Please explain nanocode. The Crusoe chip is a VLIW machine that runs code Transmeta calls code-morphing sw. If you're calling that microcode, where's the nanocode? They were referring to the morphing code that would remain proprietary, right?
It might be instructive to compare the Intel/HP strategy for VLIW (as I understand it) to TransMeta's:
Intel/HP are working on a VLIW chip that will perform X86 instructions, primarily in the same manner as TransMeta - software/hardware on the chip will translate the instruction into one or more "native" VLIW instructions. (There may also be a small amount of special hardware on the chip to handle particularly complicated X86 instructions directly.) But the details of the VLIW instructions will be made public and the intent is that 3rd party vendors will be able to write compilers, etc., that generate the VLIW instructions. In this model, ultimately, the X86 instruction set and architecture could disappear entirely. (A lot of technical folk can't wait for that to happen, I might add...)
TransMeta, on the other hand, seems to be saying they don't intend ever to make the details of the VLIW instructions public. They have a reasonable technical argument as to why they think it would be a bad idea to do so, but by not doing so they are tying their business fortunes more tightly to the continued existence of the X86, whose technical limitations are becoming more apparent every day (at least at the high end).
In other words, they are setting themselves up as a follower rather than a leader, at least from a business standpoint. (There's no doubt they've done Neat Stuff from a technical standpoint.)
Re: nanocode - some CPU architectures (and I admit I can't name one specifically) have more than one level of translation between the machine instructions the compiler generates and the hardware. The first (and often only) level is "microcode"; the next level is "nanocode". (Presumably you'd then get femtocode, etc.) One of the reasons one does this is to make the hardware simpler, at the expense of speed. A TransMeta processor probably doesn't have any nanocode (i.e., software on the chip that takes the VLIW instruction words created by CodeMorphing and breaks them into smaller pieces); a VLIW machine shouldn't need it in general. On the other hand, handling certain exceptional conditions might be simpler to do using nanocode rather than dedicated hardware...
Why would a new chip need to be pin- compatible, if OEMs were willing to do new designs? As for your MOV example, well was it pointed out today that the vliw engine also has the Northbridge integrated. Anyone familiar with PC design knows that the other 'piece' of the architecture is the Southbridge, which talks to the cpu and the Northbridge via the PCI bus and a few extra control signals.
If OEM's are willing to do new designs, the chip wouldn't need to be pin compatible. But new designs cost money.
I didn't know about the integrated Northbridge. In particular, it means the chip is tied very tightly to the PC architecture - you won't see IBM mainframes stuffed with TransMetas any time soon. I imagine it also limits the ability to use TransMetas in large-volume embedded applications - you won't see them in your Ford either...
From a business standpoint TransMeta seems to be aiming for a niche market. (Targeting notebooks and PDAs makes sense if you're not going to be pin compatible since those machines typically have custom designed motherboards anyway.) I guess I'm just a little surprised that a company with such bold and aggressive technology could have such limited business goals...
As to your last points, I never heard it claimed the chip could be plugged into a Sparcstation. Only that different versions of the morph software would allow the vliw engine to run other operating systems.I think you're confusing running various OSes with being able to plug the chip into different hardware platforms....
An operating system is tied very closely to the hardware. Imagine a Sparcstation OS running on PC hardware: "Hey, where's the hard disk? Hey, where's the keyboard? Hey, where's the video buffer?"
On the other hand, if you had a Sparcstation emulator running on WinNT, it would be nice to be able to run Sparcstation binaries at near-native speed, which sounds like it would be feasible...
Bottom line - I think TransMeta's patent portfolio is stronger than their business plan. And their marketing department may be more talented than their technical team. |
