Post of the Day
January 30, 1998
From our Intel Board (AOL)
Subject: Cacheing In and slaying the golden goose
I am puzzled, no dumbfounded by Intel's decision to remove the cache in it's Covington processors. Have we learned nothing from the IBM PC jr. fiasco of a decade ago or so?
Why is cache so important to MPU (micro processing unit) performance? Here is a little info which may prove helpful to those not technically on top of this issue.
Generally speaking there are 2 types of memory inside your computer system. D-RAM, meaning Dynamic Random Access Memory and S-RAM for Static Random Access Memory.
DRAM is created using a special type of transistor known as a MOSFET. They are very small and require virtually no support in order to hold a BIT of information. As a result, you can pack many seperate memory cells on a chip using DRAM technology. When I was a young engineer working for IBM our goal was to place 1 million such memory cells on a single chip. (At that time, 16K DRAM was the highest commercially available devices) Now of course there are some memory devices with over 256 million such memory cells.
SRAM devices require a more complex structure to prodoce a memory cell. Generally SRAM devices use Bi-Polar transistors and also need several of these transistors to complete a single memory cell. The tremendous advantage that SRAM has over DRAM is that it is significantly faster than DRAM. So you get less cells but they are faster.
SRAM also requires much more power to operate than DRAM and also produces more heat which is always a threat to computer operation. So there is a compromise inside your computer.
For the majority of programs that your computer is likely to use, much of it is just waiting to be called upon by the processor. These instructions are stored in the systems DRAM.
But there are some programs which are used several hundred times a second. These include the computers instructions to maintain contact with it's peripherals like keyboards hard drives and video as well as internal error checking. If these instruction had to be retrieved from the slower DRAM, then the computer microprocessor would constantly be wasting time waiting for a response from the slow memory.
By adding on a cache of SRAM directly on the cpu device, Intel and other MPU manufacturer's allow the instructions which the computer uses the most to be stored in a location where there is virtually no wait. Thus speeding up the device in the areas where it spends most of it's time.
The faster that microprocessors operate, the faster the memory devices should be as well. Sad to say, but that is not the way it is. Memory devices today are still running at the same speeds they did when we were using them in 386 computers. That is why the issue of a cache of fast RAM is so important to overall performance.
It is true that MPU specific functions such as floating point calculations will be faster in the 266 MHz Covington chip from Intel, but that will not be what the majority of sub $1000 PC buyers will use it for. When placed head to head with similar rated product from Cyrix and AMD, these devices will likely outperform the Intel offering in every critical area, thus lending credibility to AMD and Cyrix and causing many less astute but important PC buyers to conclude that Intel has offerred an inferior product.
What astounds me is the timing for such a move. It would seem to me that the cost of retrofitting and re-tooling and changing the "masks" for these devices will exceed the cost of Intel just releasing the superior product that is already established.
Just so you know, the actual costs of manufacturing the full product and this Pentium jr. to Intel are basically IDENTICAL. In fact it is cheaper for them NOT to proceed with Covington. There is NO ADDITIONAL cost to just release the 266 PII at a lower price to compete in the sub 1000 market than to release this suspect device.
Intel is clearly hoping that "Intel inside" will sway the buying public to purchase the inferior product. I think this is a mistake. Computer buyers of today are lightyears ahead of the crowd jst 3 years ago and have clearly demonstrated that they are not logo centric or impressed by dancing purple bunny suits.
I also think that Intel releasing a 333 MHz device and yet promising a 350 device "soon" is another slash at the neck of the golden goose.
MANY, and I mean many, many, many, many, many MAJOR buyers are reluctant to invest dollars in new tech now because of this. The amount of time from havig bragging rights to the fastest PC has gone from 1 year to 3 months now to a few weeks. It is absurd to imagine that someone who has no specific need to be 10% faster is about to plunk down thousands now on a machine which will be "slow" in another month or so.
When I worked at IBM, there was the constant challenge of when to release our "latest and greatest" at the best time from a marketing perspective. The 1 meg devices I mentioned earlier were ready for production almost a year before we actually started selling them.
OK enough about that. I hope this info has helped some understand the issue at hand here.
As to my own feelings about Intel's marketing moves, well they are my opinions. I am still holding my shares of Intel right now. And my cost basis in the stock is now around the $18 range thanks to the covered puts I have been selling for the past few months and now some covered calls. I am actually thinking of selling all my shares of Intel tomorrow, if not I will sell the Feb 80 call first thing. Still have to think it out. I want to own Intel, but I also feel I could also sell some calls here and end up keeping the money. Either case, I will buy puts on mighty Intel tomorrow. As I did today.
Still on the next dip, I may start selling a whole new batch of puts. I am not decided on that. But I am very decided that Intel will not be at 100 in a few weeks or months. Again, that is my opinion.