 1650W PSUs - and that was the single-core No-Rev-E support board. And trust me, quad 172mm high pressure Delta blowers, 14000RPM 40mm's and twin 92mm Deltas is more than enough cooling.

Oh, and they originally promised to support Dual-Core with a BIOS flash on the S4882. Guess what? NOPE.

http://www.supermicro.com/Aplus/motherboard/Opteron/nForce/H8QC8.cfm
http://www.supermicro.com/Aplus/motherboard/Opteron/nForce/H8QCE.cfm
http://www.iwill.net/product_2.asp?p_id=93&sp=Y
http://www.iwill.net/product_2.asp?p_id=94&sp=Y
http://www.iwill.net/product_2.asp?p_id=107&sp=Y

THOSE are monsters.

... and yes the H8502 has PCIe16x. Two of them.
 
holy shit, that QK8S needs a 6 Pin and an 8 Pin plug, PLUS the 24 pin, wonder if its running a quad CPU server
rolleyes.gif



edit: is this Abusive not the brightest crayon in the box? all the posts I've seen from him are terrible.
 
DaCoOlNeSs said:
holy shit, that QK8S needs a 6 Pin and an 8 Pin plug, PLUS the 24 pin, wonder if its running a quad CPU server
rolleyes.gif
Click to expand...

Actually, no.
The K8QS requires 24+8+8+4+4.
I'm not joking. It will not post unless every last one of those is connected. The additional Molex is a halfassed hack for the dualcore because they didn't put enough VRM horsepower into the S4882 originally. Under maximum CPU load the VRMs start dropping volts because they're not good to 90W!
Just a load of unpleasant surprises.

edit: is this Abusive not the brightest crayon in the box? all the posts I've seen from him are terrible.
Click to expand...

Having seen some of his other spam elsewhere, I'm gonna go with not the sharpest crayon in the box as being an improvement. Clue levels are clearly in the negatives.
 
Stang Man said:
PCI-X is the way to go in a server board
Click to expand...

PCI-E has lower latency and higher bandwidth, how is this not better than pci-x?

Especially when you begin to consider the massive bandwidth requirements some servers are needing for scsi arrays. PCI-X cannot touch PCI-E in that market.

Tha fastest pci-x 133 hits a bit over 1GB in bandwidth, pci-x 566(not out yet) is stated to hit about 4.3GB in total bandwidth. All on a shared bus type setup.

A 16x pci-e slots offers 4GB(8GB full duplex) bandwidth which would be fantastic on a high end scsi controller or gigabit nic. It's a point to point bus, so no shared bandwidth between slots.

I fail to see how PCI-X is better for servers other than the fact it's an established technology.
 
defakto said:
A 16x pci-e slots offers 4GB(8GB full duplex) bandwidth which would be fantastic on a high end scsi controller or gigabit nic.
Click to expand...
Well, no arguments except for the numbers. 4GB/s is the signaling rate*channels, true, but there's 8/10b coding, so you're really only sending 3.2GB/s of data.

And why would you need 2 gbit*16=32gbit for gigabit network, or even 10gig network? x4 would be plenty for a server-type 4-port giganic (or 8 if you could physically fit that many) and almost enough for a 10gig link. Other than physically holding the card up, there's no need for a bigger slot than that for network.

 
unhappy_mage said:
Well, no arguments except for the numbers. 4GB/s is the signaling rate*channels, true, but there's 8/10b coding, so you're really only sending 3.2GB/s of data.

And why would you need 2 gbit*16=32gbit for gigabit network, or even 10gig network? x4 would be plenty for a server-type 4-port giganic (or 8 if you could physically fit that many) and almost enough for a 10gig link. Other than physically holding the card up, there's no need for a bigger slot than that for network.

Click to expand...

Yes, the option of a gigbit nic on x16 would be a waste. It was a bad example as most gigabit nics on pci-e are x1 slots.

I'm also aware that 4GB(not bits, bytes) is the maximum, theoretical throughput, without signaling taken into account. However, the same applies to PCI-X as well because there will always be overheard in the system for error checking and what not.

Still however, your arguments only enforce mine that pci-e is a much better alternative to the server market because it has excess bandwidth which would allow for higher end hardware later on without the need for a massive upgrade to the server.
 
Oh, I'm not arguing that. It's been needed for a while, and I'll be surprised if it only goes to x16. The need for more bandwidth isn't going to stop any time soon. I can't think of an application for more than 3.2GB/s of bandwidth, but build it and they will come ;)

 
I've seen some pretty massive scsi setups that could push 3.2GB/s easy. Just need the controller that could handle it.
 
unhappy_mage said:
Well, no arguments except for the numbers. 4GB/s is the signaling rate*channels, true, but there's 8/10b coding, so you're really only sending 3.2GB/s of data.
Click to expand...

