The IBM 6x86MX PR200

Note: For up to the minute 6x86MX information please visit my 6x86MX FAQ.

I think that you will find my review of the IBM 6x86MX to be one of the most comprehensive 6x86MX reviews available. I spent many hours working on it. I hope you enjoy it.

On April 2, 1997 AMD officially announced their K6 microprocessor and the focus of the industry shifted. Everyone suddenly forgot about their disappointing K5 microprocessor and AMD became the darling that everybody wanted to take to the dance. The 6x86, without MMX technology and Pentium MMX performance, suddenly became a has been and yesterdays news.

Although the 6x86 Classic offered performance that topped an equivalently rated Pentium Classic, it got off to a slow start and has seen only limited success with top PC makers. However, it has done extremely well in the market served by predominantly smaller vendors. Cyrix/IBM hope to build momentum from the successes of the 6x86 and capture additional marketshare with the 6x86MX.

The IBM 6x86MX (formerly code-named M2) is the successor to the IBM 6x86. The IBM 6x86MX was officially announced on May 30, 1997. With the introduction of the 6x86MX, Cyrix and its partner IBM once again force themselves in the spotlight. The cpu battleground gets more congested with its entry. The total number of viable microprocessor choices in the channel is greater than ever. The 6x86MX will go head to head against the Pentium MMX, K6, and the slower Pentium II chips. Cyrix/IBM have to not only worry about competition from Intel, but also from AMD. AMD wants to keep its newly obtained crown of alternative cpu maker of choice. Cyrix/IBM want to take it back. It's going to get ugly.

Caveat: This website focuses strictly on 6x86 processors and related issues. I include hardcore 6x86MX benchmark numbers in this review, but I will also spend some time analyzing the issues surrounding the 6x86MX and give my opinion on things other than the Winstone scores.

The Announce

Cyrix/IBM remained very tight-lipped about the 6x86MX and did very little to generate channel interest prior to its official announcement. You would think that Cyrix/IBM would want to start generating interest so that people would look for the announcement and pay close attention to it. Instead the 6x86MX was released with very little fanfare and it took a lot of people by surprise. The fact that the 6x86MX was released so quietly is absolutely amazing considering that the 6x86MX is the most important product launch of 1997 for this Richardson, Texas based microprocessor company.

After asking Cyrix marketing personnel questions about the 6x86MX, I was referred to their public relations agency. All 6x86MX questions that I had were to be fielded by them. The end result was that I received no information from Cyrix about the 6x86MX. Requests for information went unanswered by their PR agency and a commitment by them for a 6x86MX remains unfulfilled. Fortunately, I was able to obtain early IBM 6x86MX parts, information, and marketing materials in advance of the IBM 6x86MX launch.

I sincerely hope that the 6x86MX partners have an advertising budget and work hard to promote the 6x86MX to retailers, resellers, and consumers. I see AMD K6 and Intel Pentium print advertisements everywhere these days.

The Name

I personally dislike the name 6x86MX very much. The only good thing about this name is that I don't have to change the name of my website. It sounds as though the primary competitor to the 6x86MX is the Pentium MMX rather than the Pentium II. The Pentium MMX is quickly becoming an entry level processor. Why would Cyrix/IBM want to associate the name of their fastest processor to Intel's entry level processor? How about the name 6x86 II? Cyrix/IBM already have a competitor to the Pentium II 233 with the 6x86MX PR233.

Although the 6x86 is a recognized brand in the computer industry, it still conjures up a lot of negative images in peoples minds. By using a completely different name you avoid the association to the problems of the past. Problems such as excess heat, poor FPU, and Windows NT 4.0 problems even though these issues have long been addressed and corrected. A lot of people, salespeople, retailers, and consumers will not touch a 6x86 chip. To give it a different name would imply a totally different chip even though it still uses a 6x86 core. It's called marketing. I think that they should of stuck with M2 or a derivative of it as the name. Everybody in the computer industry already knows what the M2 is because it has been around for so long.

