Computers are all about data; to be more specific, they are about doing things with data. Lets take this a little further and say that a computer performs three main tasks with data.
Processing data
Storing data
Moving data
Your CPU and video card process data, your hard drive and CD collection hold data and then in between are millions of electronics devices scurrying around moving the data. As home computer systems have become more and more developed weve seen faster processors, faster memory and--yes, you guessed it--faster storage devices. As each of these areas becomes more advanced and speedier, it needs the rest of the system to keep up. This is where problems arise.
Over the last ten years weve seen major developments and breakthroughs for the home PC, but with each new advance, the bottleneck of the PC is pushed to another device or sub-system. The limiting factor today is not storing data, or even processing it, but moving it.
If we take a look at the original GeForce 256, it was released in an SDR (Single Data Rate) memory version. Then a little while later the DDR (Double or Dual Data Rate) versions came out. The only difference between these two cards was the memory, or to be more accurate memory speed. The bottleneck was memory bandwidth, not fill rate. The processor or--GPU as nVidia would call it--was not the limiting factor. We can see the same problem with todays GeForce 2 GTS cards: memory bandwidth is holding the card back.
This article is not about the graphics market, but rather the CPU industry. Weve all watched as Intel was tripped up by the birth of the Athlon. Since that time the tug owar match for the customer PC has been raging. This editorial is about what to expect over the rest of the year, and maybe even a little into 2001.
AMD
The superior FPU and larger L1 cache of the Athlon has helped keep it above Intels PIII range for some time. However with the introduction of the Coppermine, featuring an on-die L2 cache, Intel was able to gain ground on the Athlon at near GHz speed. Even though the Athlon had a larger L2 cache, it was off-die, running at only a fraction of the processor speed. As I mentioned earlier, this is where the current bottleneck lies: transferring of data.
AMD certainly knows of this weak point and has prepared a beefed-up Athlon to fight round two. The prizefighters this time are the Thunderbird and the Duron (formerly code-named Spitfire)
The Duron will be going head-to-head with Intels Celeron as a low-cost processor. Its an interesting trend from the two CPU kings, as all theyve done to supply the value market is create a crippled version of their high-end chips. The amount of L2 cache used on the Duron is slightly uncertain at the moment, but it be enough to give it more than a fair chance against the Celeron. However, the Duron wont be as crippled as the Celeron; for a start it will be using the same 100 MHz DDR EV6 bus (effectively 200 MHz). This removes a few of the memory bandwidth problems painfully apparent on the relativity weak 66 MHz bus of the Celeron. Packaged as a 426-pin Socket A, the Duron will hit the streets around the 533600 MHz range, with 667 MHz speeds expected by June or July.
The Thunderbird is the real successor to AMDs current high-end processor, and it promises to win back the edge over the PIII. As with the Duron, the Thunderbird features an on-die L2 cache--a beefy 256 KB at core speed, instead of the existing 512 KB off-die cache running at only a fraction of the processor speed. Two versions will be manufactured, a socket A and a slot A version, but the latter will only be issued to OEMs. This casts doubt on an upgrade path for existing Athlon owners. Even if you could lay your hands on a slot A Thunderbird, it seems that only the decrepit AMD 750 Irongate chipset will be able to accommodate them. The new KX133 based motherboards will not work with the Thunderbird, apparently due to the traces on the boards being too long, leading to fatal signal degradation.
VIA are responding to this apparent flaw with the KX133 chipset by bringing out the KZ133 for the launch of the Thunderbird. However, the KZ133 is nothing more than a KX133 with support for socket A.
This leads me neatly onto the next issue: which motherboards will be able to use these new chips? At the launch it seems that only AMDs 750 Irongate and VIAs KZ133 will work. This rules out the use of DDR memory with the first wave of chipsets. I find this to be a little surprising, considering that AMD were smugly modeling a Socket A Thunderbird DDR system at WinHEC 2000.
If we look a few months down the line to the Christmas end of 2000, AMD will be releasing the Mustang. This will also mark the release of the AMD 760 chipset, featuring DDR SDRAM. Around the same time AMD plans to roll out the 760MP, which will allow for 2-way SMP Athlon systems.
From this brief overview of the rest of 2000, we can clearly see AMD has some big plans, certain to keep them on the Intel hate list. AMD now has the time to create a product for each market segment with a processor based upon the original Athlon.
The crowd, the world, and sometimes even the grave, step aside for the man who knows where he's going, but pushes the aimless drifter aside. -- Ancient Roman Saying
Intel created a winner with the P6 range, which has certainly been stretched a long way since the debut of the original Pentium Pro. The creation of the BX chipset is another great piece of work from the Intel camp; some people still use it to run the new Coppermines, even though this pushes the BX spec beyond its limits. But what of the future? The BX chipset is outdated and the P6 design has run out of life, so Intel has made some revolutionary plans for the months to come.
