3 ms·
It's even worse than that. Intel acquired Alpha, tried to jointly license it with Samsung, and that failed too. I have no idea what the status of that IP is. Th
by nickpsecurity 10y ago
It's even worse than that. Intel acquired Alpha, tried to jointly license it with Samsung, and that failed too. I have no idea what the status of that IP is. They should probably try to open it up like OpenPOWER to see if anything comes of it. Meanwhile, its legacy remains in Shinwei processors that likely came from stolen I.P. and power supercomputers. ;)
Far as their failures, I think the reason is that they always do something simpler and better but then add some weird stuff that screws it up. Gotta maintain backward compatibility where possible plus make compilers easy. A great example of doing it right would be i960 w/out the BiiN crap, with C compatibility, and x86 emulation a la Transmeta or Loongson. Instead, they did no x86 or C compatibility plus Ada support. Countering market demand killed it even if market was being short-sighted.
Better approach is Cambridge's CHERI which provides both benefits and tooling compatible enough for FreeBSD:
https://www.cl.cam.ac.uk/research/security/ctsrd/cheri/ https://www.cl.cam.ac.uk/research/security/ctsrd/cheri/
Another good example of doing it right is Hardbound which experimented on x86 ISA:
https://www.cis.upenn.edu/acg/papers/asplos08_hardbound.pdf https://www.cis.upenn.edu/acg/papers/asplos08_hardbound.pdf
These two both drastically improve an important area of CPU's, mainly security or reliability, while maintaining as much compatibility as possible with existing investments. The CPU's and compilers do most of the work actually. Many of the good designs also modify only a small part of CPU investment and/or compiler to reuse those too. So, my recommendation for Intel is to stop trying to get clever (Itanium, SGX) in favor of copying stuff that works, makes sense, is backward compatible, and has major reuse of x86.
- rrmm 10y agoIntel has always used the x86 to add incremental improvements which are aimed at improving bottlenecks (Pentium 4 not withstanding). The bad side is that x86 is a multilayered blob of random features cobbled together; (I'm glad AMD got to x86-64 first). But it's produced a processor that offers useful things to people along with backwards compatibility. And that seems to be what people want. I wonder if deprecating and eventually removing the ugly parts of the architecture over time would work for Intel (even if they had to bifurcate the architecture). What doesn't work for Intel (and its customers) is huge revolutionary new archs with straight-out-of-research advanced features and no compatibility.
- yuhong 10y agoAs an example, it is probably not difficult to create a new x86 version that is user mode compatible with most modern programs but lacks things like segmentation and real mode. New OS versions would be required, but most modern user mode programs would work with few if any modifications.
- nickpsecurity 10y agoThey're already sort of doing that on x86-64 but keeping it in there if you need it. I'm particularly going to miss segmentation as it's phased out as it was used in prior, high-security work plus some current work. Example below: http://dslab.epfl.ch/proj/cpi/ http://dslab.epfl.ch/proj/cpi/ In prior comment, I illustrate with Hardbound they can do a lot more than this. It remains compatible at source-level with x86 apps with good performance while enforcing spatial safety. I could imagine a split version where certain components, esp binaries, run totally unsafe just for compatibility while it enforces safety on whatever you can compile. Alternatively, people wanting backward compatibility with no source could run with safety off whereas anyone (or vendor) with source could compile a version that works. Even the binary stuff might be isolated a bit with techniques that decompile, instrument, and recompile the binaries. Or separate, assembly-level methods. So, Intel could do a lot more than it's doing without totally scrapping backward compatibility or performance. They just... don't. I think they're afraid of loosing billions again or just don't know about the stuff I'm mentioning. Or both. (shrugs) Love to talk to them about it sometime.
- r00fus 10y agoAMD64 wasn't just first; EMT64 (intel's version) is essentially the same thing. If it's x86-64, the processor running your laptop is AMD tech. Period.
- mrweasel 10y ago> I have no idea what the status of that IP is Part of it made it's way into the AMD Athlons, the EV6 bus they used are from the Alpha.
- nickpsecurity 10y agoDidnt know that. Cool that part of it survived. Im mainly interested in the processor IP's. A clone would be a fast RISC with PALcode. Not to mention run VMS, UNIX, etc. Not too practical, though, unless aiming for low-cost given other holdovers (POWER, SPARC) have more software (esp FOSS OS's) plus state-of-art implementations.