by Shui-Che Lim
Slot-1 Only?
Intel says that AGP and 100MHz host bus will only be available on Slot-1. This is absolutely true if you stick with an Intel chipset solution. Intel has not and will not support AGP or 100MHz host bus on Socket 7. However, if you were one of the early Pentium II adopters buying a board with an Intel 440FX chipset, well, then you’re out of luck since AGP was first introduced (along with SDRAM and Ultra DMA support) on their recent 440LX chipset. Furthermore, 100MHz host bus support won’t be enabled until the 440BX chipset arrives sometime in second quarter of 1998.
What Intel won’t tell you is that these very same features can be found on Socket 7 boards coming out starting the first quarter of 1998 supporting alternate CPUs such as AMD’s K6 and Cyrix’s Mx686. Now that Intel has officially dropped the ball on Socket 7 support, AMD and the remaining third party chipset vendors such as ALI, VIA and SiS are all providing AGP and 100MHz host bus support for future Socket 7 boards with an initiative called Super 7.
AMD’s Super 7 Initiative was announced during last November’s Microprocessor Forum along with the K6-3D and K6+3D CPUs. Currently ALI (Acer Labs), SiS (Silicon Integrated Systems) and VIA Technologies are in the final phases of getting AGP and 100MHz host bus chipsets ready for market.
| Chipset | AGP Support | 66/75/83MHz Bus | 100 MHz Bus | SiS 5591 | YES | YES | NO | VIA VP3 | YES | YES | NO | ALI Aladdin V | YES | YES | YES | VIA MVP3 | YES | YES | YES |
Motherboards based on the SiS 5591 and VIA VP3 chipsets are just now starting to show up in the market. Full 100MHz host bus motherboards based on the ALI Aladdin V and VIA MVP3 chipset should be available in February/March timeframe. Well known motherboard makers such as Asus, Abit, Atrend, Biostar, FIC, Gigabyte, Microstar and Shuttle as well as a whole host of others are fully behind the Super 7 Initiative.
These chipsets are no slouches. They support all the latest features such as SDRAM, Ultra DMA33, serialized IRQs, and even larger 1MB L2 caches. Basically, they all do everything Intel’s 430TX chipset does, only better. Furthermore, some of these chipsets such as the ALI Aladdin V and VIA MVP3 are mobile capable – meaning that AGP and 100MHz host bus notebooks fully supporting the Super 7 architecture and alternate CPUs from AMD, Cyrix and IDT will be available in 1998.
Super 7 or Slot-1?
What are the relative merits of both architectures, and are there any performance issues that you should be aware of? This simple question can lead to a lot of very technical discussion. I’ll try to simplify wherever possible. Also, to truly do justice to these issues, I’ll also take the liberty of playing Devil’s Advocate to some of Intel’s marketing propaganda concerning Socket 7 vs. Slot-1.
Pentium II: Superior Architecture?
Intel asserts that the Pentium II CPU architecture is the most advanced consumer CPU available. Let’s do a little comparison to see how this argument holds up.
| Processor Features | Pentium II | AMD K6 | Pentium w/MMX {P55C) | RISC Core | Yes/5 issue | Yes/6 issue | NO | Speculative Execution | YES | YES | NO | Out of Order Execution | YES | YES | NO | Data Forwarding | YES | YES | NO | Register Renaming | YES | YES | NO | X86 Instruction Decoders | 1 Sophisticated, 2 Simple | 2 Sophisticated, 1 long, 1 vector | 1 Sophisticated, 1 Simple | Execution Pipelines | 5 | 6 | 2 | Branch Prediction | YES | YES | YES | Advanced 2 Level Branch | YES | YES | NO | Branch History Table | 512 Entries | 8,192 Entries | 256 Entries | Branch Target Cache | 0 Entries | 16 Entries | 0 Entries | Branch Predict Accuracy | 90% | 95% | 80% | MMX | YES | YES | YES | L1 Cache | 16K instruction + 16 K data | 32K instruction + 32 K data | 16K instruction + 16 K data |
On the whole, the AMD-K6 holds up very well architecturally when compared to the Pentium II. This should come as no surprise since the AMD-K6 is a true 6th generation design. The Pentium w/MMX (P55C) chip is totally outclassed by both the Pentium II and AMD-K6 as it is based on Intel’s 5th generation design.
