Conference ricks::dechips

643.0. "Intel & RAMBUS Team up - Comments?" by CSC64::D_DONOVAN (SummaNulla(The High Point of Nothing)) Wed Jan 29 1997 13:18

	http://www.pcmag.com/news/trends/t970122a.htm

This article at PC Magazine's Web Site caught my eye and was
wondering if this can be construed as a future "opportunity"
or a "threat"?

Dennis

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Memory of the Future 


Intel and Rambus team up on high-speed memory design. 

(1/22/97) -- PC system and processor designers
have known for some time that existing memory
architectures will in short order be too slow to
keep up with microprocessor speeds. Now that
Intel has made a deal to license memory
designs from Rambus a developer of a
super-high-speed memory interface technology,
Intel has effectively anointed Rambus technology as the main
memory architecture of the future. Together, Intel and Rambus
lan to design the next generation of Rambus dynamic memory,
called nDRAM, which could reach 1.6 gigabyte per second by
1999. 

SDRAM, the current performance leader among main memory
architectures, is expected to run into serious technical hurdles as
system-bus speeds surpass 100 MHz in the next couple of years.
Today, Rambus ships RDRAM at 600 MHz and plans to deliver
speed increases on a semi-regular basis. 

Big Breaks 
Founded in 1990, Rambus has received praise from the technical
community for its 600-megabytes-per-second DRAM interface,
RDRAM, which is designed to run ten times as fast as
conventional DRAM interfaces. But the company has won few
actual contracts for the product. Last year, Rambus had its first big
break with Nintendo 64, the RISC-based game machine, which
during last year's holiday shopping season posted sales second
only to Tickle Me Elmo. 

RDRAM has always had industry-leading memory bandwidth and
has made great strides to minimize the one weakness in the
architecture--high latency prior to the initial transfer of data. Intel
and Rambus are now shooting for 1.6GBps nDRAM. According to
industry newsletter Microprocessor Report, Intel's upcoming P7
processor, code-named Merced, has a voracious appetite for
memory, and Intel needs to keep the rest of the system moving
fast enough to make Merced look good. Intel expects Merced to
be ready in about a year and expects nDRAM to be the
mainstream memory architecture by 1999. 

The only potential competitor to RDRAM is SyncLink DRAM, a
specification that semiconductor companies are currently
hammering out. Intel's endorsement of RDRAM appears to doom
SyncLink--at least as a high-volume main memory
architecture.--Larry Seltzer 

re .0

This is gonna be tricky. Like SDRAM, RDRAM will be expensive until the
first high volume use on Intel systems. We will eventually need to use RDRAM,
or whatever RAM technology Intel dictates. We can continue to use the older
technologies if they provide adequate bandwidth. If we adopt RDRAM too
early, we will pay a substantial price penalty for our main memory chips.
If we adopt RDRAM too late, we may have a memory bandwidth disadvantage and
we will be seen as technology laggards.

On SDRAM, we have executed the transition fairly well. We now have two 
product families that use SDRAM (the Miata workstations and the Rawhide
servers) and much of the remaining product is being transitioned to these
platforms. The price of SDRAM is within 10% of EDO RAM already, so we are not
paying a serious penalty, even though SDRAM isn't the mainstream Pentium
Pro memory yet.

If the RDRAM transition goes as smoothly, we won't be hurt by this. If we
can squeeze more usable bandwidth out of RDRAMs than Intel does, it could
even help us.

--RS