Yet it seems either VMs weren't in focus when they made the Bulldozer design, or there's something in DosBox in particular, but this doesn't only make BD ugly, it's a complete disaster:
http://pctuning.tyden.cz/hardware/procesory-p … sktopu?start=13
Haswell runs DosBox 2.5 times(!) faster than BD (clock-to-clock). This is why not having a Trinity test bothers me, but I don't expect a huge improvement. I also couldn't find a comparable test with older CPUs, so I'm somewhat in the dark.
I can only hope they keep to the habit of testing with DosBox at least until Kaveri arrives. That's said to have changes in its architecture to alleviate a lot of bottlenecks (like they'll double its decoders). We'll see if the problem is there, or somewhere else. BD (Trinity, Kaveri) is rumored to have quite a long pipeline anyway (I haven't seen any official data about it). Probably not as long as the infamous Prescott, but surely longer than K10 or Core designs.
And that long pipeline didn't work too well, it had a very unbalanced performance. It could have been decently strong (and outstanding in a few areas) if they could double its clock rate (they planned a 10GHz version later), but it proved impossible.
Bulldozer seems to be in the same trap: it's unbalanced and looks to be designed for higher clocks which it can't reach. My question is that if it's because of a long pipeline, or other bottlenecks? If it's the former, they'll have to do something to rev it up within a reasonable thermal envelope (energy consumption), or look for some other design (Jaguar, maybe?). If the latter, they can still hone the existing design to make it better.
(Because it looks very interesting. That shared FPU and the future common address space with the IGP never ceases to amaze me!)
Shame on us, doomed from the start
May God have mercy on our dirty little hearts