How will the Megahertz Wars be remembered?

Any computer will run slowly, with a sufficiently crappy compiler. One real issue with speed is that to maximize it you must use a compiler that outputs optimised code that takes advantage of the good features of a particular CPU and minimizes the weaknesses. That's one reason why programming demi-god Donald Knuth still thinks that teaching machine language and knowing the computer architecture is still important (see Why have a machine language? section):

I still don't know the truth of the Mhz War, but if both were using relatively similarly well done compilers, the Mhz might actually matter.

Apple was definitely publishing some inflated numbers during the whole MHz wars. I remember that Apple advertised the dual 1.42GHz G4 as benchmarking up to 22 GFLOPS, and the closest I could get to that number with the Altivec Fractal thing was 14 or 17GFLOPS (which is still fast). Intel's chips have become much faster, too. The Core 2 architecture is significantly faster than the Pentium 4 architecture, and SMP is a great thing to have.

If the G4 never came out, then I think Apple would either still be using PowerPC processors or they would be out of business. IBM could push the G3 to 900MHz before Motorola ran into the 500MHz wall. I think the Altivec performance was what saved them, really.

If Intel hadn't dropped NetBurst (P4) then AMD would have several more leads than Opteron. The Mhz Wars are one reason drove Intel to use NetBurst, as they could publish the multi-Ghz numbers, even if the CPu was really slower. Know why the Tualatin P3s were dropped? One reason is that unless all code was recompiled and optimised for P4 it really wasn't faster, and in some cases it was slower. Luckily they didn't fully shelve the design, and now we get the ICA with a more promising future.

hmmm, perhaps future operating systems / compilers / hand assembly will put some speed into aging PowerPC Macs. Yellowdog

Yeah, sort of.

Flash back to the early days of the PowerPC, and you'll find yourself positively overwhelmed with discussions of RISC vs. CISC archtecture, all of which of course is heavily slanted towards the idea that RISC is "cleaner" and fundimentally superior to CISC. Which is horse hockey from a number of angles, but it sounded good. (Despite the fact that arguably the PowerPC doesn't really qualify as a true RISC architecture.) The other point harped on was that x86 designs paid this huge penalty by retaining hardware compatibility with legacy 16 bit code, and that somehow that complicated their architecture so much that it was impossible for them to ever be efficient. (And therefore, impossible for them to ever be "fast", even though the two arn't necessarily the same thing.)

Anyway. There was some truth to these arguments back in 1995. The original Pentium design had to sacrifice a fairly large chunk of its silicon real-estate to accomodate some very fancy decoding and instruction reordering hardware in order to achieve the goal of being a "Superscaler" (multiple instructions-per-clock) processor and still handle variable length non-aligned machine code instructions. The PowerPC was able to offer similar if not better performance per clock with substantially fewer transistors and lower power consumption, and of course would look more "scalable" by comparison. When Intel introduced the Pentium Pro it actually reinforced some of the accusations of the PowerPC crowd when it was found it was actually *slower* then its predecessor in many benchmarks. Clearly x86 was doomed.

In truth, really, the advantage of PowerPC was largely illusionary. In designing the Pentium Pro Intel decided to sacrifice some performance with 16 bit code in order to optimize its 32 bit performance, which was actually quite good. However, at the time most people were still running 16 bit DOS and Windows 3.1 programs, so it looked bad. However, it still ran "legacy code" a lot faster then a Macintosh, which had to use a software 68040 emulator to run the bulk of the programs on the market. (Including most of the OS, sadly enough.) As time went on more and more software was targeted at the 32 bit Intel ISA, which is at a much smaller efficency disadvantage compared to native PowerPC software, and thus a comparatively smaller and smaller percentage of silicon real estate is dedicated to legacy code. If you compare a modern PowerPC CPU to a comparable x86 CPU, such as the AMD Opteron Vs. the PPC 970 (G5), the transistor counts are almost identical, and the chips provide similar performance. Ironically the "transistor bloat" in the G5 is largely due to having to maintain compatibility with and provide good performance for what for the G5 is "legacy" binary code.

You can actually make a good argument that the "RISC philosophy" is fundimentally incompatible with the consumer computer market. For "RISC" to work at its best you need to be able to recompile optimized object code for every new processor design that comes along. If you're in the position where you have to make a new processor run "legacy" object code, even if it's just for last year's model, you'll find yourself having to resort to the same brute-force on-chip performance enhancers that x86 designs have to use, such as instruction reordering/out-of-order execution. The consumer market consists of people who want to run off-the-shelf software *in compiled, object-code format* fast, and further and they want their new computer to run all the old software they already have faster then their old computer did. For that you need a CISC-y design, by definition. At this point no x86 design on the market actually runs x86 instructions in their core. They translate them to "micro-ops" which are executed by the hardware processor, basically akin to an emulator in hardware. You can almost think of the x86 ISA as having evolved into a "psuedo-machine language", akin to P-Code or compiled Java. It's the gold standard for portable software, and to be successful in the consumer market you need to be able to run it fast.

Apple could always find a benchmark their machines ran faster. (Even if it was just a lame set of Altivec-accellerated Photoshop benchmarks.) That's really what it boils down to. In the early days, meaning the PowerPC 601 through 604 era, PowerPC could generally comfortably best Intel clock-for-clock, at least at floating point performance. Integer performance, however, was basically a wash, and that's more important for most consumer software.

You might find this interesting reading. It contains references to SPEC benchmark results over the early PPC era, and also points out some of the contributing factors to Apple losing the "performance crown". (A major factor being the mediochre motherboard chipset and memory bus designs.)

So how is it that the Intel (and AMD) managed to creep up on and then exceed PowerPC in performance? There's this old saw about economies of scale: "Everyone who buys a Chevy now could be driving a Cadillac for the same price, if only they'd agree on which Cadillac to buy.". The x86 ISA is the Chevy of CPU instruction sets, but the massive demand to run it faster has produced inexpensive Cadillac hardware to do it. If PowerPC had ever achieved true mass-market status (if, for instance, IBM and Apple had actually gotten along and released a real, inexpensive, widely available and licensed alternative to Windows, rather then Apple hoarding its IP and gouging for hardware margins) PowerPC would be the cheap Cadillac. As it is, it's the Lincoln Continental of ISAs. Nice, cushy, *expensive*, and a bit slow off the line compared to its less elegant but now *much* cheaper competition.

"Losing". Well, maybe not quite, but they were definately barking up the wrong tree. There's nothing fundimentally wrong with PowerPC, and if history had played out differently and it'd been adopted by the industry as a whole it probably could be faster then x86 is now. But in that alternate history Dell would also be selling PowerPC-based Precision workstations for Steve Jobs to compare the Power Macintosh G6 to. We'd still all be together, but I suppose you could pretend you're at Target instead of Kmart. Either way, well... you don't get to be special anymore.

I suppose that's what hurts the Macintosh "true believers" the most. "Thinking Different" just didn't translate to being "better".

--Peace