Following is a serious quote from Slater (of the MicroProcessor Report), this was meant to be analysis, but really shows how gullible PC buyers are -
"I think MMX will be successful with consumers, not so much because there's applications there that are compelling -- because there just aren't very many of them today -- but [because] consumers don't want to buy something that's going to become obsolete any faster than necessary. Consumers are going to be wary of buying a non-MMX processor, because Intel has made a big point of saying that MMX is a big deal."
MMX offers a small amount of value to some users, and even less so if users already have 2D or 3D or Sounds support chips (like most users do). Because of the redesign of the Pentium chip and the larger L1 Cache, it does give it speed bumps of about 10-15% (in many cases). This is not tremendous or anything - just incremental. It is not NOTICABLY faster - but it is faster.
MMX still isn't a PowerPC
The PowerPC's had a big advantage over Pentiums in certain NSP / DSP like functions (dealing with streams of data for signal processing - like Sound, video, graphic filters, etc). Often the PPC's are twice as fast or more than either the Pentium or PentiumPro's. MMX is not anything revolutionary - just Intels way to respond to this deficiency. The benchmarks are showing that the MMX chips roughly double the performance for certain very specific tasks (as compared to the regular pentiums and PPro) -- but that it is just a little faster or slower than the PPC's (1).
(1) The MMX Pentium are about 8% faster than a PPC (603e or 604e) at the same clock rate -- however, PPC's are available in much faster clock rates (20-80%) -- and the G3 is far faster at the same clock rate (and available at about 20% faster, and expected to go faster still).
Also PowerPC's have been getting cheaper more ubiquitous MP support (multiple processors). So you can buy stock PowerPC processors that are faster at doing MMX tasks today than MMX, much faster at doing everything else, and that margin looks like it will increase in the next 6 months to a year. Plus it is easier to throw multiple processors at the problem with PowerPC's. There are even going to be PPC chips that are really four-processors on a single chip -- something the RISC design allows the PPC to do, that the CISC based Pentiums will have a much harder time with (do to heat and space considerations).
MMX vs Trimedia
So MMX is a proprietary instruction set, that will give users a relatively small performance increase, will be able to leverage a few OS tools, will require some custom hand-tuned assembly code in each app. There are gotcha's like penalties in performance if not real careful - cant use FP and MMX interchangably, and do not want to switch too often. All this so that Intel can demand a premium for these chips of about $200/chip. (MMX requires lots of extra gates on processor and they pass that on to the consumers). Remember that users get more performance by just staying with dedicated video and sound accelleration chips that are usually already built in to their systems today.
Apple is partnering with Philips to use their Trimedia Chip. This chip is rated to be anywhere from 10 to 100 times faster than MMX (4billion instructions/second VLIW chip), and it off-loads the main processor from having to do the work, and users can have more than one chip. So while MMX hangs the computer while it is doing its work, the TriMedia-PPC solution allows all to work in parallel. So not only is TriMedia faster, but it helps make the main processor(s) faster as well by offloading work. This solution is more open, gives dramatic performance increase, supports multiple Trimedia chips (MP), will require less custom code than MMX (and comes with libraries to make it easier for programmers to use), does not have the performance penalties of MMX, and programmers can still can use FP simultaneously. Cost for starters $50/chip for the TriMedia (about 1/4 the cost of MMX) though system costs will probably be more than MMX until the Trimedia chip is integrated on to the motherboard (probably 6 months to a year later). Another strength to the TriMedia solution is that users can upgrade their multimedia capabilities and chip seperately from their main processor - so there is more versatility.
Philips trimedia - http://conquest.oakridge.com/philips_semiconductors/trimedia/
MMX is nothing new -- it is just a new name for an old concept -- one that (ironically) IBM and the Power Architecture (what the PowerPC is based on) helped innovate. The PPC is going to get a full 64 bit architecture and VMX (MMX like functions, but far superior) in 1998. We will see what that does to MMX.
Read What is MMX for more.
MMX requires that Apps take advantage of special code and a proprietary instruction set and only ONE processor to get the advantage (Intels new MMX instructions). MS can help with VfW (video) support, and Direct3D - but many different developers use many different libraries to do the same things in the PC world, so your applications will vary. On the PC side the support for MMX has to be pretty spread out to get results. Lots of work for developers - and little return, esecially considering that MMX can reduce performance due to context switch times.
MMX has overhead with switching from floating point math to MMX instructions, and programmers can not do both at the same time. So in some cases the overhead out-weighs the advantages. The PentiumII uses MMX2 -- which is not nearly so bad as MMX is at this -- but the whole design still needs more work.
Apple on the other hand has some more standard middleware like QuickTime and QuickDraw3D that are used more universally. Apple supporting those alone will allow most programs, that need this type of technology, to get results of the specialized chips IMMEDIATELY. TriMedia has also already created libraries for audio/video/filtering/phone/DSP/etc. to help programmers leverage the functionality of the chip directly. And MetroWerks has already comitted to a CodeWarrior version to utilize it as well. The results will be a few TIMES more dramatic than using MMX. And the trimedia chips are more open. They are made by only one company for now, but anyone can buy from Philips, and more importantly Apple/IBM/Motorola are not using proprietary instructions to lock you into their architecture.
MMX is not a bad way for Intel to respond to its weaknesses as compared to the PowerPC, and it almost catches up. However, the PowerPC camp is rapidly increasing the general speed of their processors - while MMX is mostly a way to increase only a sub-set of its functionality. If Intel is not really careful they will fall dramatically behind not only in multimedia but even further behind in general.
If you want multimedia capabilities dedicated chips for modems, graphics accelleration, 3D and sound are already far superior to MMX, and do not slow your processor down - so in this regard MMX is a step backwards. However, it does allow system designers a little more flexibility in creating mediocre performance on the low-end. Mid and High-end users will still want their specialty chip solutions that will far outperform MMX.
Trimedia is a way to create a much more far reaching extension to computing capabilities in regards to multi-media. It will enable much more multi-media capabilities, is less proprietary, is more versitile, and offers much better performance at possibly LESS cost.
Also check out the article on RISC vs CISC.