
Original Link: https://www.anandtech.com/show/254
We all remember the platform shoes, and the tacky suits on the bodies of the dancers that dominated the dance floor, bouncing to every bump of bass in the catchy beat of the Beegee's Stayin' Alive. No, we're not talking about your high school dances in the 70's, rather CPU manufacturer, Intel's MMX campaign of 1997. | |
Both experiences should bring back memories, good and bad, of disco, truly "unique" fashion, and one of the most empty promises in the history of PC processors, otherwise known as MMX technology. Just as Stayin' Alive is one of the most well remembered songs of the Beegee's dance music reign during the 1970's, Intel's MMX technology still is one of the most well remembered contributions the company has made to the industry. Two years after the introduction of the 57 MMX instructions into the x86 core, Intel's MMX technology has been licensed by competitors and has probably caused more processor sales than the name of AMD's flagship or Cyrix's one-hit-wonder. Upon walking into a computer retail store, the rainbow emanating from Intel's MMX logo seemed to attract the attention of just about every single customer, without even bringing up the question as to whether or not the 57 new instructions actually provided any tangible benefit to users. |
Times were good, simplicity was bliss, and Intel's MMX technology was forced into the market just as easily as disco caught on to bold teenagers two decades before. It took little more than a few reviews of Intel's Pentium MMX to prove to the market that the "wonderful" MMX technology was nothing more than a marketing ploy, offering a very tiny performance benefit to the end user. Unfortunately, because of Intel's incredible power over the market, MMX was a marketing requirement, without the "MMX Compatible" logo on their processors companies like AMD and Cyrix/National Semiconductor would have had a difficult time even attracting attention to processors sitting in systems next to Intel MMX based computers. After months of legal disputes, and senseless political protection of the "precious" 57 MMX instructions Intel held so dearly to themselves, the technology was finally licensed out to AMD for use in their upcoming K6 processor, thus began the demise of Intel's perfect marketing strategy. Instead of providing a technology that they would be the sole proprietors of, Intel managed to allow competing companies to leech off of their marketing success with the MMX campaign by licensing out the MMX trademark. It wasn't too long before end users could go out and buy an AMD K6 MMX processor, or a Cyrix 6x86MX processor that offered MMX compatibility at a cost generally lower than Intel's own Pentium MMX. MMX was indeed a failure in the eyes of Intel's marketing team, as well as in the eyes of the industry.
As the old maxim goes, "once bitten, twice shy," the industry was not about to allow for another MMX campaign to be force-fed to itself. It is because of this shyness that competing manufacturer Advanced Micro Devices, AMD, had so much trouble rounding up support for their step towards the direction of additional enhancement instructions to include in the core of their K6-2 processor. AMD's 3DNow! instructions, just weeks before their official unveiling at the 1998 E3 Expo in Atlanta, Georgia, were being torn to pieces before the eyes of the public by comparisons to Intel's MMX instructions that made their debut about a year earlier. It wasn't until AMD displayed the true, tangible benefits of their 3DNow! instructions that the market began to give the struggling company a chance. AMD took the weight of the blow MMX inspired at the introduction of their 3DNow! instructions, and with the initial wave of resurrected hatred towards any MMX-like instructions gone, Intel felt it was their time to step forth with a new instruction set as AMD was quickly gaining performance and popularity as a result of their efforts. Intel promised revolutionary new instructions to be integrated into the next version of their highly publicized Pentium II processor that would offer benefits similar to those brought by AMD's 3DNow! instructions, Intel promised a set of enhancements that would be everything MMX wasn't; and they called them, MMX2.
Luckily, for Intel's marketing sake, the name MMX2 didn't last too long as it was soon replaced by the name KNI, or Katmai New Instructions, named after the processor codename they were to be implemented in. The name KNI was nothing more than an internal codename, and Intel finally settled on a much more descriptive name which, although doesn't roll off your tongue as well as MMX or KNI, does convey Intel's newfound preference of quality over marketing ability in the instructions. The name? Streaming SIMD Extensions, or SSE for short. The processor? Intel's third generation Pentium processor, we've come a long way since the days of the Pentium classic processor and Intel is ready to carry on tradition with their latest concoction, the Pentium III processor.
