I purchased a s3420gplx for a server. I am having issues. I appreciate any words of wisdom.
Are there any documents specifying the registered memory to populate all three channels at 1333mhz and not be forced to 800 mhz? The memory configurator software only lists the memory in two channels. If I populate only the 4 slots will it run at 1333? I tried 2x6 kingston from the supported list and it runs 12gb at 800 mhz.
The other is wheather to use the matrix raid ot the esrt2 raid configuration? I have read the support documents and am still confused what is a better fit for my application. What will give the best performance and reliability for a raid 10 volume.(wd black wd2003fyys 2tb x 4)
The last is what is the largest volume that can be created in server 2008 r2 standard. I tried 4-2tb drives and my boot volume is reporting a total of 2tb, I expected to see 4tb. I assume the remaining 2tb needs to be an extended partition in the volume. I thought there might be a hotfix or procedure for this limitation.
Here are the answers to your questions:
1. Xeon 3400 series series processors have two memory channels, with up to three DIMMs per channel. In a 3 DPC configuration, memory speed is limited to 800 MHz. 2 DPC @ 1333 is possible, if you're using single rank or dual rank DIMMs. It is documented clearly in the S3420GP TPS, section 184.108.40.206 (page 24).
2. See Comparison of Intel® ESRT2 and RST (Matrix) RAID. If you need RAID 10 under Windows, both should be OK. Personally I prefer ESRT2 for server applications, because the management utility (RAID Web Console 2) provide advanced functions like remote management, email notification, etc.
3. Currently booting from a volume larger than 2TB is not supported, for either ESRT2 or RST. But you can use a volume larger than 2TB as a data volume. See http://www.intel.com/support/motherboards/server/s5520sc/sb/CS-031582.htm. Also, the following threads might be helpful:
Thanks for the info. The comparison documentation of the two raid chipsets looked very similar.
One more question. Why would Intel put three memory channels on a board that supports 1333, but if you use all of them it reverts to 800? Is it a limitation in the architecture of the board/chipset, or is the third channel strictly if you plan to use unbuffered dimms? This doesn't make sense to me.
It is related to timing issues between the DIMMS on the same channel.
GP has 2 memory channels with 3 dimms on each channel. for a total of 6 dimms.
As the data path lenghens, the timing requirements for the DIMMS becomes more critical.
The option ends up being slow the bus down or increase the tolerance specification on the memory to the point it could become very difficult for memory vendors to produce memory that would work. ( very difficult = higher cost to you and more than what most people are willing to pay )
In pratice, the memory speed difference between 800 and 1333 in dual channal mode is not noticable in most applications.
The difference between memory running at 800 and memory running 1333 I would think would be noticeable. If a Xeon processor has a 1333 fsb, the performance impact of running the memory at 800 mhz, I would think, would be like night and day compared to same hardware running the memory at 1333. Wouldn't the processor would need more clock cycles (@800) to get the same throughput? Am I thinking about this all wrong? I do remember the days where the 800 fsb cpu's ran 400 memory in dual channel. Is this the same idea where 1333 memory (running at 800) is running in dual channel and the combined speed is faster than the fsb of the processor?
Time to stump the professor. I guess I should ask what would be better performance, 2 - 8gb 1333, 4 - 4 gb 1333 or 6 - 2gb 1333 (running at 800).
What is the memory requirement for a this board with a 4 core Xeon 3470 (2.93ghz) supporting 10 users on Windows Server 2008 r2 standard x64
(application/file server / dc). In the old days on Novell servers there was a guideline for system memory per user attached to the server. I remember seeing a recommendation/suggestion for memory per core in my original research, but I can not find it again. I am not running exchange or sql server. (I found that)
I just wanted to get some more theory behind the server hardware and memory configuration from the experts.
Thanks everyone for the replies......
You may try some type of "load-reduced" memory.
Check out Netlist HyperCloud which makes a low-latency memory that allows operation at 1333MHz at 3 DPC - however you need to check whether your motherboard is supported by them explicitly at 1333MHz (or if only 1066MHz at 3 DPC will be achievable with your motherboard).