Prince Opteron unloads on AMD, Intel and the future of memory
Fred Weber removes stealth shield
Radio Reg Fred Weber had the look of a mad man.
There he was at some chip conference in 2002, waving sticks of memory around and talking about a new age of server computing. Weber, then CTO at AMD, planned to unleash 64-bits on the x86 realm. Intel claimed this was a horrible idea. Weber's, er, unwavering - sorry - enthusiasm told a different story.
AMD released Opteron a short-while after Weber's memory episode and proved the chip designer right. Server customers needed those 64-bits and access to loads of memory bad. You've met VMware, right?
In Episode 13 of Semi-Coherent Computing, Weber walks us through his early days as a physics student at Harvard, his time with DEC legend Gordon Bell and some slow going days at NextGen. Then we get to the good stuff with Weber's work on Athlon and Opteron chips for AMD, and many, many battles with Intel. Finally, Weber dishes a few tidbits of dirt on his new project - a start-up called MetaRAM that's set to de-stealth any day.
Okay, okay. So this interview runs a bit longer than usual at just over an hour.
Still, I think this puppy is worth the effort. And by "think," I mean "know."
Weber explains his central design philosophy which is, "If it can be done, you must do it." He also hits on where AMD excelled and why it has now fallen behind Intel.
Away from the x86 market, Weber confesses that he sees little future for discrete graphics, while adding that betting against Nvidia is like betting against Ethernet. Go figure that one out.
Yeah. We cover that too. It's all about memory, 3-D stacking and maybe even optical components - if a miracle occurs. Have a listen.
Thanks for your ears. ®
Wow, remembering the Encore Multimax...
Wow, this brings me back. My school brought in an Encore Multimax to reaplce a DEC-20 as the general computer for the school. 20 processors, 64mb of RAM. A real screamer! Cost us $250k at the time.
I helped move it in and it worked well with a bunch of DECstation 3100s as the front ends, replacing VT100/220s.
Three years later, we ditched it for $25k... talk about depreciation!
It was a pretty decent machine, watching compiles fly by with 'make -j 8' was cool.