So without any further ado, let's follow in the words of the Beegee's and try to understand the New York Times' effect on man, or more specifically, Intel. Has the idea of pleasing the customer been crowded by the marketing and publicity Intel has received since the company's original claim to fame was made?
The Pentium III - "Revolutionary" or "Revolving" Technology?
Imagine a Pentium II, except outfitted with 64KB of L1 cache, operating at clock speeds derived by the 133MHz Front Side Bus (FSB), and combined with the release of a killer chipset that would boast AGP 4X acceleration as well as the introduction of Rambus DRAM. The processor you're imagining is what the majority of the market envisioned the Pentium III's release as being, and as you can probably conclude by now, the vision we all shared of the Pentium III was a bit on the fantasy side.
The fact of the matter is this, the Pentium III, in spite of its dramatic name change, is nothing more than what the Pentium MMX was to the Pentium classic. The Pentium III offers the same basic features the Pentium II brought to the table, with a major difference, the introduction of the 70 new SSE instructions as previously mentioned. While the Pentium MMX doubled the amount of L1 cache of its predecessor, the Pentium III remains virtually unchanged from the original Pentium II, leaving many critics disappointed with Intel's lack of originality with the Pentium III processor.
For those of you that aren't already familiar with the Pentium II, it features an internal 32KB Level 1 cache split evenly between data and instruction set caches, and support for an external, closely-coupled, on-package Level 2 cache operating at 1/2 clock speed. The Pentium III's L2 cache running at 1/2 clock speed, as with the Pentium II, is located on the processor's Slot-1 cartridge in the form of two 256KB modules placed adjacent to the processor core. Since the speed of the L2 cache is derived from the clock speed of the processor and not the FSB speed of the system, future Pentium III processors will continue to offer almost linear raw performance increases as a large percentage of system performance is courtesy of a fast L2 cache. As described in AnandTech's AMD K6-3 review, the function of all system level cache in a computer is critical to performance:
Cache is nothing more than high speed memory that is located closer to your CPU for faster access to frequently used data. The first place your CPU looks for data is in the cache, and more specifically, the cache located on the CPU itself, referred to as Level 1 or L1 cache. If the data the CPU is looking for isnt present in the L1 cache, or it fails to retrieve it in the current clock cycle, it then looks for it in the secondary cache, if present, otherwise it retrieves it from your system memory. Assuming that there is a secondary cache present (L2 cache), the processor can then retrieve it from a source slower than that of the L1 cache, yet still faster than if it had gone all the way to the system memory to retrieve the data. This process continues with however many levels of cache your system has before the processor has no other option than to retrieve the data from system memory, the slowest option out of them all.
Unlike AMD's K6-2, whose L2 cache was generally stationary at the supported 100MHz FSB frequency, the Pentium III's L2 cache speed scales with the processor's clock speed. At the same time, unlike the Celeron A and AMD K6-3 processors, the Pentium III does not offer the integrated L2 cache running at clock speed that other newer 6th generation processors offer, including those from Intel themselves. The Pentium III, like its predecessor, is still outfitted with 512KB of L2 cache and as the Celeron A and K6-3 processors have already shown us, the performance of a system with 256Kb of L2 cache running at clock speed is approximately equal to that of a system with 512KB of L2 cache running at 1/2 clock speed in most cases (although AnandTech will get into the performance differences later on).
A Cool New Look - SECC2
Intel made a mistake with the introduction of the first Pentium II's Single Edge Contact Cartridge (SECC), the mistake was made in an attempt to eliminate the competition by introducing a new standard into the market (sound familiar? MMX…). By enclosing the entire Pentium II processor in a plastic casing, and by making the printed circuit board (PCB) the Pentium II was placed on so large, the costs of producing a single Pentium II processor almost prevented the technology from being introduced into the sub-$1000 PC market. Intel threw together a quick fix for this problem by introducing the Celeron processor, which removed the plastic casing around the Pentium II. This "naked" design also allowed for direct contact to be made between the processor and the heatsink/fan combo which helped improve cooling issues the higher frequency processors introduced. Intel's final admission of their mistake came with the introduction of the socketed Celeron processor that completely removed the processor card design and reverted back to the days of Socket-7 Pentium processors. Intel insisted that this move back to socketed interfaces was for the low-end processors to drive costs down for the newly established sub-$600 PC markets.
