Transmeta (Semiconductor) (TMTA)

[weaksidehelp]

Chips Off the Old Block
PC Magazine, February 26, 2002, by Warren Ernst

http://www.pcmag.com/print_article/0,3048,a=22124,00.asp

Are you shying away from that less expensive new computer because it lacks the "Intel Inside" sticker? If so, you're not alone. A large segment of the computer-buying public believes the only way to get a truly Intel-compatible processor is to buy one from Intel.

That's not the case, though. Modern non-Intel x86-based CPUs are really just as Intel-compatible as Intel's chips. (Of course, we're not talking here about the chips that power the Mac; those evolved independently, are not x86-based, and are completely incompatible with Intel chips.) You'll see why if you understand how a CPU does its job, what has changed in CPUs and what has remained the same over the years, and what your software expects your CPU to do.

<snip>

The technical aspects of the chips notwithstanding, there are many indications, though not necessarily proof, that compatibility is not an issue in the real world. Perhaps the best example consists of reviews that include results of benchmark tests run on AMD- and Transmeta-based systems. (See the sidebar "And What of Transmeta?" for a discussion of how the Crusoe chip achieves compatibility.) Look closely: You never see "unable to complete the test" in those tables. Run after run, these chips complete the stressful tests just fine. For the purposes of compatibility, ignore how high the scores are; just recognize that if chips were incompatible, they wouldn't get benchmark test scores at all.

And why wouldn't non-Intel chips work well with software anyway? Consider that Microsoft, Oracle, Red Hat, and most other software companies use AMD-, Intel-, and Transmeta-equipped machines to develop their software. If products crashed repeatedly on a certain type of chip, these companies would get rid of the problematic machines in a heartbeat. They don't. In fact, counting the patches listed in Microsoft's own Knowledge Base, you find far more for solving problems occurring with Intel-based CPUs than with competitors' products. And let's not forget the well-publicized Pentium math bug of a few years back. A chip isn't necessarily perfect just because it comes from Intel.

So getting back to your new computer: Should you worry about that "Intel Inside" sticker? In a word, no. Instead, worry about having enough RAM, extra hard drive space, a fast video card, and strong customer support. The compatibility of the processor is one of the few constants in the PC universe.
Warren Ernst is a computer consultant, author, and journalist in southern California. His home site is at www.warrenernst.com.

Sidebar to Article:

And What of Transmeta?

In the design of alternative Intel-compatible CPUs, the apple rarely falls far from the tree. The Transmeta Crusoe, on the other hand, isn't even in the same orchard.

The basic instruction sets (though not the multimedia extensions) of the modern PIII, P4, and Athlon processors are very similar to those of the ancient 386 and 486 CPUs. And although current AMD and Intel microarchitectures are quite different, both decode and process instructions using millions of hard-working, on-chip transistors. These complex internal designs permit a relatively old 32-bit processing scheme to scale up to gigahertz speeds by brute force. Why stick with this old 32-bit scheme? Because being compatible with software written for the 386 and 486 is a necessity for a modern Windows-based PC.

The Transmeta Crusoe works very differently. Its microarchitecture is nothing like that of the 386, the 486, or anything else seen in a PC with an "Intel Inside" sticker. Instead, the Crusoe uses a small, efficient 128-bit VLIW (very long instruction word) processing core with native instructions that are totally incompatible with the x86 instruction set. In many respects, this is similar to the internal RISC-like instruction sets of Pentium and Athlon processors. But the Transmeta core has far fewer transistors, because the x86 instruction translation, decode, and scheduling functions are handled off-chip by a Code Morphing software layer, which translates x86 instructions into Crusoe's internal VLIW instruction format.

During execution, the Code Morphing software actually has a few different modes of operation, as determined by real-time feedback from instruction processing. The software either interprets or translates x86 instructions into the internal format. Less frequently executed x86 instructions might only be interpreted. More frequently executed x86 instructions are translated into VLIW code, and the translated code can be optimized for speed and cached for subsequent reuse. The more often the code is executed, the more Crusoe tries to optimize it, up to a point. The Code Morphing software, which initially resides in ROM located on the processor module, is copied to system DRAM during processor initialization for faster execution.

Additionally, Crusoe monitors the processor load and is able to adjust its frequency using its LongRun technology to conserve power and reduce heat. The Crusoe is known for being a low-power CPU and is now seen in some mini-notebooks that are providing one and a half to two times the battery life of those based on competing processors, though factors such as active power management of other devices and more efficient LCD technology may also influence the results.

If this seems like a lot of extra effort just to work differently than Intel processors, well, it is. But Transmeta rightly believes the benefits of less power consumption and lower heat generation are critical to mobile processing environments. Crusoe allows the design of systems with longer battery life and without CPU cooling fans. And so far, with all of Transmeta's wizardry, Crusoe CPUs seem to be extremely compatible with standard Intel x86 processors.


Great article to reassure consumers re concerns of compatability between INTC and AMD and, even more so, TMTA. This article is also found in the Feb. 26, 2002 issue of PC Magazine. The full article also includes a nice history of x86 chip development, which was snipped above in the interest of brevity.

Regards, wsh