6x86MX Architecture

Although the 6x86MX is based on 6x86 core technology, the 6x86MX wrings extra performance out of slower clock speeds from architectural tweaks designed to let it handle data more efficiently and an increase in the size of the cache memory. The L1 cache has been increased from current 6x86's 16KB to 64KB, twice the Pentium II's 32K. Cache memory is integrated into the processor and can feed data to the processor at high speeds. Internal memory-management tricks reduce the need for the Cyrix chip to continually access standard RAM, which is slower than internal memory. In the 6x86MX-PR233 Cyrix uses a 75-MHz system bus which is faster than the 66-MHz bus used by Intel and AMD.

Like the 6x86, the 6x86MX chip will offer most of the Pentium Pro's most attractive features - including register renaming, out-of-order completion, and speculative execution. Unlike the 6x86, the 6x86MX will boast optimization for 32-bit code and for MMX.

MMX is an extension to the x86 instruction set that allows multiple bytes of data to be packed together in a single register and operated on simultaneously. MMX consists of 57 new instructions for speeding multimedia tasks. The trick to these new instructions is that they operate on many different data elements at once, which increases the overall amount of work the processor can do. Examples range from operating on multiple pixels in a bitmapped image to multiple sound channels in an audio stream. MMX is not intrinsically a graphics or multimedia acceleration technology, but those are the types of applications that stand to benefit most. MMX will provide a substantial performance boost for some new applications that are coded to use MMX (more so for home than for business applications).

Cyrix/IBM needed MMX and higher performance in order to stay in the cpu race with AMD and Intel. Cyrix settled an MMX trademark dispute with Intel in early April paving the way for Cyrix to call its chips MMX-compatible.

The major differences between the 6x86MX and the 6x86 processors are listed below (Taken from the Cyrix website).

Feature Comparison

Architectural Comparison

Other notable features of the 6x86MX:

• 2:1, 2.5:1, 3:1, 3.5:1 core to bus clock ratio
• .35 micron 5 layer metal CMOS fabrication process; expected to transition to a .25 micron process in early 1998
• 6.5 M transistors compared with AMD K6's 8.8 million, Intel Pentium MMX's 5.5 million, and Intel Pentium II's 7.5 million.
• Uses existing Socket-7 P55C compatible boards and chipsets

The PR (Performance Rating)

The 6x86MX will initially be offered in 150MHz, 166MHz, and 188MHz flavors.

Because the 6x86MX is architecturally more efficient that the Intel Pentium MMX and Pentium II processors, it has equivalent performance at a much lower internal clock speed. Lower megahertz speeds are an advantage, reducing power consumption and heat, making the chips more energy-efficient and more reliable. The 6x86MX has the lowest power consumption / performance ratio. This is a big change from the early days of 6x86 Classic.

Unfortunately we continue to have to live with the PR rating. Everyone should recognize the PR rating because it was carried over from the 6x86 Classic. Remember that PR stands for Performance Rating and not Pentium Rating. The folks at Intel are responsible for this change in meaning. They threatened Cyrix/IBM with legal action for naming their 6x86 processors Pxxx+. Intel thinks that they own the letter P and Cyrix/IBM responded by changing the name to PRxxx+.

Notice that all of the internal clock speeds of the 6x86MX processors are 2.5 x bus speed.

The 6x86MX has the most confusing PR rating I have ever seen. It is a two tiered rating that compares the 6x86MX to both the Pentium MMX and Pentium II. Also, the PR rating used by 6x86MX is completely different than the PR rating used by the 6x86. A 6x86 Classic with a 150MHz internal clock speed is compared to a Pentium 200. A 6x86MX with a 166MHz internal clock speed is compared to a Pentium MMX 200. This is because the Pentium MMX is faster than the Pentiums Classic.

How does the Pentium Pro fit into this rating? It doesn't. Cyrix/IBM don't consider the Pentium Pro to be a competitor to the 6x86MX. They are targeting the 6x86MX towards Windows 95 users. Windows 95 users don't normally buy Pentium Pros.

Honestly, although I understand the need for a Performance Rating, I still hate it. An 6x86MX running at the same internal clock speed, but at a different bus speed invalidates the Performance Rating. For example, an M2 doing 75 / 150 instead of 60 / 150 will spec out higher than a PR166. The same thing goes for an 6x86MX performing better than a PR200 at 83 / 166 instead of 66 / 166.

