Re: h/w and s/w remap of 4 GB
From: General Schvantzkoph (schvantzkoph_at_yahoo.com)
Date: 09/27/05
- Next message: John-Paul Stewart: "Re: what if I double the ram 1 to 2 g"
- Previous message: serge: "what if I double the ram 1 to 2 g"
- In reply to:(deleted message) Tatanka Yotanka: "Re: h/w and s/w remap of 4 GB"
- Next in thread: Tatanka Yotanka: "Re: h/w and s/w remap of 4 GB"
- Reply:(deleted message) Tatanka Yotanka: "Re: h/w and s/w remap of 4 GB"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Tue, 27 Sep 2005 09:09:39 -0400
On Tue, 27 Sep 2005 13:30:50 +0200, Tatanka Yotanka wrote:
> On Mon, 26 Sep 2005 19:47:48 -0400, General Schvantzkoph wrote:
>
>>> Sorry this maybe doesn't help the op's issue, but I have some
>>> observations.
>>>
>>> 1) The handbook of MSI K8N states that (page E-2-9):
>>>
>>> "Due to the South Bridge resource deployment, the system density will
>>> only be detected up to 3+GB (not full 4GB) when each DIMM is installed
>>> with a 1GB memory module"
>>
>> If you enable the memory hole remapping it sees all 4G of RAM. Also as
>> long as you use a 2.6.12 or later kernel Linux seems to see all of it.
>> With earlier kernels you can put a mem=5G on the boot line in grub.conf
>> and that seems to make it work also.
>>
> OK, don't shoot on me ;-). I'm just reporting it's manual.
> AFAIK it refers to first versions. They have modified their K8N series,
> with flatter condenser as well in order to allow mounting of larger
> heatsinks. Seems they have corrected the memory problem as well.
>
>> The problem with the 3800, 4200 and 4600 chips is that they have 1/2M
>> caches. I've found that cache size makes a huge difference for the
>> things that I do, Verilog simulations and FPGA place and routes, on
>> NCVerilog the difference is 2 to 1. I'm sure that there are many
>> applications that aren't as caches sensitive and for those you might be
>> happy with a 3800+. As for main memory speed, I found that the
>> difference between running the DDR interface at 200MHz and 167MHz was
>> about 5%. I don't know if using DDR4000 RAM would allow you to to run at
>> 200MHz with a quad double sided DIMM system but I wasn't able to get my
>> system stable with OCZ DDR3200 RAM at that speed. When I first got my
>> system I attempted to overclock it to see what margins I had, it wasn't
>> much. The system ran OK with a 5% overclock, when I tried 15% I couldn't
>> even access the BIOS (had to do a CMOS reset to get the system to boot
>> again). The 3800+ is the bottom of the line chip so it probably has more
>> margin then the 4400+ but you shouldn't count on it.
>
> Thanks. Do you have a comparison with a, say, 3000+ single core, and a
> common usage without verilog simulations? I have just read the opposite,
> i.e. that cache usually doesn't count much in a workstation usage.
These are my results for the 3800+ and the 3400+, I haven't added the
4400+ to the page but they were as expected, i.e. the 4400+ has dual
memory channels but is otherwise just two 3400+ cores on the same die
which means that each individual core outperforms a 3400+ by about 10%.
For GCC and XST the bigger cache on the 3400+ make the performance equal
to the 3800+ which is one step up in clock speed and has dual memory
channels vs a single channel on the 3400+. It's NCVerilog that extremely
sensitive to cache size, thre the performance is two to one. Someday when
I have some free time I'll do another set of benchmarks and include the
4400+.
http://www.polybus.com/linux_hardware/index.htm
- Next message: John-Paul Stewart: "Re: what if I double the ram 1 to 2 g"
- Previous message: serge: "what if I double the ram 1 to 2 g"
- In reply to:(deleted message) Tatanka Yotanka: "Re: h/w and s/w remap of 4 GB"
- Next in thread: Tatanka Yotanka: "Re: h/w and s/w remap of 4 GB"
- Reply:(deleted message) Tatanka Yotanka: "Re: h/w and s/w remap of 4 GB"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
