Re: What is important is INTC's long term future prospects...
It's the short and medium term that are something of a question. AMD's motherboard nightmare is ending, Intel's Rambus nightmare is just beginning.
You know that FEDEX commercial where some guy is hiding under his desk while his boss is walking up and down the hall calling his name because some vital package missed it's deadline? I've got a feeling that Craig Barret is walking up and down the halls of Intel looking for whoever convinced him that it was a good idea to make Intel completely dependant on the success of Rambus.
=======================================================
theregister.co.uk
Does BX chipset with SDRAM outperform 820 with Rambus?
Just a few hours before a gaggle of British hacks headed out of overheated Palm Springs, we overheard a conversation that stopped us in our tracks.
We were sitting by the poolside bravely quaffing some Michelobs and Banquet Coors, when a group of Inteleers and customers came out on the veranda for a sly cigarette or two.
One customer was asking why it was that the i820 with Direct Rambus memory is outperformed by a BX mobo using SDRAM (synchronous DRAM).
He had been to an Intel presentation where a graph demonstrated that anomaly and had asked the speaker to explain it to him.
The Intel speaker couldn't, but said: "If it says it on the slide, it must be true."
Ahem. Regular readers will be well aware that there is currently a massive shortage of BX parts. Without being conspiratorial about it, there couldn't be a connection between the two, could there?
At IDF, we also picked up a copy of a Platform bandwidth test for the i820 which we hope to take a shrewd look at in the near future.
Note, this is not a benchmark. As Pat "Kicking" Gelsinger said in a press conference earlier in the week: "There are lies, lies and benchmarks". ©
=======================================================
theregister.co.uk
Intel abandons server Rambus efforts
Reliable sources have told The Register that Intel's efforts to implement Rambus technology in its Carmel chipset for standard high volume servers have come to nothing.
Carmel is Intel's server Rambus chipset, which can optionally use the Repeater chipset for machines with very large memory arrays.
Over the last six months, a team of Intel engineers have attempted to design a Carmel chipset based motherboard for a quad server but that project has now been abandoned.
Instead, Chipsetzilla will now use the Reliance chipset for its quad processor, which will use DRAM, instead.
As reported here from Computex in June, engineers are doubtful that large arrays of Rambus RIMMs will ever run at 400MHz (800MT/s). Board firms will also have trouble producing reliable boards to run at 400MHz, the sources added.
One of the problems is that Rambus technology are microwave circuits and engineers at mobo manufacturers do not have the technical skill to produce such designs.
The 354MHz Rambus channels only tipped up on roadmaps over the last three months and specs for the DRCG clock generators use a divider ratio of 8/3.
Translated into system terms, the REFCLK frequency of 100MHz generates a Rambus channel frequency of 267MHz, which does not give any performance gains at all.
But if the REFCLK input is changed to 133MHz, identical to the front side bus (FSB) frequency, the rabbit pulled out of this piece of Intel magic gives a frequency of 355MHz.
This gives a peak channel of 1.42GB/s and there is enough margin to run a reliable Rambus channel by most mobo makers.
Sources added that when Intel tested the first prototype Rambus boards earlier this year, they discovered problems due to intersymbol interference and power plane noise. The divide ratio popped out of nowhere shortly after this discovery.
No-one at Intel was available to comment on these reports on Saturday night at 8 pm. ©
===========================================================
Dan |
