Search
Sign in with
Sign up | Sign in

Why use a server motherboard

Last response: in Systems
Share

Hi forum.

At work we have a process ... compile a ARM-based code to load into a cell phone ... that takes about 3 hours on our current machine:

HP xw6600 Workstation
-- Quad Xeon E5405 2.00 GHz, 2x6 MB L2, 1333MHz FSB
-- Intel 5400 chipset
-- 3 GB DDR2-667 ECC
-- SATA, 500GB

At another location, they managed to get the speed down to 40 minutes using the following configuration:

HP DL380G5 Server
-- Dual Xeon E5335
-- 8Gb FDB PC2-667
-- 6x 146dB 10krpm SAS configured in some kind of a RAID

So now I am trying to build a system to improve speed at my site. I'm looking at duplicating the HP DL380GS Server solution but what bothers me is I cannot find a server motherboard that takes the latest DDR3-1600 memories. So I'm thinking instead of the server solution I will try a variation on the current $4500 System Builder's Marathon:

-- Q9650
-- 8GB DDR3-1600
-- 2 300G 15krpm SAS configured as RAID 1
-- SAS PCI or PCIE card

My thinking is I think the Q9650/DDR3-1600 is higher performing, and I'm not seeing an advantage with having a server motherboard. Can I have the expert opinion of folks in this forum as to whether I'm failing to consider something here?

Thanks so much I would really appreciate your feedback.

More about : server motherboard

Register or log in to remove

What is the PC for mainly? purpose? What are you trying to achieve actually? speed? stability?
Server mobos have different chipset and layout. There are other features that the users' mobos don't have.

I think you should go for DDR2 and save up the cash for other parts of your new build.

Wouldn't the compile speed primarily be related to the available CPU muscle? If so, you'd want a multi-processor setup. The difference in times between the two setups (2 hours vs 40 minutes) implies that whatever you're doing will utilize as many cores as the system can offer... that said... you need a server board that supports multiple processors. Going from a 2 GHz quad core to a 3 GHz quad core should yield you a 50% reduction in time... or 1 hour. If you're happy with that, you're on the right track. If you were aiming for 40 minutes or less, you'd probably want a multi-processor setup.

dori_ansari said:
Hi forum.

At work we have a process ... compile a ARM-based code to load into a cell phone ... that takes about 3 hours on our current machine:

HP xw6600 Workstation
-- Quad Xeon E5405 2.00 GHz, 2x6 MB L2, 1333MHz FSB
-- Intel 5400 chipset
-- 3 GB DDR2-667 ECC
-- SATA, 500GB

At another location, they managed to get the speed down to 40 minutes using the following configuration:

HP DL380G5 Server
-- Dual Xeon E5335
-- 8Gb FDB PC2-667
-- 6x 146dB 10krpm SAS configured in some kind of a RAID

So now I am trying to build a system to improve speed at my site. I'm looking at duplicating the HP DL380GS Server solution but what bothers me is I cannot find a server motherboard that takes the latest DDR3-1600 memories. So I'm thinking instead of the server solution I will try a variation on the current $4500 System Builder's Marathon:

-- Q9650
-- 8GB DDR3-1600
-- 2 300G 15krpm SAS configured as RAID 1
-- SAS PCI or PCIE card

My thinking is I think the Q9650/DDR3-1600 is higher performing, and I'm not seeing an advantage with having a server motherboard. Can I have the expert opinion of folks in this forum as to whether I'm failing to consider something here?

Thanks so much I would really appreciate your feedback.

You could check out the Asus P6T Deluxe (X58) Motherboard for the upcoming release of Intel Nehalem series processor the Core i7, that will be using DDR3 Tri channel ram. The P6T Deluxe features 2 SAS ports.
http://www.cpu3d.com/review/6163-1/preview-asus-p6t-deluxe-x58-motherboard/introduction.html

The DDR3 Tri Channel ram is starting to show up on Newegg. Most of the main Ram manufactors will be coming out with them shortly, i.e. Corsair, G.Skill, Patriot and Kingston.
http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2010170147+1052129233+1052345114&Configurator=&Subcategory=147&description=&Ntk=&SpeTabStoreType=&srchInDesc=

The lowest priced Core i7 which will be 2.66ghz will be roughly $280 and the Asus P6T will most likely be around $350. Which should be out within a month.



