bikeman

Distinguished
Jan 16, 2002
233
0
18,680
Did anybody read <A HREF="http://www.anandtech.com/showdoc.html?i=1600&p=1" target="_new">Anand's report on Cebit, Part II</A>? And did anybody notice on <A HREF="http://www.anandtech.com/showdoc.html?i=1600&p=4" target="_new">page 4</A> of that article? I did, and I saw this:

<i>"SiS's Pentium 4 chipset SiS655 was also first to be seen at CeBIT. It features two DDR SDRAM memory channels, and supports Intel's future P4 processors with a 133MHz FSB clock (FSB533)."</i>

Hmm ... SiS using nVidia's technique used on the nForce Mobo, now in conjunction with a processor that could actually use the extra bandwidth? As stated in previous threads, the AXP doesn't benefit [more than 10%] of the TwinBank memory architecture, since it's FSB is limited to 64 bit @ 133 DDR = 2.1 GB/sec, which equals a single channel PC2100 DDR-RAM module's bandwidth. The double bandwidth available on the nForce was only useful for the integrated GPU, and maybe for DMA-accesses to the RAM, or AGP-texturing for an 'external' graphics card.
But now, using that very same bandwidth with a processor that has a 64 bit, 100 MHz (and future models, which will be supported according to the quote above, 133 MHz) QDR FSB, implying a 3.2 (or 4.2) GB/sec bandwidth, could lead to FENOMENAL performance improvements in comparison to nowadays available single channel DDR-solutions for the P4, I guess. Would this mean the end of RAMBUS memory, or will they be used in quad-banked configurations in the future (as they will be in Xeon MP-configurations)? Although ... I somewhere read that RAMBUS would be evoving to 32-bit-wide memory modules instead of 16-bit. Let's go and look some things up. I'll be back in a moment ...

Bikeman

<i>Then again, that's just my opinion</i>
 

bikeman

Distinguished
Jan 16, 2002
233
0
18,680
Ok, I'm back. Found some interesting links concerning this topic.

First thing I found is <A HREF="http://www.anandtech.com/chipsets/showdoc.html?i=1588&p=1" target="_new">this</A>. An article on AnandTech.com about the E7500-chipset, a dual channel DDR-solution for Xeons. I guess SiS is not the only one trying out that configuration, although I don't think Intel will get to making dual channel DDR-solutions for the desktop P4's
Then, at TomsHardware, in the dual Xeon vs. dual AMP article, <A HREF="http://www6.tomshardware.com/cpu/02q1/0203131/images/dual18.jpg" target="_new">this picture</A> is shown, with a ... huge amount of RIMM-slots, arranged in a way that really makes one think it's about a quad channel setup being used on the i860-chipset. The text underneath the northbridge photo, on the contrary clearly states <i>"Intel chip for the memory interface (dual channel RDRAM)"</i>
To be sure about this, I visited Intel's site, where I found <A HREF="http://www.intel.com/design/chipsets/860/" target="_new">this document</A>, that clearly states that the memory controller features a dual channel RDRAM setup, though, and this is confusing at a minimal level, there is something called the <i>82803AA RDRAM-based memory repeater hub (MRH-R)</i>, which splits every channel into two seperate ones. This does not improve the bandwidth, though, and is only used in order to make it possible to go over 2 GB (or are there 1 GB RIMM-modules?) of memory.
Then there was the 32-bit thing I read somewhere. When I googled "32 bit rimm", the first link that showed up was one pointing to <A HREF="http://www.rambus.com/rdf/presentations/RIMM3264_06.12.pdf" target="_new">a pdf document</A> from Rambus.com, which gives rather detailed specs of (future?) 32 and even 64-bit wide RAMBUS modules. This does give a bandwidth up to ... 8532 MB/sec ... *SHIVER* ... OK, I gues RAMBUS is not quite dead yet ... if only they find manufacturers that will be using their technology, of course. But at last I found what I read first, namely <A HREF="http://www.anandtech.com/cpu/showdoc.html?i=1590&p=5" target="_new">this page</A> of the report of Anand about IDF day 1. I states that in the future, RDRAM won't be used with a pair of RIMM-modules anymore, but that two 16-bit modules would be implemeted on one RIMM. They call it a 32-bit module, but I don't know wether this is what the document at rambus.com meant.

My conclusion? (Dunno how many of you will have read every link I posted here, or even have made it to the end of this post :wink: ) I think SiS is going to kick ass, if they will be able to implement what they plan in an efficient way (what they proved capable of with their recent 645DX). Intel is working on dual channel DDR-Ram, but I think they target only the high-end market, with prices according to it. RDRAM still has it's potential, but in pure bandwidth they [i.e. PC 1066] will be matched by dual PC 2100 and considering the high latencies encountered in a RDRAM-system ... But of course, who sais that 655 will be targeted at the low- to mid-end market we all belong to ...

Any opinions?

Bikeman

<i>Then again, that's just my opinion</i>
 

bikeman

Distinguished
Jan 16, 2002
233
0
18,680
Hey! Why isn't anybody replying to this thing? :frown:

Anyway, found some new things concerning the topic.

Intel is planning on TWO dual channel DDR-chipsets for the P4 (Xeons or not, that's the question): Granite Bay and Springbay.

Via is planning on a dual channel DDR-chipset for the P4: Via P4X600.

SiS is planning on a dual channel DDR-chipset fot he P4: SiS 655.
I even read somewhere that Abit is going to release mobo's using this chipset in June already. Via mobo's would be out same time. Although I find that hard to beleive ...

So, does anybody have a comment on this? PLEASE?

Bikeman



<i>Then again, that's just my opinion</i>