Sun’s x86 Clone

Cyrix, Rise Technology, Transmeta, IDT’s WinChip, Meridian, Metaflow, MemoryLogix, Montalvo – all of these were failed x86 clone attempts – and, according to the New York Times, there was nearly another x86 clone from Sun.

Sun, apparently, recognised what everyone has recognised, that Intel’s soft-underbelly is its inability to do low-power.

So Sun decided to produce a low-power server chip – clearly something that is needed to reduce the staggering power requirements of server farms.

In pursuance of its low-power, x86 server clone, Sun, in 2008, bought the assets of  Montalvo an x86 multi-core processor developer with a similar approach as the Cell processor in off-loading tasks to specialised subsidiary cores for efficient processing.

Sun also hired away some AMD talent to work on the project.

But nothing was brought to market and then Oracle took over Sun.

Will Larry Ellison resurrect the project?

Has he the balls to take on Intel in the x86 market?

Well Ellison has balls but, maybe, not that big.


Comments

13 comments

  1. And then there was the mysterious PowerPC 615, which never made it out of the starting gate (Compaq and Eckhard Pfeifer was pretty keen on that part, for awhile).
    One could easily argue that RISC, specifically PowerPC, made Intel hyperefficient by causing them to go to overlapping product development cycles and by investing in fabs that can put out high-speed chips (which, truth be told, DEC’s Alpha taught us all a lesson).
    Intel’s Atom, specifically Moorestown, looks pretty tasty: http://en.wikipedia.org/wiki/Moorestown_(CPU)
    Still no DSP though to help with video encoding (?)
    As for Mr. Ellison, one should know by now that a.) Larry clanks when he walks (i.e. he has brass ones) and b.) he doesn’t do things “just because”, and certainly not because a blog ribs him for it!
    Transmeta’s technology lives on, and Cyrix became the AMD Geode line (which was relatively successful in certain markets like thin clients). Via does a good job these days with low power x86.
    But as for me, I’m now into health care. Why? The processor we are using, Texas Instruments’ OMAP 3530, cost at least $70 million to design, and is fabbed in a Very Expensive Factory that costs – what? billions of dollars/Euros? And I pay less than $25 a part for all that technology! Its good to not be on the wrong side of this transaction!
    OK, to complete this rambling post, here’s one: if you wanted to license a DSP core for your new SoC, where exactly would you go? BOPS, Siroyan, Improv Systems, 3DSP, all gone … Ceva, perhaps, but for high performance?
    Cheers,
    Alan

  2. Those who haven’t tried, Jakob

  3. Przemek Klosowski

    @SteveH: servers only have to be built on x86 if they are running Windows. Linux always ran on everything, so if there’s a low-cost server platform that’s competitive in TCO per performance, people will use it, unless they are stuck on Microsoft.

  4. That was exactly what I believed that Sun would do when they bought Montalvo. I really expected a Niagara-style server-oriented chip to come out of Sun, with an x86 instruction set. Would have been interesting to see that, but seriously, who can think that competing with Intel is a good idea?

  5. Surely no one does, Steve H, PowerPC and SPARC are proof of that and even ARM is being prepped for use in servers

  6. Curious that everyone assume a server has to be built on x86. Surely if power consumption is becoming an overriding concern, then there will be a market for servers built on a better CPU, burning less power, even if there’s an overhead of migrating any software to that platform.

  7. IMO, Transmeta made a decent attempt at emulating the deeply-flawed x86 ISA. But the few products I saw using their chip set had abysmal performance. They should have also directed their microcoding finesse to a few more-efficient architectures, which is what they initially proposed.
    As to Intel, I’m sure they have the engineering talent to implement a low-power design, but neither Intel nor anyone else can overcome the x86’s inherent inefficiency to combine Pentium-class performance with a respectable power envelope at today’s feature size.

  8. Torben, I think the key phrase in your comment is ‘moved towards’ – Yes, Intel has moved towards low power but it hasn’t got there yet. An x86 which had the power-efficiency of ARM-based stuff (currently a 10x difference) would be a viable market contender.

  9. Indeed, Lefty, but he also goes in for cash-burning activities like sailing – described as like taking a cold shower while tearing up £10 notes – so the x86 clone business might appeal to him.

  10. Robert, you’re absolutely right – sorry for that – I should have remembered.

  11. David,
    you’re giving TI a free pass, don’t forget their 586 debacle. It was a completely new design approach, new process, new fab, new…..
    If memory serves me correctly they achieved a whopping 10Mhz main CLK operation. It was so embarrassing that the whole X86 strategy got canned (they were already shipping a viable 486 part)

  12. Lefty Goldblatt

    Does Larry not make quite a lot of dosh selling software? One processor markup falls into insignificance to a software license does it not?

  13. Torben Mogensen

    While Transmeta’s x86 processors were not hugely successful, I would no really call them a failure. They were used in a number of subnotebooks from, among others, Fujitsu.
    Similarly, Cyrix had reasonably success with its x86 clones, in particular the 6×86 series.
    Many of the others you mention never made it into products at all.
    In any case, this is beside the main point of your blog: Will Oracle revive Sun’s attempt at making an x86 clone. I seriously doubt it. For one, Intel has moved towards low power, so the market that Sun sought is already covered by Intel. Also, x86 is a very tough market where it is extremely difficult to make an impact.

Leave a Reply

Your email address will not be published. Required fields are marked *

*