To keep costs down on the Slot-1 Pentium II/III processors that Intel had been claiming were so "superior" in their slot design the company decided to introduce a new standard for the slot-1 cartridge, SECC2. By removing the front half of the SECC packaging, and allowing the processor's heatsink to make contact with the processor itself, Intel is able to cut costs, as well as improve cooling efficiency. SECC2 will be standard on all Pentium III processors, and it has been available, albeit in limited quantities, on Pentium II processors.
The Pentium III is also a complete implementation of Intel's new Organic Land Grid Array packaging process, a deviation from the standard Plastic Land Grid Array packaging which is much larger in comparison. | |
The OLGA package of the Pentium III can also be seen, in limited quantities, on newer Pentium II processors, such as some Pentium II 450 units. |
Cracking Down on the Bad Guys
Alongside the release of the new Pentium III Intel introduced two highly controversial features into their latest processor line, the first of which being Frequency Locking. As we're already familiar with, clock locking, or the limiting of the available clock multiplier pins on a CPU to a single setting (i.e. 4.0x clock multiplier on the Pentium II 400) was a feature Intel introduced into the Pentium II/Celeron line of processors back in 1998 in order to eliminate both the overclocking and remarking of Intel processors. Since Intel has had such an extremely high yield on their processors, some unscrupulous vendors were actually selling overclocked processors and remarking them as their overclocked counterparts and making quite a killing off of the sales. The problem with this, from Intel's perspective, was that when the consumer finds out that they've been had, the first in line to receive a nice little blaming would be Intel, provided that the vendor mysteriously "disappeared." Intel did not want any sort of defacing of their name in this manner, and in order to combat remarkers, they introduced a number of measures to reduce the quantity of remarked chips floating around on the market.
This still didn't stand in the way of Intel as remarkers quickly found a way around the clock locking mechanisms driving Intel to a more rash method of combating the problem, frequency locking. The Pentium III will be the first frequency locked processor from Intel, meaning that the processors will only operate within a certain range of clock frequencies before shutting themselves down. Since no clock generator can produce a solid 450MHz or 500MHz frequency, Intel has built some variation into their frequency locking technology, allowing a Pentium III 500 to operate within a few MHz above or below the 500MHz mark (i.e. 495MHz - 505MHz). Although Intel would not specify the exact tolerance the Pentium III processors would provide for, nor would they discuss the details of how they happen to be achieving the frequency locking on the processors themselves, it is obvious that Intel is trying their best to kill two birds with one stone here by eliminating overclocking and remarking. Let's just hope, for our sake, that we can find a way around these limitations.
The second, and by far the most controversial feature Intel chose to start implementing with the Pentium III is the introduction of the Intel processor serial number. There's no better way to start to describe the Intel processor serial number than a direct quote from Intel's processor serial number FAQ:
Q: What is the Intel. processor serial number?
A: The Intel processor serial number feature consists of silicon devices in Intel's Pentium III processor that are programmed to a specific number during manufacturing. The Intel processor serial number serves as an identifier for the processor, and, by association, its system. It is similar to the serial numbers on other electronic devices or products, except the Intel processor serial number is implemented electronically, rather than being placed on the exterior of the product.
What does that mean? Basically, every Pentium III processor will have a unique ID, sort of like a name tag, associated with it that would allow it to report such features as rated clock frequency, week of manufacture, etc… The controversial part of the processor serial number is the fact that, when enabled, your unique ID could serve as a unique tracking identifier for you and your computer while on the Internet. Theoretically the processor serial number is intended to only offer a method of informing users of the rated clock speed of their processor while also allowing for greater security during on-line transactions since your unique ID can only be assigned to a single processor, and therefore a single computer, yours.
Many users expressed dissatisfaction with the processor serial number as they felt that it would turn into nothing more than a means of violating the privacy of Pentium III owners, however Intel stands firmly on the grounds that the processor serial number will cause more good than harm possible. Once again, referring to their FAQ on the topic, Intel's official statement regarding the invasion of privacy is best expressed by the following questions/answers:
Q: Will Intel track each processor to know who gets which serial number?
A: No. While this is theoretically possible to do, as part of our manufacturing process Intel will not track or correlate which processor serial numbers go to which consumers, nor will Intel entertain any offers to develop such a list.
