Demystifying PC 100 memory specifications


For questions about your generic "PC-100" mother board please try your mother board manufacturer at http://www.pcchips.com/   or the vendor that you purchased it from. PC-100 is a specification, not a brand name.

Myth: There is no economical way to protect from malicious hacker attacks. | click here for the truth |

PC133 PC2100 PC1600 RDRAM  What do all the new names mean?
PC133 SDRAM is just about the same as PC100 SDRAM except for the bus speed is 133 MHz instead of 100 MHz. That is about one third faster. A CAS latency of 2 is pretty much standard for most PC133 memory. Memory with a CAS 3 latency is a few per cent slower than CAS 2 memory. PC 2100 is the DDR (double data rate) version of PC133. PC1600 is the DDR version of PC100. PC2100 is only a few dollars more than PC1600 so it gives you the most value for your money, even if you are going to put it with a motherboard and CPU with a front side bus speed of 200 MHz for now. The effective speed of double data rate PC2100's 133 MHz clock is 266MHz because data is transfered on both the rising edge and the falling edge of the 133 MHz clock. As far as RDRAM goes, just say NO. The only way to get any real speed out of RDRAM is to use a bank of four modules. As if they wern't expensive enough one at a time. The latency of RDRAM is so horribly long that in most stuations it more than eats up any clock speed advantage. One of the worst features of RDRAM is that it can only transfer 16 bits at a time (16 bits = two 8 bit data Bytes). SDRAM transfers data in blocks of 64 bits at a time (64 bits = eight 8 bit data Bytes). So when they jump up and down about the effective speed of 800MHz RDRAM that only means a peak data transfer rate of 1600 MB/s, that is 800 MHz times 2 Bytes at a time. The term PC2100 comes from the peak data rate of 2100 MB/s, or 266 MHz times 8 Bytes  = 2128MB/s A CAS latency of 2.5 is the usual for both PC2100 and PC1600 SDRAM  PC100 and PC133 both use the same old 168 pin SDRAM socket. PC2100 and PC1600 SDRAM have 184 pins. RDRAM uses RIMM sockets. Have fun!

F A Q  How to tell the real PC100 Enchilada  


Computer Memory Market is Undergoing a Transition Phase.
 I am not suprised at all to have just heard that Hitachi and Lucky Gold Star have been sanctioned for dumping. Rambus® dRDRAM® RIMM's® are supposed to be shipping in sample quantities. I should see them here in the next few weeks or next month. Some PC-66 memory has been hard to find. Mosel Vitalic is now reputed to be shipping the fastest PC100 memory available, significantly faster than even Samsung or NEC. The future of PC100-222 Cas 2 memory is in doubt with the release of PC100 dimms with PLL and Register. Cas 3 modules seem to be fast enough for most folks. The new registered dimms are not compatible with other PC100 dimms. Micro$oft now recommends a minimum of 64MB for Windows 95® or for the $100 bug fix otherwise known as Windows 98®. I don't know myself, I just said fooey on the whole mess and am now running 256MB of Mosel Vitalic PC100-222 SDRAM in my 350MHz AMD system with a 100 MHz bus. If I reboot once a day I don't hardly crash at all anymore ever. Stoooopid computers.

  Anatomy of a true PC100 dimm.

Note the PC100 lable. The -222 indicates a cas 2 @100MHz rating. Notice the manufacturer's logo. Viking has been building grade A oem memory modules for ten years and our other factory has for twelve years now.

 Note the source resistor packs RP5 and RP6 in the picture to the left. Note the little dipsy doodle zig-zag in the trace to the right of the U6 designator text. This is part of a printed circuit layout utilizing equalized trace lengths.
  Note the spd eeprom chip in the picture below.


  Both of our factories use only grade A quality components and ISO-9001 or ISO-9002 manufacturing procedures that ensure that quality is built in, not just something tested for after the modules are built.

 Demand is up. Prices have rebounded slightly.

Samsung makes the best chips, but I'm having a bit of trouble getting them off and on. They evidently went ahead and shut down for a few weeks here and there like they said they were going to do. For now NEC, and Toshiba are it for same day delivery for the most part, and Micron in the cas3 parts. If you need Samsung only you may expect to pay a 5% or so premium. Please email us or call first, it may also take one to three days extra.  I had one Samsung 128MB 16x72 cas 2 available today, 31 July, and will see what they have for me on Monday.

Asian Monetary Crisis and Deflationary Cycles

Bottom line. There has never been a better time to add memory to your computer. Many prices have dropped in half over the last few months. The raging price plummet has leveled off for now and has rebounded slightly. Hitachi has left the DRAM market all together. Samsung will be shutting down for a few days each month till things get better. Fujitsu has closed a European fab. Thank goodness, I stock nothing and always ship factory direct.

