If the 480 is 2300 GCN cores and the 1080ti is 3600 CUDA cores

>bigger chip with MOAR COARS
what is Vega

>>59811186
I think AMD is focusing more on the low-end GPU market, since that's the vast majority of GPU sales.

also >>59811335

>>59811335
On top of being a major architectural rework compared to all previous GCN cards

>>59811335
Ok fine, but what's the delay? Slamming on an extra 1300 cores doesnt seem like rocket science...

>>59811452
the delay is, amd makes substantial less money than nvidia and has to make compromises when choosing careful strategies

>>59811470
>choosing careful strategies
What careful strategy? Explain to me like if I were an idiot on the Internet

>>59811452
Even if it's "all the same", a tapeout of a new microprocessor of that size costs millions of dollars and eats away Fab resources AMD might have used for other products.
AMD are not that retarded (for the most part). They will have considered doing just that but it was most likely just not profitable enough.

>>59811506
you see the revenue of amd is like 80% less revenue than nvidia in the consumermarket.

this gives you the advantage of doing unusuable, yet probable stuff which is called competition.

but due to our monopoly paradise, it's hard to climb up that ladder unless you have an absolute edge advantage.

amd is brave enough to try out different routes, but this for the sake of less adoption because most people don't like change, even if it could mean a way better product design in theory

>>59811506
GPUs in the midrange segment net the most profits due to the sheer amount of cards sold and the relatively low cost of manufacturing.
A high end GPU like the Titan or Fury might give you much more profit on a per product sold basis, but in the end there will be far less products sold

So let me get this straight: everyone pretty much knows that AMD could compete if they wanted to, but they chose not to? Boy do I feel stupid!

>>59811452
>>59811186
GCN cores are more complex than CUDA cores

>>59811630
Then why CUDA cores perform better than the same amount of GCN core Rajeesh?

File: 1491703452516.gif (1MB, 218x218px) Image search: [Google]

1MB, 218x218px

>>59811630
??

>>59811654
because developers need to make use of the complexity first. but if the general mass has the simpler nvidia card, why even bother

>>59811785
source on pic related?

>>59811654
also a core is basically 1 engine. and engines differ in size, efficiency and power

>>59811630
What complexity? Thats factually false, stop lying! IS THIS INCEPTION??

You see, adding in moar corz even reaches a limit for the GPU, AMD already has a 4096 core design called Fury X, its problem was FEEDING those cores, so in a lot of shit the cores were left sitting idle.

That what AMD plans to change with Vega, they're keeping the same 4096 shaders, which is pretty much scandalous considering their Fury X was 28nm, and 14nm can fit in almost 2.5x times more transistors per area.

So what's happening with Vega is AMD doing a complete overhaul of their cores, the count will be the same but they'll be much faster and run at much higher clocks (Fury X was 1050MHz)


So yeah, Vega is a complete architecture overhaul like Kepler was, without relying all too much on dieshrinks, will be really nice to see what the result is.

>>59812270
>So what's happening with Vega is AMD doing a complete overhaul of their cores, the count will be the same but they'll be much faster and run at much higher clocks (Fury X was 1050MHz)
That should logically make it *even harder* to feed the cores, not easier! For the last time: STOP LYING TO ME. I think I am entitled to the truth at this point.

File: 1481454591018.jpg (119KB, 796x805px) Image search: [Google]

119KB, 796x805px

>>59812309

>>59812309
What feeds the cores is the scheduler or what AMD calls it, the command processor, it needs its own fixes, Titan XP has like 200 cores less and it has no issues feeding them, what's so hard to believe this is possible?


See Polaris with 2300 cores, being only 10-15% slower than a Fury X, which is 4096 cores with TWICE the ROPs

Per core performance has dramatically improved, this is fact.

>>59811654
Because complexity doesn't necessarily equal performance. In terms of raw compute power, AMD's cards wipe the floor with Nvidia's. But there's more to gay men than raw compute power. The Fury X, for example, was crippled by an anaemic back end (and paltry VRAM), something which was an inherent limitation of the GCN architecture at the time (a maximum of 64 ROPs, i.e. the same as the 290X).

>>59812309
They could have competed with high-end Pascal from the beginning, but thing is that they're contractually bound to use a shitty chip factory (GloFo which used to be part of AMD) making the cost of manufacturing those cards much higher.