So here is what you can tell all of your friends. I just bought a 6x86MX PR166 which is equivalent to a Pentium 166 MMX, but I am using a 75MHz bus speed instead of a 60MHz bus speed so it performs like more like Pentium 200 MMX . That will clear things up for them.

How I Tested

 Components:

• IBM 6x86MX PR200 Revision 1.31
• Video board - Matrox Mystique 2MB SGRAM
• Hard drive - Seagate Medalist 2140 / 2.11 GB Fast ATA-2
• Motherboard - Abit AX5 512K
• Memory - 32MB 10ns Macrotron SDRAM

Test condition:

• OS - Windows 95 OSR2
• Video driver - Matrox version 3.41, 1024 x 768, 16 bit
• The test system is loaded with WIN 95 OSR2 and benchmark programs only.
• I used a program supplied by IBM called M2OPT to set the bios for safe 6x86MX performance. 6x86MX bios support for my Abit AX5 has not been completed.

In order to isolate the chips as solely responsible for any performance differences, I installed an IBM 6x86L PR200+, IBM 6x86MX PR200+, and AMD K6 200 in the reference system described above. As a result, the only variable in the benchmarks was the cpu.

I choose to use the Abit AX5 which is based on the new Intel TX chipset even though it does not support linear-burst mode. Linear burst mode is supported by Cyrix/IBM processors and can boost 6x86MX performance by approximately 3 to 5 percent. However, there are no Intel chipsets which support this feature.

Cyrix/IBM 6x86MX processors will still run faster on Intel HX and TX chipsets even without linear burst mode support. You are fooling yourself if you think otherwise. I can make this statement because I also own and use a board that supports linear burst mode. Intel chipsets perform significantly better than chipsets from VIA, SIS, and VLSI.

I now own a Intel HX, Intel TX, and SIS 5711 chipset based motherboard. These boards are the Asus P55T2P4, Abit AX5, and M-Tech Mustang R534. The Asus P55T2P4 was donated to me by Universal PC in Lake Worth, Florida. Thank you Miguel. I will back up this statement with a benchmark comparison of these chipsets and show which one is fastest when used in conjunction with Cyrix/IBM processors. I should of done this a long time ago and I expect to roll it out over the next few weeks.

Benchmark Tests

I ran the following tests on a IBM 6x86L PR200+, IBM 6x86MX PR200+, AMD K6 200.

• Business Winstone 97
• High-end Winstone
• CPUMark 16
• CPUMark 32
• Intel Media Benchmark

The 6x86 vs the 6x86MX

The 6x86MX PR200 offers roughly a 15% improvement over the 6x86 PR200+ Classic for non-MMX application execution at the same Performance Rating and a 13% improvement at the same internal clock speed. For those who are Windows 95 users and run primarily Windows business applications, this is a nominal increase. The High-end CAD / 3D benchmark numbers give us a good representation of the strength of the FPU. It hasn't changed. The 6x86MX PR200 offers a 8% improvement over the 6x86 PR200+ Classic at the same Performance Rating primarily because of the higher clock multiplier of the 6x86MX PR200. There is only a 3% improvement at the same internal clock speed.

Windows 95 Performance of the 6x86MX.

I have stated many times that I would not buy a motherboard that didn't have a 75MHz or 83MHz bus speed. To coin a phrase, the proof is in the pudding. You can obtain 6x86MX PR200 performance by running a 6x86MX PR166 at 75MHz x 2 instead of 60Hz x 2.5, but only pay the 6x86MX PR166 price. The performance difference between the 6x86MX PR200 and the 6x86MX PR166 at 75MHz x 2 is only 2%. You can also get 6x86MX PR233 performance by running a 6x86MX PR200 at 83MHz x 2 instead of 66MHz x 2.5. Note, just because you meet the core frequency regardless of the clock multiplier does NOT mean that the part will work if you are running the bus frequency out of specification. You essentially overclocking the chip. The majority of times it is not a problem, but there is no guarantee that it will work.

I was not able to obtain 6x86MX PR233 benchmarks because I could not overclock my 6x86MX PR200. However, they should look very similar to the 6x86MX PR200 numbers at 83MHz x 2. My luck overclocking Cyrix/IBM processors is not very good. I am only 1 out of 6.