And…
Q: Does Intel share any of my personal information with other companies?
A: No, Intel will not share any of your personal information without your permission, as stated in the Intel Privacy Policy.
Because of the general distrust of Intel's intentions with the processor serial number, Intel immediately announced that all processors would ship with the means of enabling/disabling the processor serial number via a software utility. In contrast, many publications are questioning the reality of the status of the processor serial number, until there is a sure fire way of determining whether the feature is enabled or disabled (outside of Intel's software "utility") we'll just have to rely on Intel's official statements.
Q: Some press reports say that once you turn the processor serial number feature "OFF," it will automatically default back to "ON." Is this true?
A: No. While the processor serial number is activated in the chip, the default control utility setting will turn the feature to "OFF." The utility then allows the user to choose whether to enable the processor serial number feature, and it will then remember the user's preference and reinstate that selection each time the utility is run. For a user to enable the processor serial number, they will need to change the software setting in the control utility, which is installed in the "Start-up" folder. They will then need to reboot their PC.
A hot feature to look out for on future motherboards is the ability to enable/disable the processor serial number feature in the BIOS, as that seems to be the best method of approaching the problem if there is any concern about maintaining your privacy. (For more information about the Intel processor serial number visit Intel's overview at: http://support.intel.com/support/processors/pentiumiii/psu.htm)
The Big Question - Is SSE really MMX2?
Intel had a lot of proving to do with the Pentium III's 70 Streaming SIMD Extensions (SSE), and among the tests they had to pass, proving that SSE wasn't just another MMX was one of them. At a first glance, it's easy to say that SSE would be another flop just like MMX, however if you look a little deeper into the two technologies you'll begin to notice quite a few differences.
The name Streaming SIMD Extensions is indication in its own of the power of SSE, as we all remember from the introduction of the K6-2, SIMD is an extremely popular and powerful feature of the 3DNow! instructions. SIMD, or Single Instruction Multiple Data (in this case SIMD-FP as it applies to FPU instructions, whereas MMX offered SIMD-Int for Integer instructions) allows a single command (or instruction) to be applied to multiple sets of data simultaneously. The key to understanding the benefits of SIMD-FP instructions is the emphasis on the simultaneous execution of commonly used instructions such as multiplies, divides, and adds. The perfect example would be in the transfer of a simple cube in mathematical space to a 3D world as is illustrated by the diagram provided by Intel below:
Copyright 1999 Intel
Corporation
The 14 mults, 12 adds and 1 reciprocal function of the above transfer from model space to world space (mathematical space to 3D world space) could benefit greatly from the Pentium III's SSE as you're essentially executing the same instruction and applying it to multiple forms of data over and over again. The nature of SIMD-FP instructions allows for these instructions to be applied to multiple data structures processed by the CPU in a more timely manner, offering an actual improvement in performance.
While MMX did essentially the same thing for Integer values (whole numbers, i.e. 1, 2, 3…), most complex software such as 3D games, 3D rendering programs, image editing software, and even speech recognition software make use of Floating Point values (numbers with decimals, i.e. 0.0001), therefore there was a very tiny real-world performance improvement that Intel's MMX instructions brought to the table. At the same time, since floating point calculations already take an incredible amount of time to process, even on today's fastest x86 processors (relatively speaking, from the point of view of the CPU not the user); therefore, the application of SIMD to floating point operations as provided for by the Pentium III's SSE offers a greater tangible performance improvement than the application of SIMD to integer operations which are already quite fast on x86 systems.
A disadvantage Intel has with the introduction of the Pentium III's SSE instructions is that AMD's 3DNow! has already had quite a bit of time to build momentum and support within the industry, whereas SSE is just now making its first introduction. The software support for SSE at the launch of the Pentium III seems to be quite disappointing, however in the future, you can expect such titles as id Software's Quake 3 Arena to take advantage of SSE as well as 3DNow! with initial reports illustrating a bare minimum of a 5 - 10% increase in performance with SSE enabled under Quake 3 Arena using minimal optimization of the extensions.