This chart also brings into question Intel’s marketing propaganda of postioning their Pentium P55C against the AMD-K6. While it is true that both the AMD-K6 and the Intel Pentium both reside on a Socket 7 motherboard, the AMD-K6 compares favorably to Pentium II in terms of overall performance despite the use of, as Intel puts it, an "outdated Socket 7 architecture."
Performance figures between Pentium II and AMD-K6 based systems have been verified and reported by far too many PC publications, both in print and online so I don’t need to rehash them here. What is more interesting to note, however, is that most publications will compare (IMHO correctly) the AMD-K6 to the Pentium II while Intel is the odd man out comparing AMD’s K6 to their Pentium. Again, it doesn’t take a big stretch of imagination to figure out why.
Pentium II Bus: More advanced than Socket 7?
Intel further asserts that the Pentium II bus based on Slot-1 is architecturally superior to Socket 7. Let’s look at another chart.
| Processor Bus | Pentium Pro66 MHz | Socket 7 66MHz | Bus Width | 64 bit | 64 bit | Max Bandwidth | 528 MB/Sec. | 528 MB/Sec. | Latency | 5-7 Clocks | 2 Clocks | Bus Protocol | Split Transaction | Address Pipelining |
This chart pretty much says it all. From a performance perspective, the Socket 7 bus is as good as the Pentium Pro (P6) bus being used by the Pentium II. A noted industry pundit seems to agree. "For a uniprocessor system, the Pentium bus is just as good as - if not better than - the P6 bus. " (Michael Slater, Microprocessor Report, December 30, 1996)
The Pentium Pro bus which uses a split transaction protocol is only at an advantage in a multiple CPU configuration. This pretty much means that it holds very little value for the gaming community. When was the last time you saw a gamer running more than one Pentium II using NT as his OS? Yeah, get real.
Multiple CPUs only work when there is specific OS support, such as from multithreaded Windows NT or from UNIX. However, as of this writing, Windows NT 4.0 does not fully support Microsoft’s Direct X API (a prerequisite for a vast majority of today’s gaming) or DOS for that matter; and UNIX doesn’t even do Windows.
DIB: Backside Bus better than Frontside Bus?
Intel creatively coined a new name for a feature found on the Pentium II, which they have termed DIB or Dual Independent Bus. However, what Intel fails to tell you is that this really isn’t a new feature. It has actually been around since the Pentium Pro. Ever heard the saying, "That’s not a bug, it’s a feature!"? Let me explain…
DIB is a backside bus scheme. What this means is that the L2 cache is running on an independent bus from the rest of the system. It is on the "backside" with the CPU between the L2 cache and the rest of the system bus. Conversely, Socket 7 architecture uses a frontside bus where the L2 cache sits between the system bus and the CPU and runs at the speed of the system bus.
In a Pentium Pro system, the host bus runs at 66MHz. However, the L2 cache in the Pentium Pro CPU was running at the full core clock speed of the CPU. This "feature" was ultimately what caused the price for Pentium Pro’s to remain relatively high and inflexible. The cache chip was joined to the CPU in a MCM (multi-chip module) package. The yield from this process was very low since either a bad CPU chip or a bad cache chip or a bad join would ruin the whole module. Of course, you wouldn’t know whether the module was good or bad until it was completed so that it could be tested.
This was how the idea for the Pentium II module was born. It is basically a low cost version of the Pentium Pro chip. The CPU and cache are mounted onto a PCB instead of in an MCM. Better yields using this process results in Intel’s ability to keep pushing the prices down (although one could argue how quickly Pentium II prices would have come down were it not for the competition that AMD and Cyrix have been providing).