It seems that Intel will be churning out sub-GHz Coppermines for the rest of the year, and even sometime into 2001. These processors will be paired with the i815, i815E, i820 and i820E chipsets, due for release in June. The i815E and i820E will feature integrated LAN, dual-master USB and ATA 100. To sideline to the issue of hard drive transfer speeds for a moment, Id like to point out its a sad fact that current ATA66 drives dont offer much over ATA33 designs. The limiting factor here is mostly mechanical speeds rather than transfer rates.
The value market has recently enjoyed an injection of new technology in the form of the Celeron II. If you want to know what its made of, just think Coppermine then cripple half the L2 cache and you have a Celeron II. The use of a Coppermine core gives the low end market use of previously high-end-only features like SIMD and the improved L2 cache design. But even all the new bells and whistles cannot hide the fact that the Celeron II is still hampered by a measly 66 MHz bus. With the soon-to-be-released Duron on the scene, Intel may be forced to raise its bus speed to add a little more muscle in the low-cost market.
The fact is the Coppermine and Celeron II will have a hard time against AMDs Thunderbird and Duron chips. The price/performance scale is even more unbalanced by the adoption of RDRAM on Intels flagship motherboards. Weve all heard the SDRAM VS RDRAM debate many times, but there are several points we can agree on. For one, RDRAM is currently more expensive than SDRAM; this alone is a key disadvantage when comparing the two memory technologies. The use of DDR SDRAM running at 133 MHz gives an effective clock speed of 266 MHz. This translates to 2.128 GB/sec of memory bandwidth, a higher raw number than PC800 spec RDRAM, which can only offer 1.6 GB/sec. I know that when AMD chips start shipping with DDR SDRAM well see some impressive results, but really the future lies elsewhere. I believe this future may well be with a Rambus licensed technology.
Intels future lies in the use of RDRAM with its new P7 architecture processor, the Willamette. Early this year, Intel VP Albert Yu demonstrated a 1.5G Hz Willamette. This successor for the P6 line will begin to move into the market late this year at the 1.3 GHz mark. It was stated at the demonstration that the Willamettes 100 MHz quad-pumped bus has been designed for use with dual channel Rambus RDRAM. This quad-pumped memory bus will certainly outperform the Coppermines 133 MHz bus. With Intel grinning about quad-pumped RDRAM, its interesting to note that the high-end versions of the Willamette will use DDR SDRAM. Quite a strange twist of events, which is hard to explain. Some have suggested that higher end configurations would need large amounts of memory, thus making RDRAM an overly expensive solution. The trouble with this theory is that the lower end Willamettes will be using RDRAM, which is a more expensive coupling than a system with SDRAM.
The Timna will also make use of SDRAM, which is not really surprising for a value chip. Although originally this SOC (System On a Chip) solution was meant to feature an RDRAM controller on-chip, it seems that Intel reconsidered. Timna integrates video and the Northbridge chip onto the CPU die, making for a cheaper system. Intel will really be pushing the Timna as a price, rather than a performance, solution. Expect to see 600 MHz Timnas hitting the streets the second half of this year, with 700 MHz versions in early 2001.
Conclusion
This article has provided an overview of the battle plans from both sides of the chipmakers war, but its still too early to say how the battles will play out. One of the key issues so far this year has been L2 cache and memory technology, both of which have been leaned on to open up the memory bandwidth bottleneck. The problems in this area can normally be resolved at the cost of a price premium. As time moves on, technological advances will bring price down, so the price/performance trade-off will be less of an issue. Its been a painful time for RDRAM so far; the introduction of such a high-priced RAM has hurt Intel. With numerous problems experienced with RDRAM chipsets, Rambus has had to try and keep people looking to the future, rather than focussing on its rather messy present.
DDR SDRAM will certainly give current SDRAM a major bandwidth boost, but beyond that the future of SDRAM is uncertain. Although the technology for TDR (Triple Data Rate) RAM does exist, Ive heard no firm plans for its use in PCs. My recommendations for the coming months is to take advantage of the high-speed battle, and grab a cheap Coppermine or Athlon for a nice price.
The future of the CPU is uncertain, but with the highly competitive environment exemplified by this battle, we are sure to see a new age of high-powered home computers.
The crowd, the world, and sometimes even the grave, step aside for the man who knows where he's going, but pushes the aimless drifter aside. -- Ancient Roman Saying