Also notice that a lowly 6x86MX PR166 (60 x 2.5) will outperform a 6x86 PR200+ Classic (75 x 2). The 6x86 Classic will soon die because Cyrix/IBM are going to stop allocating precious silicon wafers for their production. Inventories may still be available for some time.

The 6x86MX vs the Intel Pentium MMX

Three chips already outperform the Pentium MMX overall: the Pentium II, AMD K6, and Cyrix 6x86MX. Yes, the Pentium MMX offers better FPU and MMX performance than the 6x86MX and K6 but these operations do not make up for other serious performance shortcomings that result in significantly lower Winstone scores. Pentium MMX analyses and comparisons have been published countless number of times. You can find benchmark comparisons of the AMD K6 and 6x86MX vs. the Pentium MMX within most major computer magazines on the web or on Tom's Hardware Guide. I am not going to waste my time talking about something most of you already know. I recommend not buying a Pentium MMX.

The 6x86MX vs the AMD K6

The results here are mixed. The 6x86MX PR200 is faster in some areas, but not all. The 6x86MX takes the blue ribbon for Windows 95 performance and business applications while the AMD K6 takes 32-bit performance and high-end applications. The AMD K6 200 runs 32-bit code 12% faster than the 6x86MX PR200 according the the CPUMark 32. The CPUMark 32 is a synthetic benchmark and does not always represent results obtained by real world applications. While in general the conclusion is correct, this figure can be misleading.

High-End applications represent particularly demanding classes of work that are heavily dependent upon FPU operations such as application development or image editing. The AMD K6 FPU is stronger than the 6x86MX FPU. However, for these kinds of applications you are best off with a Pentium II because the AMD K6 FPU is also weak in comparison.

MMX Performance

For lack of a better benchmark, I had to use the Intel Media Benchmark to gauge MMX performance. This benchmark is Intel's and it may show the AMD in a more favorable light because AMD has cross-licensed the MMX instruction set from Intel and will include it with all versions of the K6. Cyrix, always one to work from scratch rather than cloning Intel designs, decided against a licensing agreement and created their own version of MMX. Unfortunately, it looks like that they didn't do a very good job. From the looks of these benchmark results, MMX performance of the AMD K6 kills the 6x86MX. It is 35% faster overall.

I am in the process of trying to obtain a MMX optimized application so I may add additional benchmarking information to this section of my review. I hope that ZD Labs releases a MMX benchmark sometime soon. One is desperately needed.

Availability and Pricing

General availability of the 6x86MX will be approximately 4-6 weeks from May 30, 1997. Expect the first shipments to go directly to OEMs. I think that you can expect a very similar situation to what we have seen with AMD as far as availability goes. AMD announced the K6 233 in April, but just go out and try to buy a boxed K6 233 today. It is difficult. There will be many 6x86MX PR166s, fewer 6x86MX PR200s, and negligible amounts of 6x86MX PR233s in the channel for some time. In my opinion, Cyrix/IBM sacrificed yields and quantities in order to get the product out of the door. The official announcement came just days after they had production silicon.

Cyrix expects to ship tens of thousands of the 6x86MX chips in June and to increase this to the hundreds of thousands in the third quarter. By the end of the year, the company expects to have shipped about 1 million of the processors. IBM is not expected to have volume until August.

Cyrix states that PR166, PR200 and PR233 versions of the Cyrix 6x86MX are available now, priced at $190, $240 and $320, respectively, in quantities of 1,000. These three versions of the 6x86MX cost considerably less than their AMD-K6 ($244, $349, and $469) or Pentium MMX ($270, $492, and $594) equivalents (All pricing is per chip in thousand-unit quantities.) The company also expects to introduce a version rated at 6x86MX PR266 in the fourth quarter and a 6x86MX PR 300 in the first quarter of next year. Although the 6x86MX is similar in performance to Intel's faster MMX Pentium and newly introduced Pentium II processors the price is decidedly different. Even Advanced Micro Devices' K6 processor, considered a low-price leader to date, costs significantly more than Cyrix's new high-end 6x86MX. The new Cyrix chip family offers Pentium II and Pentium MMX-class performance at less than half the current per-chip pricing of Intel's offerings, and well below Advanced Micro Device's K6 pricing.

