Special: Inside the development of Cyrix's 6x86MX

On a visit to the UK this week, Michael Woodmansee, project director for Cyrix's 6x86MX (or 'M2') and UK managing director Brendan Sherry gave a rare insight into the development of a state-of-the-art microprocessor.

On the development of the chip...

MW: At the peak of the project I had about 25 people. We started in August or September 1995 and at that time we had just a handful of people.

On operating system and multimedia optimisation...

MW: [M2] is for Windows 95 and NT. We didn't want to optimise for 95 at the expense of NT and vice-versa. Also, we didn't know what the MMX standard was going to be but we knew we wanted to accelerate multimedia and be aggressive about that. Ultimately, we decided to follow the industry standard - MMX. The M2 has a couple of unique features that could be really compelling to multimedia application performance. I think against K6 the M2 will do very well but most of the MMX-optimised software being developed is hand-coded for Intel so PII is going to be faster [for MMX] most of the time.

On AMD...

BS: We're not really going to war with AMD and they don't want to do that with us... We're both focused on taking share from Intel.

MW: They've got a 64Kb cache like us. Their cache is dual-ported but can only support one read/one write per operation and that's a pretty big limitation in Windows. You'll often see many reads in a row... that'll really slow down your pipeline.

On Socket 7 versus Slot One...

BS: There's a lot of things that can happen. We'll be watching very carefully what happens with the new slot. We don't believe that many vendors will want to go exclusively with Slot One. Dell and Gateway probably will, they're so closely tied in to the Intel mother ship. Will Compaq? I'd be surprised. For us, the business drivers are more important than the technical.

On mobile CPUs...

MW: I think this can go straight into mobile products. It doesn't need enormous heat sinks. Right from the off we planned a low-cost dye from a manufacturing view and a low-power chip to enable mobile solutions.

On floating point performance...

MW: We've made incremental improvements to floating point but it's not fully pipelined. It's about 10-20 per cent faster than the classic 6x86 if you run a floating point benchmark. If you run something like polygon calculations, it's definitely a step behind Pentium II. But if it's a game, that's not the only thing they do. [Using a floating point benchmark] is kind of like using Norton SI to say this is what Windows performance is like.

Newsletters

You have been successfully signed up. To sign up for more newsletters or to manage your account, visit the Newsletter Subscription Center.
See All
See All