The Cat is Out of the Bag

I was not even attempting to get this guy or anything, but it's kinda funny how the truth always comes out. So here's the story. As a matter of course, I sell memory to a couple of customers who do what is called competitive analysis. You know, folks that get paid to fry chips and fun stuff like that. This isn't like industrial espionage or anything. Everybody in the business of making memory chips does this as a matter of course, to help them keep up with the Jones's. So I asked one of my "competitors", who by the way wanted to sell me memory and was happy to tell me what I wanted to know a couple of weeks ago, what brands of chip that he was selling this week? That's how I buy for the burnt chip guys, I get them samples of everything out there. He has me explain why I want to know what brand again, and then he tells me that he is not interested. So I ask him, "you aren't interested in selling me memory dimms?" He again said, "no." I'm kinda incredulous by this point, but just tell him, "OK, thanks, bye" and hang up the phone. I don't know about you, but that makes me feel kinda suspicious. I've been doing this kind of thing for years now, without anything like that ever happening. What's he got to hide anyway? I'm being up front about this, it's not like I'm plotting against him or anything. I'm in the business of buying and selling memory. I thought that he was too, I must have been mistaken.

Routine Bait and Switch

OK, here is what we are up against. They take "-7" or "-6" chips many of which are really -10 chips if they were honest. That one had me going for a while. They put them on a dimm together with a spd eeprom that says to run them like  true -8's and sometimes they work and sometimes they don't. This is where all of the so called "compatibility problems" are coming from, there are just no safety margins at all. You can overclock if you want to, but to be sold dimms that are overclocked from the factory without anyone saying anything, is just not good business. The Pentium II processor cache limits memory access bandwidth anyway, so if you don't mind running at only 66MHz with cas 2 this could be an acceptable 90% to 95% performance compromise with decent reliability. I refuse to cheat my good customers by not saying anything about it. I intend to be in business when these garbage memory  brokers have dried up and blown away from all of the lifetime warranty claims that they can't fulfill. You still get what you pay for.

  Order out of chaos. The Intel® PC100 - xxx - xxx specification label.

The test question seems to be "do your cas 2 parts operate at cas 2 at 100 megahertz, or only at 66 megahertz?" The fastest cas 2 parts that operate at cas 2 at either 66 MHz or 100 MHz will have a PC100 label on them that says PC100-222-xxx. Cas 2 parts that operate at cas 2 at 66 MHz and cas 3 at 100 MHz will have a PC100 label on them that says PC100-322-xxx. The 222 or 322 refers to the actual data programed into the spd eeprom chip located on the memory module that informs the motherboard exactly what the sdram module is capable of. Anyone with a logic analyzer can measure this. There should be a piece of software that can measure this parameter directly but I have not seen it yet. All I can offer for now is the shareware benchmark programs seen below. Plug in the brand x module, run the benchmark. Swap to the brand y module, run the benchmark again and see how the numbers compare.

 Find out how fast your memory really is. Download any of three win 95 shareware benchmark software packages now! - winbench is large but is an industry reference standard -nbench has individual modules to run various tests on your memory and processor etc., there is also a version for win nt.  -rbench opens and closes wordpad and solitaire and does a few screens full of graphics. I am convinced that you usually get what you pay for and am willing to prove it.

  | winbench 5.7 mb |  •   | nbench.zip 72 kb |  &   | rbench.zip 3 kb |

Demystifying PC 100 memory specifications
  SDRAM timing specs.

of chips
cas @ 
cas @ 
many -12   83 MHz   66 MHz  3  n/a  n/a
many -10 100 MHz   83 MHz  3  n/a  n/a
most -8 125 MHz 100 MHz  2  3 322-620
Samsung  -G8 125 MHz 100 MHz  2  2  222-620
Gold Star -7J 100 MHz 100? MHz  2  3 322-620
Gold Star -7K 100 MHz 100? MHz  2  2 222-620


We have the fastest parts made either ready to ship or one day lead time 8 nano-second 125MHz in Samsung.
The parts that will operate at cas 2 @ 100 MHz are the fastest parts.
Some eight nano-second cas 2 parts can operate at cas 2 @ 100MHz or @ 66 MHz
while others can only operate at cas 2 @ 66 MHz and cas 3 @ 100MHz.
The 7ns -7K parts will operate at cas = 2 @ 100MHz or @ 66 MHz.
Some 7 ns parts are capable of this, some are not.
The 7ns -7J parts operate at cas = 3 @ 100MHz or at cas = 2 @ 66 MHz.
The -6 parts are not real 6ns, they are actually 10ns, even the chip makers themselves have been adding to the confusion. The data that I have been finding at many of the chip makers web sites is behind. The foundries that have been usually ahead of everyone like Siemens and Micron are behind the curve on this one. I find it confusing myself.

