Positive Reports From Transmeta

utopicillusion writes "The register reports : "More cash flowed into Transmeta in the second quarter than it spent, the company said late last week as a teaser for its upcoming results announcement." This is about after a month that CNN predicted that Transmeta was going under. "

Comments

Re:I find this suprising

[m4dm4n]

FTFA: "Transmeta has been aggressively attempting to cut costs ever since it decided earlier this year to get out of the chip manufacturing business and focus instead on processor technologies it can license to other vendors."

You're probably going to battle to find any chips with the name transmeta on next year. However, many other chips may contain licensed technologies in them that will be bringing revenue into transmeta.

That's because they sold their processor business

[Nova Express]

I shouldn't be surprised by Slashdot's short institutional memory, but the rteason Transmeta showed a profit was that "it sold off most of its chipmaking business for $15 million to Culturecom Holdings." It's cash flow positive in the same way a family which auctions off all its belongings is cash-flow positive: temporarily.

Re:IBM's new chips = Transmeta revival?

[TheRaven64]

The PS3 / XBox 360 CPUs schedule instructions relatively poorly. This is not a problem for a console - it is a fixed platform and the CPU in it will be the same in a year's time. You can compile a game for the console with a compiler that's explicitly aware of the instruction timings of the deployment chip and it will work correctly on every console because they have the same CPU.

PCs and workstations are more varied. Every CPU revision can throw your instruction timings off and make your tightly compiled code suddenly much slower.

The only CPU I am aware of that is taking a similar approach to Transmeta's offerings is Itanium, which has a few quite neat tricks to get around the standard limitations of VLIW. Itanium bundles all instructions which can be executed in parallel together to allow additional execution units to be added later without and still used by existing code. Transmeta use x86 dynamically translate x86 instruction set code to VLIW code on the fly, so the compiler doesn't need to know about the native VLIW instruction set (not very useful, because it still means they need to do all of the things that VLIW is supposed to eliminate - although perhaps they can just do them once and then cache the results).

Oh, and no one is stripping out branch prediction. With current CPUs, the cost of a branch prediction miss is horrendous (it can be over 200 instructions on a P4, for example), so removing branch prediction would absolutely cripple performance. This is one of the reasons why languages such as Java have high-level syntactic constructs like exceptions - to allow the compiler to provide a hint for the first time past the branch (the exception path is known to be lower probability).