The AMD Problem.

[Najuno]

Kind of funny how people like to believe in graphics card manufacturers like they were their spiritual gurus .

 

So many well educated people here, that know so much and yet they fight like dogs over a bone... but isn´t topics like this that lead to these discussions?!

[App4that]

Kind of funny how people like to believe in graphics card manufacturers like they were their spiritual gurus .

So many well educated people here, that know so much and yet they fight like dogs over a bone... but isn´t topics like this that lead to these discussions?!

LOL!!! It is like talking about religion isn't it?

[Coaxialgamer]

600 series was unmatched by AMD. 7000 series were loud, power hungry and slower. That's why 7970 ghz edition had to come out and shit like that.

Well originally the 7000 series where set to compete with the 500 series, in fact the 7970 came out months before the 680.

[i_build_nanosuits]

snip!

AMD is where they at right now due to 10+ years of very poor decisions and bad thinking...it's not gonna change overnight.

[Coaxialgamer]

The architecture is getting old and it has a lot of weaknesses indeed...somehow people seem to think those are brute-force powerhouses because they consume a lot of power, they run hot and they ship with a lot of memory...this to me is inefficiency, and lots of useless video memory...not brute-force.

here, this user submitted benchmark database rank the 390 above the 970, the 390 according to the benchmarks is 4% faster than the 970 overall when no driver issues are stopping it from being faster...but if you look at it closely the 970 is faster at rendering, NB calculations, lighting and reflection...all very nice to have, meanwhile the 390 is better at texture due to the better memory configuration but that can only go so far when it comes to gaming:

http://gpu.userbenchmark.com/Compare/Nvidia-GTX-970-vs-AMD-R9-390/2577vs3481

The thing is you mostly judge a card's compute performance by looking at floating point calculations, and in that department, gcn beats maxwell by a lot.

[Coaxialgamer]

cutting the price of a product, even a Hawaii core, which is a very mature process by now, will eat into your margins...

now... how much margin do you think AMD has on the 390?

100 bucks? 200 bucks?

add in shipping, handling, import/export taxes, marketing, production costs etc... ill say AMD most likely have 150$ margin on their 390... probably less.

if you cut the cost by 50-80 bucks, you gotta do so out of AMDs margin. because no retailer on earth will cut THEIR margin to sell an AMD product. Just aint happening.

So that means AMD makes A LOT less money per card. Given their low market share, they do not have the "brand name" to do the marketing for them. So they would simply make less money, have less money to spend on marketing. Thus selling even less cards.

what AMD needs is to release FEWER PRODUCTS....

do a Nvidia, just cut down cores..

why have Bonaire, Pitcairn, Tonga, Hawaii and FIji???? WHY?

then you have a "Pro and XT" version of these.... WHY?

just do "Tonga" for lower end

Hawaii for mid to high range

Fiji for max

you could easily disable some ROPs/SPs in Tonga in order to "cut it down" enough to perform at 260X levels...

by doing so, a single fucking GPU core could span 3 cards... which means LESS COST....

Pit Hawaii XT (290X/390X) vs the 970.... you would absolutely demolish the 970

if a 390X cost the same as a 970....

then just leave Fiji to deal with the rest.

THIS would have been a good idea... this way you need THREE PRODUCTS.... you can just disable SPs/ROPs/TMUs in BIOS, or do it on a hardware level....

they could just say "Custom bios = no warranty.... so good luck bios modding your 260X Tonga chip to a 380X tonga chip.... cuz we wont care what happens"

Thats exactly what they do :

Pro and xt are both an iteration of one chip. For example :

380 and 380x are both the same chip ( tonga pro and xt) 390 and 390 are both grenada (xt and pro) fury, fury nano and fury x are all based on the fiji core.

Every chip manufacturer today sells cut down versions of chips in order to increase yields.

[Prysin]

AMD is where they at right now due to 10+ years of very poor decisions and bad thinking...it's not gonna change overnight.

i know... atleast they are on the right track...

 

but, it takes no economics genious to understand that it costs more to produce multiple products, then it costs to produce fewer and choke them...

[Prysin]

Thats exactly what they do :

Pro and xt are both an iteration of one chip. For example :

380 and 380x are both the same chip ( tonga pro and xt) 390 and 390 are both grenada (xt and pro) fury, fury nano and fury x are all based on the fiji core.

Every chip manufacturer today sells cut down versions of chips in order to increase yields.

not quite.

 

what they do is make

Bonaire GPU = R7 360 (GCN 1.1) (128bit GDDR5)

Pitcairn GPU = R9 370 (GCN 1.0) (256bit GDDR5)

Tonga Pro GPU = R9 380 (GCN 1.2) (256bit GDDR5)

Tonga XT GPU = R9 380X (GCN 1.2) (256bit GDDR5)

Grenada Pro GPU = R9 390 (GCN 1.1) (512bit GDDR5)

Grenada XT GPU = R9 390X (GCN 1.1) (512bit GDDR5)

Fiji Pro GPU = R9 Fury (GCN 1.2) (HBM)

Fiji XT GPU = R9 Nano (GCN 1.2) (HBM)

Fiji XT GPU = R9 FuryX (GCN 1.2) (HBM)

 

now.. here you have 5 GPUs. 5 different architectures. 5 different production lines and methods. 3 different generations of GCN....

Sure, they are all 28nm (thank god for that atleast)... but they need different etchings. different PCB power delivery, 4 different memory systems which means FOUR DIFFERENT MEMORY CONTROLLERS...

 

you see how this will add A LOT of costs?

they cannot just buy a massive bulk of ONE part for power delivery or memory chips. they need to buy different chips, that is compatible with different bus sizes, transfer speeds, VRM solutions and whatnot.

 

this all surmounts to money money money money.... wasted.