The rest of the details
PC100 is the Intel specification for a new special type of sdram computer memory that can go faster than regular sdram, without experiencing data errors. PC 100 memory is required for the new motherboards with the seattle 440 bx ( se440bx ) chip set from Intel for maximum performance. Many people seem to have been taking advantage of the confusion surrounding the release of a new product line and selling pc 66 memory for pc100 memory. Intel made provision so that the new chip set will recognize either type of memory to facilitate compatibility. Any device running in "compatibility mode" cannot give maximum performance.

There has been a preliminary specification for 10 nano-second pc100 memory since late 1997. Memory that conforms to this specification has basic compliance and can function at 66 MHz , 75 MHz or sometimes even up to 83 MHz under real world conditions. The 100 MHz that is claimed for it is a theoretical limit only. The Intel spec for these parts is referred to as pc66 version 1.0. Memory that conforms to the complete Intel version 1.0 pc100 specification is called full compliance pc100 memory and can function reliably at full speed, that is at the 100 MHz bus speed. The theoretical limit quoted for this memory is 125 MHz.  Full compliance 8 nano-second pc100 modules can perform reliably with a cas latency as low as 2. Basic compliance memory can only function reliably with a slower cas latency setting of 3. CAS =  Column Address Strobe -OR- Column-Address Select  Cas latency is one of the memory timing settings controlled by the bios in the mother board chip set. A nano-second is one billionth of a second, that is 1/1,000,000,000 second.

The faster that your system goes, the more critical memory specifications become. The major consequence of this, is that the quality of your memory modules are now more important than ever before. The most important thing for any type of memory is the quality of the chips, grade A chips are the best that you can buy. They are selected from the heart of the silicon wafer. The heart of the wafer has the best focus of the image mask that is used to create transistors out of raw silicon. Grade A chips do not require extensive testing and matching to find chips that will work well together. The grade A chips from a particular wafer match one another as built and almost always pass quality control testing. Reliable quality is much better built in than tested in.

With the new pc 100 memory specification everything else now is also very important. The connecting traces must conform to required maximum and minimum length, as well as width, spacing and thickness specifications. This is so that each connecting wire printed onto the circuit board has a constant impedance. The use of carefully matched constant impedance traces facilitate exact matching of the time that it takes the data to get from one end of the connection at the chips to the other end at the pins. The printed circuit board construction must now be six layer instead of four layer to further improve timing accuracy. Each trace now also has a series source resistor to help cancel reflected signals. The memory chip data signals are so fast now that all of these things are done just like the transmission lines used to connect radio equipment together.


Because of the real world characteristics of the connecting traces used on memory modules, the data signals travel a good bit slower than the speed of light. It takes a finite amount of time to get from one end of the wire to the other. The real trick is to get all of the individual bits that make up each memory data word to show up at the motherboard connector all at the exact right time for the processor to read and write to them. Many factors like temperature and humidity changes can affect this critical timing relationship. Timing is everything. You can't see what is inside the memory chips or eeprom chip but a visual dead give-away for a non full compliance module would be if the little source resistor packs were missing from the printed circuit board traces between the chips and the contact fingers.

 Some of the first "full compliance" pc 100 modules to hit the market left out two other very important requirements of the specification. Some early chips did not have the correct signal clamp diodes as specified. Some early modules also did not have the correct data programmed into the serial presence detect eeprom chips to match up with the new bx motherboards. The eeprom contains information that is read by, and used to identify the memory modules to your system. If you find that you got stuck with either of these problem non compliant 8 ns modules they will work just fine in the 440 BX motherboards at 66 MHz. If you put them into a 440 LX motherboard you will have extra good performance margins for over clocking. However, as for overclocking, what good is a system that goes 10% "faster" but has data errors twice as often? If it were me, I would be looking for a rebate or replacement modules in that case. I am already hearing bait and switch stories about folks selling either non-compliant or basic compliance modules for full compliance.

 Am I trading in fear tactics? That is not my intention, my regular customers understand that my prices are good for the quality of my goods delivered. For me that means that I can sleep well at night knowing that I have delivered the best value to my valued customers. If you see a price that looks high, please call or email anyway, it might be a few days old.

A fully informed customer is a happy customer with a bright future in store
 www.pc100.com is from the fine feathered folks at  the Memory Store 
and is in no way affiliated with Intel Corporation,
all trademarks are the property of thier respective owners &cetera.


| stupid computer y2k trick |

PC 100


y2k 128mb 256 mb 512mb 1gb 2gb abit giga byte asus super micro dec alpha ddr esdram sldram rdram drdram y2k 128mb 256 mb 512mb 1gb 2gb abit giga byte asus super micro dec alpha ddr esdram sldram rdram drdram