One of the most common questions heard today is “What platform provides the best upgrade path?”. There seem to be as many opinions on this as there are users, mostly centered around whether Slot 1 or Socket 7 provides the best option. Though some have claimed that Intel will move to Slot 2 in the near future, making Slot 1 another ‘dead end’ in the Intel roadmap. most believe that Slot 1 is the answer. Many have also said that Socket 7 is soon to be a platform of the past and that Slot A will replace it by the end of 1999. Interestingly, some new information seems to indicate that the real answer may be quite surprising.
Though Socket 370 has not been cited as the next big platform, it’s introduction actually is the beginning of an apparent trend by Intel. A recent editorial on the MicroDesign Resources website claims that Intel will be abandoning the Slot 1 and 2 packaging altogether by 2000. This article indicates that by shifting to a Socket 370 package, Intel can save $10 per processor for the Celerons for a total savings of about $300 million in 1999 alone. The article goes on to say that Intel will take advantage of this cost savings for the rest of their product line as well.
One of the reasons given for the shift to a Slotted package in the first place was for cost savings. Intel felt that in order to gain the performance edge they needed the L2 cache had to be put close to the processor so it could be removed from the slow system bus. The options were to put the cache on-die or to package it on the same cartridge as the processor. The latter option was chosen because Intel felt it was cheaper to purchase commodity SRAM chips than to try to fit it into a larger processor package. According to MDR’s article, just putting 128K of cache onto the Mendocino chip adds about $10 to the cost due to the relatively large die area it requires with .25-micron transistor sizes. Once the .18-micron process has been implemented, however, 256K cache will fit into the same area.
The basis for the claim that Intel will move away from slots and back to sockets is that all manufacturers are now shifting to on-die L2 cache, or to larger L1 cache sizes. It is well known that AMD will be putting 256K on-die L2 cache on their upcoming Sharptooth and that the K7 will include 128K L1 cache. IDT has indicated that they will also put 128K L1 cache onto the successor to the Winchip-2. According to the information in the editorial, Intel has said they will be including on-die L2 cache with the Cascades, Foster and McKinley chips up to 2MB in size. What is most interesting is the claim that Coppermine, Cascades and Foster will *all* be offered as a socket processor – perhaps even Socket 370!
If Intel does indeed switch all of their processors back to a Socket 370 package, this could create a bit of a problem for AMD. The main justification for the Slot A is that it is almost identical to the Slot 1, with the exception of a few pinouts. This means that motherboard manufacturers can make both Slot 1 and Slot A boards with almost no modification of their production line, making the decision about whether to make a Slot A board relatively easy since there would be little or no additional cost over their current process. If the manufacturers no longer have a need to make Slot 1 boards, they would be less likely to make a special Slot A production line unless AMD has sufficient market share to justify it.
The MDR editorial indicates that AMD may decide to adopt an on-die L2 cache also and turn the Slot A into a Socket A, however a call to the AMD PR department revealed that AMD does not currently have plans to do this. On the other hand, the spokesman for AMD did indicate that they plan on producing Socket 7 processors through the end of 2000, and perhaps beyond. From this information, it seems that AMD my believe that they can push the Sharptooth processor to speeds well beyond the anticipated 600MHz top end that most people have thrown around.
It would actually make sense that AMD might be able to support faster processor speeds, since both L1 and L2 cache are going to be full speed. With the L2 cache no longer tied to the system bus, there is almost no loss of performance as the processor multiplier goes higher. It has already been proven on this site and others that the 100MHz system clock of the Pentium II provides very little benefit over 66MHz due to the 1/2 speed L2 cache, so it makes sense that the same would apply to a full speed L2 cache as well. It also makes sense that as the transistor size shrinks, more L2 cache would be possible to fit into the same size package, and that the speeds can increase. It is feasible that we could see a Sharptooth processor with 512K L2 cache by the end of 1999 running at over 600MHz.
If the information here is correct, the answer to the original question now becomes much different. Most have written off Socket 7 as a dead-end, while dismissing Socket 370 as a low-end option that doesn’t really merit any consideration for future upgrade possibilities. Up until recently, the odds-on favorite for future upgradability has been the Slot 1 package, but it appears that the entire situation has now been turned upside down! As strange as it seems, Socket 370 may actually be the best upgrade path for future high-end processors, while Slot 1 becomes the dead-end. Looking at the alternatives, a paraphrase of Samuel Clemens immortal words seem to apply here – “The reports of the demise of Socket 7 have been greatly exaggerated”.
Be the first to discuss this article!