It seems AnandTech has got another world's first - this time AMD's Barcelona core was examined. Anand Lal Shimpi reveals all the details you will need regarding the overall chipdesign, it's memory controller, pipeline and the other major and minor of it's features.
Have a good read but don't expect any benchmark figures... ;-)

On the presentation show of AMD's first chipset created in cooperation with ATI, the 690G, it flexed it's muscles in presenting a quad core Barcelona based PC system together with a pair of R600 graphics cards. The cards were running in Crossfire mode and the whole system was tweaked not for gaming but for calculating computational algorithms. It could squeeze out 10^12 floating point operations per second which equals a TeraFLOP/s (a term that get's quite popular today). AMD didn't seem to be tight lipped regarding the postponing of R600 at this event.

At least EE Times and The Register have a good deal more information packed into their articles.

At the press center of AMD there is now a press release covering the story.

For a long time, we've been nicknaming AMD's next major chip release K8L. In truth, AMD never really called their next generation CPU "K8L". The only time they ever nicknamed this new chip K8L was in imaginary roadmaps that were used a long, long time ago. The "L" in K8L actually stands for "low power", which has been used in the Turion 64 series for a year or two now. AMD has officially stated that their next generation of CPU's has the codename of K10. AMD also assures us that the K10 micro-architecture will be a very large step from what AMD currently offers in the K9 micro-architecture.

AMD has released some information about its Shanghai processors - the next generation of server CPUs after the Barcelona cores. The Shanghai processors will be the company's first 45nm chips, which should bring power consumption and heat benefits over 90nm and 65nm equivilents. They will also have 6MB of L3 cache and use the socket 1207 interface, so should be able to work with current Socket F motherboards providing BIOS updates are released. The Shanghai is likely to compete with Intel's 45nm offering, known as Penryn.

Some new details on AMD's next generation of desktop processing have been released, with three different models to go alongside AMD's Barcelona (the processor for servers). The desktop equivalent of Barcelona, with the codename Agena, will feature four cores and should have clock speeds of 2.4GHz to 2.6GHz, with a 2MB L2 cache and 2MB L3 cache per CPU - it's scheduled for the third quarter of this year. The mainstream processor is called Kuma and should launch with clock speeds of 2.0GHz to 2.9GHz using a dual core architecture, with 1MB L2 cache and 2MB shared L3 cache. Rana, the replacement for the Sempron family, will have clock frequencies of 2.1GHz to 2.3GHz, feature dual core and 1MB of L2 cache, but no L3 cache. The Rana is expected for the forth quarter of this year.

Although AMD's next generation of server processing hasn't been a complete secret, here are some of the latest details about it. Codenamed "Barcelona", the processor is likely to go head-to-head with Intel's Xeon 5300, both being quad-core processors. However, one key feature of the Barcelona core is that (unlike the 5300) it will natively be a quad-core CPU, not two dual-cores built into one unit. The chip will continue the advance into 64-bit computing, and like the rest of AMD's 64-bit processors it has a memory controller build into it, another difference between this and the Xeon 5300. Despite much higher performance than older models, the processor will still only use 95 watts or 68 watts of power depending on the different designs, which apparently gives a 60% performance increase per-watt over the older generation Opterons. It can also support up to 256TB of memory, now that the memory controller supports full 48-bit addressing. AMD's benchmarks suggest that it performs much better than the Xeon 5300 due to fewer bottlenecks, but this could obviously be biased against Intel.