Intel's goal for SSE is not the same as their goal for MMX was back in 1997, instead of advertising the Pentium III's 70 Streaming SIMD Extensions, which virtually any CPU manufacturer (if history chooses to repeat itself) could eventually license from Intel, the microprocessor giant decided to push the Pentium III name which only Intel will hold the rights to. It is because of this attempt that veering the market in the direction of the Intel Pentium III alone that you'll see manufacturers releasing processors that follow the same nomenclature, isn't it funny that the K6-3 has been renamed to the K6-III? At the same time, you'll notice that the cute MMX logo on the Intel Pentium and Intel Pentium II Inside stickers is mysteriously absent, as it has been for a few months now. It looks like everyone's trying to forget about the days of MMX. Don't you all just love the marketing/politics involved in the CPU industry?
So in answer to the question is SSE really MMX2? Yes and no. Yes in that the software support for SSE at the Pentium III's launch will be limited, but no in that SSE actually offers a more visible and tangible performance improvement over MMX, which did little more than inspire a series of repetitive commercials staring Intel's now retired "Bunny People."
Taking Advantage of it all
What do you need to take advantage of the Pentium III and its SSE instructions? First of all, the current Pentium III processors will run perfectly fine in all Slot-1 motherboards provided that they have been upgraded to the latest revision of the motherboard's BIOS with support for the Pentium III. Ideally the voltage regulator on the motherboard should have support for the 1.8v core voltage the Pentium III requires, although the sample AnandTech received did not exhibit any erratic behavior at the 2.0v setting. The long term effects of running a Pentium III at 2.0v have not been investigated, and could potentially be hazardous to the life of your processor, so your best bet is to keep the processor running at the specified 1.8v setting. There have been reports of Pentium III processors specified for operation at 2.0v core, however as far as Intel is concerned, the Pentium III is specified for operation at 1.8v only (1.6v for mobile units).
Other than that, the Pentium III runs using the standard 100MHz FSB with a 5.0x clock multiplier, and since the multiplier is locked at 5.0x your motherboard does not even have to support the 5.0x clock multiplier setting (clock multipliers are functions of the CPU not of the motherboard), the bare minimum requirements being support for the 100MHz FSB and the 1.8v core voltage.
On the software end of things, there are three major ways to take advantage of SSE: 1) if the software is specifically written to take advantage of SSE, 2) if the video card drivers are specifically written to take advantage of SSE, or 3) if the application uses DirectX 6.1 which has native support for SSE, although the performance differences are seemingly negligible if DirectX 6.1 is the only thing taking advantage of SSE.
Video card manufacturers such as ATI, Matrox and nVidia have already announced Pentium III SSE compliant drivers, you can expect more to follow in the future as the standard grows.
The Test
The Socket-7/Super7 Test System Configuration was as follows:
- AMD K6 233, AMD K6-2 300, AMD K6-3 450 (engineering sample)
- FIC PA-2013 w/ 1MB L2 Cache
- 64MB PC100 SDRAM
- Western Digital Caviar AC35100 - UltraATA
- Matrox Millennium G200 AGP Video Card (8MB) - All other Benchmarking
- Canopus Pure3D-2 Voodoo2 (12MB) - Glide Tests
- Canopus Spectra 2500 AGP TNT Video Card (16MB) - OpenGL/Direct3D tests
- VIA AGP GART Drivers v2.9
- VIA Bus Master IDE Drivers
- VIA PCI IRQ Remapping Drivers
The Pentium II comparison system differed only in terms of the processor and motherboard in which case the following components were used:
- Intel Celeron 300, Intel Celeron 300A, Intel Pentium II 400, Intel Pentium II 450, Intel Pentium III 500
- ABIT BX6 Revision 2.0 Pentium II BX Motherboard
The Pentium Pro comparison system differed only in terms of the processor and motherboard in which case the following components were used:
- Intel Pentium Pro 200 (256KB L2), Intel Pentium II OverDrive 333 (512KB L2)
- Octek Rhino P6 Pro Socket-8 FX Motherboard
The following drivers were common to both test systems:
- MGA G200 Drivers v1677_426
- nVidia TNT 0.48 drivers (Detonator Drivers were used in the SSE Drivers Comparison)
- DirectX 6.1
- Quake 2 v3.20 w/ 3DNow! Support enabled when applicable
The benchmark suite consisted of the following applications:
- Ziff Davis Winstone 98 under Windows 98
- Ziff Davis Winstone 99 under Windows 98 & Windows NT4 SP4
- Quake 2 v3.20 using demo1.dm2 and Brett "3 Fingers" Jacobs Crusher.dm2 demo
- Naturally Speaking Professional Speech Recognition Software
- Microsoft Netshow Encoder
- Adobe Photoshop 5.02
- Dispatch by Rage Software w/ SSE support
All Winstone tests were run at 1024 x 768 x 16 bit color, all gaming performance tests were run at 800 x 600 x 16 bit color. 3DNow! support was enabled when applicable.
