Transmeta letters to the editors
Orvalds becomes a legend in his own Linus-time
An unprecedented flood of emails arrived after our Transmeta stories this week. Some of them were hate mail which was interesting. Exactly why do people get so wound up about this, we ask ourselves? Anyway, the replies were so interesting that here's a selection of the ones that didn't use obscene language! The silicon chips run underground... The rumor around here in Silicon valley, is that Transmeta is building an x86 clone that internally uses VLIW techniques (ala Merced). The rumor is that they got prototype silicon some months back. It was running at 375Mhz , however this was considered too slow (remember that the chip has to convert x86 instructions into native vliw instructions so it might need very high frequencies to be competitive). The guy's name is Dave Ditzel (not Dinzel). He's somewhat of a joke among the microprocessor design community. Appeal for psychotherapy What part of "Gestalt" don't you understand? Welchen Teil von "Gestalt" verstehen Sie nicht? Burden of ancient instruction sets Transmeta's new chip shouldn't be burdened with an instruction set that's more than 15 years old. By concentrating on an efficient, modern design they should be able to make a chip that can emulate x86 code in software at a respectable speed and race ahead when running native code. Look at Acorn: when they came to design the next machine after their very successful BBC Micro, they could have looked for a speedier processor compatible with the 6502. Instead they weren't afraid to start again from scratch and build their own 32-bit RISC chip, at a time when everyone else was getting started with 16-bit CISC. They bundled a BBC Micro/6502 emulator (actually two) with every machine to run the extensive catalogue of educational programs available for the Beeb. The result? A decade on, you don't hear much about 6502-compatibles, but you can hardly have failed to notice the ARM and StrongARM. A soft answer averteth wrath I think you're full of shit. I think you're making this up to get hits. This is the last time I will visit your site. Irresponsible journalists can suck my left nut. Lord have Merced on us I doubt that anyone but Intel can get Merced compatability: too many of the architectural features are covered by patents. What they have to be hoping for is Merced-like performance. That may be possible, even with a 386-instruction set (derivative?). I'd be surprised if anyone but Intel could introduce a new instruction set at this stage. Twas brillig and the Alpha wabes Well, even Alpha-based NT systems have to run "DOS software" in the form of peripheral-card BIOSes; video cards in particular. They use software emulation, which is OK, because that's mostly a boot-time operation, and not a performance issue. Still, writing a PC emulator is not trivial. If Linus made it, I'll buy it... I don't think it really needs to run legacy DOS apps.. ONLY if it's going to be mainly a server chip. There are still a lot of DOS games out there that run better than their windows equivalents.. So its not going to be a "legacy" gamer chip. As for Linux..... I'm almost positive it'll run it. And Speaking for the linux community. I know I'll probably buy one just because Linus helped make it. When you're cleaning Windows (George Formby) Simply put. All the windows applications are compiled and built for x86. Porting to a new arch is not going to happen quickly or readily. Unless Transmeta's plans are WAY out in years, there won't be anyone for them to sell to. Leaning on the Linux on the corner of the street (George Formby) Actually, it makes no difference whether the transmeta chip supports dos, win 3 or even windows "nt'. The most important OS it needs to support is Linux - The moot is moot Any new hardware product should be able to easily run 16bit x86 code at 486 DX/50 speeds by using 100% software emulation. The question is moot in todays (sic) environment. Eek, it's the PowerPC Well... IMHO x86 compatibility is kind of like the VGA mode of your graphic cards... slow, somewhat inefficient (compared to other graphic mode), and limited in feature... I think maybe a processor with a translation chip either inside or outside the die, and with the ability to be turned off so it can run native RISC code, it might be a good idea... But take a look at NT, they abandoned MIS and PowerPC, now only x86 and DEC/Compaq Alpha are really supported... if Transmeta is not shooting for x86 compatibility, what are they shooting for? DEC/Compaq Alpha? I don't they they'd have the illusion of draming that Microsoft is gonna port the thing over for them... (AFAIK IBM did the work on PowerPC porting of NT...)
Sponsored: RAID: End of an era?