Cyrix keeps its low-cost strategy after the price and positioning of the previous 6X86 chip demonstrated the difficulty of competing with Intel in the mainstream and high end. This is an extremely bright move. Cyrix's focus on the low-priced PC segment is successful because that is not an area that Intel seems intent on dominating. Even AMD has difficulty here because they have to pay for their new Fab 25 that is weighing heavily on their balance sheet.

IBM Microelectronics manufactures Cyrix processors and sells them under their own label. Although IBM targets the same low-cost PC market as Cyrix, its processors are priced about 25 percent higher than Cyrix parts. IBM states that they have a policy of pricing consistently with competitors and will adjust prices accordingly when needed. The IBM 6x86MX microprocessor is available now in limited quantities. General availability of the PR166 and PR200 is August. General availability of the PR233 is September.

Expect my sponsor Real World Technologies to be one of the first vendors to have the 6x86MX available for sale in quantities.

Conclusion

The 6x86MX is based on existing technology that we see in the 6x86. The 6x86MX is essentially a 6x86 on steroids that includes MMX technology. It is not a complete redesign. As a result, the same strengths and weaknesses in the 6x86 are evident in the 6x86MX. If you didn't like the 6x86 Classic, I doubt that you will like the 6x86MX.

Personally, I am very high on the 6x86MX and there is no doubt that the 6x86MX will be a successful cpu. Cyrix/IBM have certainly crashed the AMD party and gave Intel another black eye. Whether or not Cyrix/IBM manage to sign up any large systems vendors to resell this chip remains to be seen, but I really don't think that it matters. Cyrix/IBM is initially targeting its existing customer base and value conscious consumers.

Production volumes and yields are what concern me. Cyrix/IBM need to ramp up the yields of the 6x86MX as fast as possible. Most people won't want to buy a 6x86MX PR166 or 6x86MX PR200. They want a 6x86MX PR233 and above. Inadequate volumes and yields will hold Cyrix/IBM back. Consumers don't want to wait 4-6 weeks for a Cyrix/IBM 6x86MX PR233 when they can buy an alternative now.

Price/performance is biggest selling point of the 6x86MX. Cyrix/IBM continue to own this and AMD and Intel can't beat it as of today. The 6x86MX is priced right and delivers blue ribbon performance under Windows 95. Windows 95 is what the 6x86MX was built for.

I am sure that all of you will bash the 6x86MX for its weak FPU. Some of you became delusional before the launch and thought that Cyrix had redesigned the FPU for the 6x86MX. This is absolutely not true. I now agree that it is time for the FPU to be improved.

Because of the slower internal clock speeds, intensive CPU operations are slower than what the Performance Rating represents. This is a failing and misrepresentation of the Performance Rating system.

If you are considering buying a Pentium MMX then you are wasting your money. The AMD K6 and 6x86MX both outperform the Pentium MMX. Even the newly released Pentium MMX 233 is a non-contender. This is a CPU that nobody should waste their money on.

Although the 6x86MX PR233 offers comparable performance to the Pentium II 233, the Pentium II is still king of the performance hill because neither Cyrix/IBM nor AMD currently offer Pentium II 266 or Pentium II 300 performance. If you have got to have the fastest that is available and money is no object (you will have to buy a new motherboard also) then this is the right choice for you.

The AMD K6 is a strong cpu and it will remain wildly successful, but I think pricing pressure from Cyrix will force them to lower their prices. Why pay more for the AMD K6 when the Cyrix 6x86MX offers similar performance for less money? The following article http://techweb.cmp.com/crw/news/int0611.html reports that Intel is going to drop the bomb on Pentium MMX prices in July. It has always been conventional wisdom that Intel would sacrifice market share rather than margins. We always thought that Intel wouldn't get into a price war. This may not be true after all. This could mean serious trouble for both Cyrix and AMD who try to offer prices approximately 25%-50% lower than Intel's.

Who is going to win the cpu war? You and I will of course. Three competitors beating up on each is always better than one lone ranger. The end result will be lower prices and faster processors sooner rather than later.