For the in-depth gaming performance tests Brett "3 Fingers" Jacobs Crusher.dm2 demo was used to simulate the worst case scenario in terms of Quake 2 performance, the point at which your frame rate will rarely drop any further. In contrast, the demo1.dm2 demo was used to simulate the ideal situation in terms of Quake 2 performance, the average high point for your frame rate in normal play. The range covered by the two benchmarks can be interpreted as the range in which you can expect average frame rates during gameplay.
Windows 98 Performance
The AMD K6-3 AnandTech reviewed back in December still packs the same punch it did then, and gives the more expensive Pentium III 500 a run for its money under Windows 9x. There's no big performance advantage here, the K6-3 is on the heels of the Pentium III and is closely followed by the older Pentium II processors and even the priceless Pentium II Xeon. Under Windows 9x, as a business application system, it doesn't really matter what CPU you get, as long as it performs within the middle of the above charts and is reasonably priced, you'll be perfectly fine.
AnandTech's engineering sample of the Pentium III was not frequency locked (although all purchased chips will be) and therefore allowed AnandTech to push the limits of the CPU. It's definitely unfortunate that Intel chose to frequency lock the Pentium III, as the Pentium III 500 made it up to 560MHz (112MHz x 5.0) without a single crash, and with a little tweaking 585MHz (117MHz x 5.0) was possible as well, albeit not as stable. Shortly Intel will be releasing the Pentium III at clock speeds greater than 500MHz, making the 560MHz benchmarks here a hopeful indication of performance to come...
Gaming Performance
At the start of the gaming performance benchmarks we have the OpenGL performance of the nVidia TNT Chipset running Quake 2's demo1.dm2 timedemo file. The entire Intel processor family seems to dominate the boards here with the closest runner up being the K6-3 400 with a reasonable 48.7 frames per second, although the Pentium III 500's 64.6 fps is definitely an improvement over that. It's sad to note that the 3DNow! implementation in nVidia's TNT drivers is still lacking the quality necessary to receive the full benefit of the 3DNow! optimizations. Chances are, Pentium III owners will see SSE taken advantage of in a more efficient manner than K6-2/3 owners have seen 3DNow! utilized; that's unfortunately the way things work in this money driven world. In the CPU industry, the majority always wins...with the minority receiving a little attention. Could this be an electronic representation of life itself? That's just a bit too philosophical for this review ;)
The same story here, 3 Fingers' crusher.dm2 benchmark is an excellent CPU/Bus stressor test (Brett, you're da man) and as you can tell by the benchmarks, Intel's superior FPU does come in handy here with the TNT. Once again, AMD is penalized by a poor implementation of 3DNow! in the TNT drivers.
The Voodoo2 is reaching the end of its reign, virtually all of the processors here, including the K6-3 all but maxed out the single Voodoo2's potential at around 60 - 62 fps under Quake 2...make way for those Voodoo3's everybody.
The CPU dependent crusher.dm2 benchmark does show a great variation in processor performance, with the Pentium III and Pentium II processors floating up to the top, leaving the slower Pentium II's and the AMD processors to wade down to the bottom of the pile. Interestingly enough, the K6-3 400 performs just as well as a Celeron 400A in crusher, meaning that the lowest performance expected out of a K6-3 400 when equipped with a Voodoo2 running Quake 2 is nearly identical to that of a Celeron 400A. What about the greatest possible performance? That's another story.
The raw FPU performance of the Pentium III is not improved at all in comparison to the Pentium II 450, the increase in clock speed is the only difference here.
