[EDIT]hmmm.....whats going on at AMD? (and history of 3Dfx Rampage)

[BoFox]

Amd raja....need his head checked.  All that talking up he did (while talking down pascal) what a freaking joke.  

And this part is especially, i want bofox to take note of.
Take a look at the 380x the full tonga to which I believe that Fiji and the 480 come almost directly from.  The 480 is running about 30% higher clock speed and has 15% added compute units.  The 480 is not anywhere I would expect it to be given l the hype raja went on and on about, how they have created a radical newly improved design.  It's just not.  The tweaks they did to the arch were almost no more beneficial, It's just truly pathetic.

I am making such comparison to contrast all those saying pascal was reduced ipc. There are some stark differences, one being that the gp104 has a smaller bus, less bandwidth, far less rops, far less cores than the 980ti...it's logical that it would have some reduction in IPC at the same clock speed as the 980ti.  It's a much more stripped down and lean chip.

The 480 vs 380x- more compute units, higher bandwidth, higher speed, more memory....it's just not a great result.


Just clock speed increase alone, 380x times 130% and it's at 480 performance.  This is not taking in account the 15% extra compute units..which are also running 30% faster. It's just unbelievable, where is this per core improvement amd has chirped about?  Well, interestingly enough, they claim that the 480 cores are 15% "faster" than those on the r9 290.  Haven't crunched those figures yet but since the 380x has the same bus size, I didn't want to skew the results unfairly using hawaii's large bus against the 480.  The 380x is, for all intents and purposes, a pretty direct point of reference. The added cores of the 480 put the 380x at the disadvantage yet clock speed alone would account for almost all of the improvement we see.

It's very very interesting.  There has to be some reason these "improvements" amounted to very little.  Amd talks about many of them and have graphs and claims of huge gains...but they are not there in the real world.  They are simply not adding up.  How?  Amd lies with hyper exaggerated figures, compulsively....or they are really lacking on producing outcome from some of these good ideas.  Relying mostly on full computer Automated chip design would surely hold back, especially in finet, which I hear has to be worked st painstakingly with engineers piecing together like a jig saw puzzle in order to extract and harness the technology fully

Ahh, the automated chip design!  I actually forgot about this - thanks for reminding me about this.

The chart you posted shows it losing out to the 390 overall.  Anandtech's review shows it losing to a GTX 970 overall (IIRC), which is even worse.  Yet, many review sites show it to beat the 390 just by a hair, overall.  I think that in the future, it will just continue to pull a bit ahead of the 390.  

The 380x Tonga already has some stuff that the 390 does not (memory bandwidth compression, slightly improved cores with further tessellation optimizations, etc. (GCN 1.1 vs 1.2 - like Fury)  So that could probably lie somewhere in between the claimed "15% more IPC" over the 290, which is GCN 1.1, like Bonaire before Hawaii).  

I loved the 512-bit bus on my R9 290, but if I had to choose between a 390 and a 480 8GB, I'd get the 480, even if a new 390 was $50 less (and then undervolt the thing while maintaining the same stock clockspeed - and make my own custom cooler for it).  Many years ago, I tweaked CPU coolers and fitted them onto my Radeon 9800 Pro and my 7800 GTX cards - they were kick-ass awesome..  ahh, I was soo proud!