One serious disadvantage is that the cache is no longer tightly coupled to the CPU core so it can’t run at full core clock speed. Currently, the L2 cache in a Pentium II runs at half the core clock speed of the CPU. It can be seen that for similarly clocked Pentium Pro and Pentium II systems, the Pentium Pro will always beat the Pentium II in performance specifically because its cache is running at a faster speed. The Pentium II only beats out Pentium Pro systems by virtue of higher clock speeds. Feature or bug? You decide.
Another serious disadvantage is that changing the speed of the frontside (host) bus in a DIB scheme will yield far less performance improvement in a Pentium II system since the cache is separated from the speed of the system bus. Performance figures for overclocked Pentium II systems can be found on Tom’s Hardware Guide. L2 cache performance in a Pentium II improves only as the speed of the CPU scales upward. What does this all mean? Be prepared to pay big money for improved cache performance (read faster CPUs) in a Pentium II system.
However, a move from 66MHz to 100MHz on the frontside bus scheme used in Socket 7 will yield a 50% improvement in performance for the L2 cache. Keep in mind that CPUs such as AMD’s K6 are performing admirably to Intel’s Pentium II clock for clock despite having a much slower L2 cache. There is every indication that on a 100MHz host bus, AMD’s next generation K6-3D and K6+3D will beat comparably clocked Pentium II CPUs by a comfortable margin.
What to buy, Super 7 or Slot-1?
The next generation of Super 7 motherboards will offer features competitive with Intel’s Pentium II motherboard and chipset roadmap. The key difference is that the infrastructure for designing and manufacturing Socket 7 based motherboards have been around for many years. This mature infrastructure allows Super 7 motherboards to be developed and manufactured for less than a comparably featured Pentium II motherboard. Typically, Socket 7 based boards are selling for one-third to one-half less than the price of comparable Pentium II motherboards.
Even if the Pentium II motherboard was to someday manage to sell for the same price of a Super 7 motherboard (and everything else being equal) you would still pay much more for a Pentium II CPU over comparably clocked CPU alternatives from AMD and Cyrix. Also keep in mind that next generation CPUs such as AMD’s K6-3D and K6+3D and Cyrix’s Cayenne will have much improved 3D and FPU performance. AMD’s K6-3D is reputed to have 2 to 2.5 times the FPU performance of the Pentium II when using AMD’s 3D technology. (Ed. Note: Support for the K63d is built into DirectX6 so will be transparent to most applications).
Furthermore, it is not immediately apparent that you need the power (or cost) of something like a Pentium II-266 or 300 when 3D boards based on 3Dfx’s Voodoo and next generation Voodoo 2 can turn a run of the mill Pentium 133 or 166 into a killer 3D gaming rig.
What you buy, ultimately depends on the size of your pocketbook. If you’ve already taken the Pentium II plunge then your best bet would be to hold off on upgrading your Pentium II motherboard until the 440BX chipset comes available. Also, hold off on buying anything AGP until at least Windows 98 becomes available. I suspect that AGP video cards will go through all the standard growing pains like any new technology. It will probably be 2nd or 3rd generation AGP cards that will finally do justice to the technology.
If, however, you’re still using a Socket 7 based board then you can get all the benefits of newer Pentium II systems including AGP and 100MHz host bus by just upgrading to a new Super 7 motherboard and getting a good 3D accelerator. The added benefit is that this will allow you to maintain your CPU investment until you’re ready to upgrade that as well.
Email the author...Shui-Che Lim
The views presented here are not necessarily those of Combat Simulations and we make no claims for the accuracy of this information
To read letters in response to this discussion go to Letters
For further discussion of PII vs K6 architecture go to The Microprocessor
Main Back
© 1997 - 2000 COMBATSIM.COM, Inc. All Rights Reserved.