Here you can begin to see the potential SSE has for the Pentium III. Just as 3DNow! did for the K6-2, SSE brings the Pentium III to new levels of performance. Rage's Dispatch demo, a 3965 frame fly through demo of actual gameplay of the upcoming game, Dispatch, truly taxed even AnandTech's fastest Pentium II systems. The Pentium III 560 only scored just under 26 fps, however after turning on the SSE support that was built in to the game, the performance of the Pentium III almost doubled offering a 96% performance improvement. Although the performance improvement experienced in Dispatch can be considered the ideal case, you can expect performance improvements to hover around the 20 - 40% range just by enabling SSE in future games. The potential is there, the weight is placed on the shoulders of game developers and video card manufacturers to take advantage of Intel's new born baby.
Desktop Application Performance
SSE for gamers only? Not at all, using Naturally Speaking, a Speech Recognition software package that supports SSE you can take a look at the speech recognition time in seconds of an 8MB wave file. According to Dragon Systems, the makers of Naturally Speaking, 3DNow! support as well as SSE and MMX support is integrated into Naturally Speaking, explaining AMD's performance in the test. Intel's advantage here is in their control of the market, you can expect SSE optimized applications and games to dramatically grow in numbers as time goes on. This is definitely not a repeat of the MMX fiasco.
Encoding an online video show? There are large numbers of users out there that do it on an almost daily basis, and using Microsoft's Netshow Encoder you can see the small but somewhat noticeable improvement in performance due to the SSE optimizations already built in to the software.
The Photoshop benchmark provided to AnandTech was a bit geared towards SSE optimizations in the method of implementation of the image manipulations and filters, however it does illustrate room for SSE's intervention in the lives of graphics artists. Once again, we're provided with a setting where SSE already receives a benefit in an area where the more mature 3DNow! instructions have barely received any attention. Is it fair? Nope, however that's the cold, hard truth.
Windows NT Performance
Here is an interesting loss for the Pentium III, the AMD K6-3 450 actually beats the Pentium III 500 under business applications under Windows NT 4.0. The Pentium III comes back strong under the more FPU intensive High End applications due to its inherently stronger FPU, however these benchmarks also go to show you that if all you're going to be using your computer for is basic business applications, the Pentium III and K6-3 are both out of your league.
Conclusion
The thing to remember here is this, the Pentium II 450 did not get any slower just because the Pentium III was released, it still offers the same performance it once did...just remember that it's not the fastest Slot-1 processor any more. Is the Pentium III worth buying? It depends on the type of user you are and how much you're looking to spend on an upgrade. If you currently have an average Slot-1 based system, your best bet is to stick with your investment, and wait until the later part of Q2-1999 when Intel will hopefully unveil faster Pentium III processors, driving the prices of the P3 to a more manageable level and making an investment a bit easier on your wallet at that point in time. Super7 users will still find the most economical upgrade in the newly released K6-3 processor from AMD.
For those users looking to build new systems in the very new future, your options are clear, buy a Pentium III now and use up your computer funds, or buy a cheaper Pentium II and save a little extra cash which you'll most likely carelessly spend on something else (unless you happen to be very conservative with money, in which case, congratulations ;)...). Remember that if you're a gamer and go down the path of the Pentium II instead of the Pentium III, even though the Pentium III 450 might seem to be just as fast as a Pentium II 450, when it comes time to play Quake 3, you can expect your friend with the P3 450 to get a nice little advantage over you in performance.
From a marketing point of view, SSE looks very promising, it will quickly become an integrated part of everyday computer junkie life, unlike MMX which faced a long overdue death shortly after its introduction. If you're still undecided as to whether or not the Pentium III is right for you, the best thing you can do is wait it out and see if any of your favorite applications or games out there happen to get SSE support. Intel's Pentium III will go head to head with AMD's K6-3, and on the Business Application end, the competition will be fierce. Intel seems to have the lead in the gaming/FPU world for now, however initial reports of AMD's K7 look more than promising...unfortunately, it seems as if regardless of the quality of the processor, unless it bears the Intel name, it simply won't receive the attention in the industry it deserves.
AnandTech mentioned it at the end of 1998, this year will be one of fierce competition. While Intel comes out strong once again, dominating the headlines with their marketing campaigns and slogans, AMD is back on the street, stayin' alive with their K6-3.
A special thanks goes out to nowhereman, kTk, Red Dawn, Rio Rebel, Pale, and dan88z...the disco fans of the AnandTech BBS ;)