250MB per; but your signalling math is actually off there, mage. It's 2.96GB/s-3.33GB/s; we're not hard drive manufacturers. 250 * 16 = 4000MB = 3.906GB. ;)

And why would you need 2 gbit*16=32gbit for gigabit network, or even 10gig network? x4 would be plenty for a server-type 4-port giganic (or 8 if you could physically fit that many) and almost enough for a 10gig link. Other than physically holding the card up, there's no need for a bigger slot than that for network.
Click to expand...

You can physically fit 8 with a mezzanine easily, but no, your combined latency makes things ugly. 16x is overkill for a quad-port, sure. However, 4 lanes gets you 1GB minus overhead call it 750MB times 8 gives you about 6Gbit, except you're also latency stacking. Bus + Cable + Distant End, which tends to cut in very deeply. It's nowhere near enough for a 10Gbit link in the end, unless you're talking FC-AL. (But HTX is out there now for InfiniBand, so why would you bother?)

Other fun fact; 4 lanes is more than enough for even the monstrous LSI Ultra320-2E when fed the fastest possible setup you can get it. Ultra320 is, well, 320MB/s. That's 1.28 lanes per channel. Call it 3 for 2 channels, which more than accounts for your overhead.You don't start running into problems until you're at quad channel or FC-AL.
FC-AL is 1/2/4Gbit; Qlogic QLE is a 4-lane; 4 lanes gives you ~6.4Gbit. That's enough for 1xFC4G, but not 2xFC4G. You need a full 8 lanes for 2xFC4G. Same for U320; you'll need full 8x for 4 channels.

So I probably shouldn't mention that I'm pushing triple-16x physical in 16x/4x/16x. With five PCI-X 133's backing it up. :)
 
AreEss said:
250MB per; but your signalling math is actually off there, mage. It's 2.96GB/s-3.33GB/s; we're not hard drive manufacturers. 250 * 16 = 4000MB = 3.906GB. ;)
Click to expand...
Well, if you're using "thousands of MiB" to mean GB... I've given up on expecting a definition of GB that works everywhere. Sure, I know, GiB versus GB, but nobody actually believes you're doing that. GB means billions, not 2<<30s.
AreEss said:
You can physically fit 8 with a mezzanine easily, but no, your combined latency makes things ugly. 16x is overkill for a quad-port, sure. However, 4 lanes gets you 1GB minus overhead call it 750MB times 8 gives you about 6Gbit, except you're also latency stacking. Bus + Cable + Distant End, which tends to cut in very deeply. It's nowhere near enough for a 10Gbit link in the end, unless you're talking FC-AL. (But HTX is out there now for InfiniBand, so why would you bother?)
Click to expand...
Just hypothetically speaking here... Pci express is a switched architecture, right? So each lane of that x16 slot is identical to each other. Why not have a card in x16 size with 8 seperate 2-lane devices, each of which talks to 2 lanes and produces gigE on the other end? That would be a really cool thing to see... you could do it with VGA, too, for large multi-monitor displays, if you had a small enough VGA chip.
AreEss said:
So I probably shouldn't mention that I'm pushing triple-16x physical in 16x/4x/16x. With five PCI-X 133's backing it up. :)
Click to expand...
Pushing in what sense? Using to capacity? Recommending to manufacturers? Selling on street corners? :p

 
ValeX said:
Can you run that board with less than 4 processors? That would be a kick ass board if it had PCI-E 16x for SLI or even Xfire.....damn, huge power potential there.
Click to expand...

In general, these boards post and run with less than all CPUs, but you will have devices missing, as busses will be connected to several CPUs and if you don't have that CPU in you don't have that bus.

You need to look at the block diagram that they publish with the board to find out what CPU slots have to be manned to get which device. Unfortunately the diagrams are often wrong :)

Both Tyan and Supermicro have PCIe boards with 4 sockets. On the Supermicro there's one x8 slot in a x16 format. If you have a card with two GPUs on one card that's not too bad.
 
-=ABUSIVE-69=- said:
imagine this with quad core procs :confused: 16 cores, :eek: :eek:
Click to expand...

Actually boards like the one you show have slots for daughterboards. The daughterboard adds 4 CPU and 16 RAM slots. So you can have 16 cores today.

The availability of the daughterboards can be spotty, not to mention the mess it is to get a PSU for these things even with 8 cores.

%%

On the PCIe/PCI-X discussion: the simple problem is that there are many known-reliable PCI-X I/O controllers for whatever your favorite disk, tape robot or netowork interface is. Known-good PCIe cards are rare. And you are screwed if you need some crazy new interface to your new disk subsystem that doesn't happen to have a PCIe controller yet.
 
You must log in or register to reply here.
Back
Top