AMD announces GPUOpen initiative

[HEXUS]

Applying open source philosophy to drivers, libraries, SDKs and more.

Read more.

[ET3D]

AMD has always been good in terms of backing open standards, that's why I like it. Only problem is that technically AMD products are not always the best.

[lumireleon]

AMD products are usually the best, thats why the same old AMD architecture from 2012 is being used today yet its as fast as Maxwell which is 'FASTER' according to Nvidia adverts not Hexus reviews. SO guess what will happen next year.

[Corky34]

IMHO AMD hardware are usually better technically speaking but their product are not, hardware doesn't exist in a vacuum and a product is much more than just hardware.

The reason AMD have been using the same architecture since 2012, if not longer, is because like all semiconductor chip makers it takes around 5-6 years to design a new architecture so while they're designing a new architecture they make small tweaks to the existing one.

[DanceswithUnix]

The reason AMD have been using the same architecture since 2012, if not longer, is because like all semiconductor chip makers it takes around 5-6 years to design a new architecture so while they're designing a new architecture they make small tweaks to the existing one.

Don't knock small tweaks, it is what evolved humans out of the oceans and then out of the trees

Intel are still making incremental changes to their i7 which if you unravel back far enough gets you to a 1995 Pentium Pro. In 30 years, their only real new architectures have been the P4, Itanium and a couple of Atom designs of which only the last OoO design Atom is still with us.

Big architectural changes have to be driven by a change in needs. GCN was needed to support better shader programming, hence Fermi happened at around the same time. Pentium Pro happened because uop translation seemed the obvious way to create an out of order x86 cpu (hence K5 and then K6 did things the same way at the same sort of time).

AFAICS Maxwell is lots of tweaks in implementation over Fermi, GCN 1.2 is lots of tweaks over GCN 1.0. That is the natural way of things, and anyone wishing for a properly new architecture should be mindful that they might be wishing for the next P4 or Itanium

[Corky34]

Sorry if what i said made it seem like i was knocking small tweaks, wasn't my intention, small tweaks are how we've improved everything around us.

[scaryjim]

... AFAICS Maxwell is lots of tweaks in implementation over Fermi ...

I've always thought of Kepler as the generational marker for nvidia - that's where they went from double-clocked big cores to single-clocked small cores (essentially learning from AMD's GPU designs). Fermi was more the last major revision going right back to G80. AMD's changes since R600 have been more incremental, and GCN is in many regards still just a tweaked R600 core: whilst it moved away from VLIW I think the arithmetic cores remained very similar, while Kepler's much smaller arithmetic cores were very different from the big cores of Fermi.

All that said, I'm a long way from an expert on micro-architectures, and the under-the-hood differences/similarities could be very different from the superficial ones I'm basing my assessment on...