Related ressources
Ask the community
Homebuilt system Authority

You'll have an interesting challenge building a server that outperforms the HP DL380 G5. Since you already know that it performs very well for that application, why not buy another one with the same or even better hardware? 2 15K disks in RAID1 will not be as fast as 6 10K drives in RAID5 on the DL380 G5. More disks in a RAID usually improve performance, particularly when a caching controller is used.

Before you do anything else, you should determine where the bottleneck is. Is is the CPU, the memory (swapping can become a serious performance bottleneck) or disk I/O? Increasing memory and disk I/O performance on that workstation is rather easy.

Most code crunching programs scale well with more cores. If I were you I'd go for a dual socket with some moderately priced quadcores. The ram speed wont make a world of difference. And make sure and pick up a faster hdd. That will help a bit too.

Biggest difference I see between the two builds is the raid array and ram size. Servers are usually designed to better throughput over strict number crunching....

so, I guess I agree with ghislaing - do some monitoring to find the bottleneck first so that you can be sure to solve the problem.

Hey thanks for all the advice. I guess the honest truth is I don't know what the bottleneck is. During compile it goes through at least a few hundred thousand modules. Then during the link phase the memory usage grows slowly to > 2Gig.

I'm not sure how to find out the bottleneck. But I think more of you are advising me to stick to the server configuration (or wait for the new Core i7 hardware.

Thanks all!
Homebuilt system Authority

What O/S are you using? What's the average and peak CPU utilization while compiling? If the disks are very busy, i.e., the disk activity light is almost constantly on, then that's probably where your bottleneck is. If a compile goes through a few hundred thousand modules, then a lot of I/O occurs and the fast disk subsystem on the DL380 G5 server might be a major advantage. In other words, if the CPU isn't too busy and memory utilization is low enough to prevent swapping, then using more powerfull CPUs or additional memory is not going to help that much.

GhislainG said:
What O/S are you using?


Right now I'm using XP. But I will be going to Win 2k Server because that's what the other group (using the HP DL380) is using.

What's the average and peak CPU utilization while compiling? If the disks are very busy, i.e., the disk activity light is almost constantly on, then that's probably where your bottleneck is. said:
What's the average and peak CPU utilization while compiling? If the disks are very busy, i.e., the disk activity light is almost constantly on, then that's probably where your bottleneck is.


Actually I looked at the CPU utilization on my quad and not a single time during compile is it very high on any of the 4 cores. And the disks are very busy. At the end (after compile at least 500,000 c, c++ and dsp firwmare code) there is a long link phase where the CPU's stay at nearly 0, memory usage climb to over 2Gig and the disks are very busy.

If a compile goes through a few hundred thousand modules, then a lot of I/O occurs and the fast disk subsystem on the DL380 G5 server might be a major advantage. said:
If a compile goes through a few hundred thousand modules, then a lot of I/O occurs and the fast disk subsystem on the DL380 G5 server might be a major advantage.


Yes! Exactly!

In other words, if the CPU isn't too busy and memory utilization is low enough to prevent swapping, then using more powerfull CPUs or additional memory is not going to help that much. said:
In other words, if the CPU isn't too busy and memory utilization is low enough to prevent swapping, then using more powerfull CPUs or additional memory is not going to help that much.


OK! I understand what you are saying now! So the main thing is the disk configuration!

In that case I have another question. If I add a SAS controller PCI (or PCIe) card to the $4500 System Builder Marathon system (the Q9650/DDR3-1600 system) do you think it will perform less than the server solution?

Thanks
Homebuilt system Authority

The DL380 G5 uses much slower memory than DDR3 and it performs very well. About Quad core CPUs, we've been using dual and quad-CPU servers with a 32-bit OS for a long time. If you add a good SAS controller and several disks to the $4500 System Builder Marathon system, it could be as fast as the DL380 G5. It will also be as expensive and a gaming PC will never be as stable as a server.

You already determined that the bottleneck is disk I/O. If you add a few 10K or 15K SAS disks to your xw6600 Workstation (a kit is available), it would compile much faster.

If you decide to build a new system with 8GB of memory, don't use Windows 2000 Server as it's obsolete (unless your compiler won't run on a newer OS). At a minimum you should go with Windows Server 2003 Enterprise Edition. Either Windows Server 2008 Standard 64-bit or Windows Server 2008 Enterprise would be my choice. Since it has 8 GB of memory, I presume that the DL380 G5 runs Windows 2000 Advanced Server.
Register or log in to remove
 
Ask the community
Read discussions in other Systems categories
!