4x 6core xeon board...

[Airbozo]

This will be the next new build for one of our customers;

Mobo: http://www.supermicro.com/products/motherboard/Xeon7000/7300/X7QCE.cfm

Memory: 64gb (maybe 128gb once they finish testing)

Drives: 3x 1tb

In this chassis; http://www.supermicro.com/products/system/1U/8015/SYS-8015C-T.cfm

Should be fun!

[Luke122]

That's crazy! Imagine the heat generated by that system in that small chassis.. gonna need some serious cooling on that rack. :)

[crenn]

Water cooling?

[SXRguyinMA]

this should be interesting to see :D

[Airbozo]

According to intel and supermicro the heat generated in that 1U chassis with 4 procs is no greater than the same chassis with 2 older xeons. We will see though. The cost of each 6-core xeon proc is $3k!

My main issue with the 1u chassis is the noise of the fans. They are completely unbearable. Each small fan is really 2 fans in one...

http://nihil.rchomepage.com/misc/dellfan/dellfan-01.jpg

[SXRguyinMA]

$3k each! thats redonkulous!

[Airbozo]

$3k each! thats redonkulous!

Hehe I think so too. Let's see, 4 procs x $3000 + 24 4gb Dimms x $230 = $17520 + drives + mobo and chassis. (8gb dims are $1400 each)

The base cost is going to be around $25,000! That does not include my time nor any OS (Redhat), or HBA's...

Thing is, The SuperMicro's are so easy to build it will only take 2 hours once all the parts are staged. Then 2 days for burn in...

Hope they go for it. I want to take some photos and if possible some benchmarks.

[SXRguyinMA]

absolutely insane! benchmarks would be neat :up:

[p0Pe]

/me wonders if this will be his next surfing system. what would be better for browsing MULTIPLE fans in firefox than 192 gb ram -.-

[Airbozo]

Heheh, Wish that were so, but I would spend that cash on a couple 4870x2's and a board that could be overclocked...

...and I would still have money to go on a nice trip.

[TheGreatSatan]

What's the price for the board alone?

[XcOM]

them fans are bloddy LOUD,but once an os starts the generaly slow down, and rem, server equipment always without fail, runs cooler than the desktop equivelent

[AMD Killa]

As soon as I saw that board, my mouth just dropped. If only there was 4 way SLI on that board. What clock speed are the processors? Wow...24 cores and 192GB of RAM. Imagine looking in task manager in windows and seeing graphs for 24 cores. And the RAM usage would probably show as 0% for just windows.

I think installing that many DDR2 sticks would get a bit tedious after a while.

Hopefully your client will have more than 192GB of hard drive space, or that much RAM is a bit of a waste XD.

Imagine setting up four of those as a cluster. 96 cores, and 768GB of RAM :eek: If you used 5 in a cluster, you would have just under a terabyte of RAM. Time for another jaw drop.

I wouldent call this a server, I'd call it a super server. (Well, under industry definitions, I'd call it a server.) What OS you thinking of? What about ubuntu server edition, or is that not the sort of thing you are looking for?

[chaksq]

The base cost is going to be around $25,000! That does not include my time nor any OS (Redhat), or HBA's...


What OS you thinking of? What about ubuntu server edition, or is that not the sort of thing you are looking for?

Personally Ubuntu server edition is not that great for industry from what I have heard. I prefer Fedora/Redhat. If the client has need for a system that expense they can certainly afford Redhat Enterprise Edition, not to mention the Redhat license includes full support.

[monoflap]

What kind of a server is this guy making that he needs all that?

[crenn]

If you've seen how long it takes to compile or render a pixar movie, this is minor.

[Airbozo]

The server will have a small internal drive setup (~300 gb in raid 6) for the OS and applications. No need for massive drive space as most of the data used is on a NAS. The main reason for the large amount of memory is for loading of database indexes to speed up searches. The main reason for that many procs is to allow more instances of the customers application running therefore allowing more clients to connect and search. From what I understand, once the search result is found, it is passed off the the node that actually has the fastest connect to the data.

We are talking hundreds of these nodes around the country. (not the index node, the data node). Part of their software (custom built) determines the location where the data is required and migrates that data to the node closest to the most recent requests. So the first access of that data may be slower than the next customer's access depending on the frequency.

I can follow most technical conversations even if they are way over my head. I learned long ago that sometimes you have to push the "I believe" button until that specific piece of info comes along that ties everything together then "ping", it all makes sense. In this case I felt like a blathering idiot and my head felt like it was going to explode!

Oh and due to the specific nature of the software, nothing except RedHat has the ability to serve up the info as fast as the clients require. Trust me they have tested this for many years. BTW: This company has been in business since the late 1800's...