Why does AMD keep failing at making an efficient and powerful GPU?

Why does AMD keep failing at making an efficient and powerful GPU?

Shouldn't the shrinkage to 7nm make this pic related thing run much cooler and faster?

Attached: R7_Car2.jpg (1500x890, 367K)

it's a rebranded instinct, it's made exclusevely for compute, it was only released because turding is terrible. power consumption and temperatures depend just on how you tune the card, anything can be made efficient by lowering the voltage

Anyone who bought this instead of waiting for Navi is a retard

HA i remember wait for vega
Maybe once Navi is released and underwhelms, wait for 'next-gen'

Vega was a failure because it was expensive to manufacture and consumers had to pay the brunt of it. Navi wont have this problem

Vega was overhyped like how Navi is currently.

At this point making the process smaller doesn't do so much towards making things "cooler and faster" because you have to deal with leakage being a problem and with the processor in an ever smaller area you're pretty much focusing about the same amount on energy into a smaller area, which is more difficult to draw away (heatpipes directly contacting the processor are more effective than heatpipes in secondary contact with the processor and so on).
It's more about just getting more processors per wafer printed, which means higher profits.

Don't forget that Raja also spent more money to shit out unnecessary products for his bois in bollywood than fixing gcn shortcomings.

Are you gonna say Vega 56/64 were also rebranded Vega Frontier Editions? Wtf you smoking, it's the same GCN architecture even if it is from leftover instincts.

Navi is an upgrade to Polaris. AMD will not have anything faster than a Radeon VII in 2019.

Dunno, but NVIDIA is able to get, sometimes massive perf gains from shrinkage. See Maxwell to Pascal, big jump in power consumption/perf.

Maybe because they have a really efficient power scheduler...

It's not even close in hype level, the fuck are you talking about. Navi is hyped to be 1080 for $250 3 years after 1080, which is just a normal progress, vega was hyped to be volta killer.

>big jump in power consumption/perf.
Because 16nm was huge jump from their previous node for tsmc, much bigger than 28nm glofo to 14nm lp node that was made for smartphones since glofo failed to make their own high perf. node.

>Thanks to this, the GTX 1660 Ti is not just 36% faster, it’s 36% percent more efficient as well.

THANK YOU BASED NVIDIA

>Vega was a failure because it was expensive to manufacture and consumers had to pay the brunt of it.
So what happened to the "Vega is actually really cheap all things considered, AMD is selling them at a loss!" narrative?

AMD cards are based if you care about Linkuks gaming.

my undervolted VII is as efficient as 2080

>radeon vii

so its a 7th rebrand of rx 480? or a rebranded fiji i guess?

>Muh undervolt

Because the die is shit, it doesn't come efficient from the factory.

No it's Fiji, GCN based. RX series is Polaris, very differnet.

It's just a shrunk down Vega 64 (with 4 less CUs) and twice as much HBM2 memory bandwidth.

It's a rejected MI50 card.

>AMD is selling them at a loss
was selling*
It's m6ch cheaper now, both chips and memory.

They're not selling it at a loss, it's just not profitable. It was estimated that the Radeon VII cost about $450 for the card and parts alone, so it's not a big profit margin.

because its not a new gpu

its a vega 60 on 7nm process

any gains to thermal performance and efficiency were used to push the clocks up out of the box

>hurr hurr my graphics card is thinner so it has to be faster

common sense is that a thicker wire has lower resistance... aka it doesnt heat up so fast and it doesnt consume as much energy... But of course you are a gaymerFag so what would you know about physics ?

Well what's AMDs plan?

Radeon VII is on par with the 1080Ti, which is a 2 year old card.

vega 64 was built to be good for everything. there was a lot of advertising regarding its gaming features. it was launched for consumer, pro and server.
radeon 7 has high dp, super-expensive 4 stacks of hbm and nothing new for gaming or pro. it's a hpc product and they never announced launching it outside server market, then they suddenly put out a consumer version of it without even bothering to cut the vram in half to make it well priced.

>take 14nm chip
>cut it in half
>it's 7nm now lol

>Shouldn't the shrinkage to 7nm make this pic related thing run much cooler and faster?

Yes, you can cut the power cost by about 50%, but you can also opt to increase the performance or clockspeed (which they did), or add more silicon (in this case improved FP units but fewer working cores, 60 instead of 64). Keep also in mind that it has 16 GB of VRAM instead of 8 of the Vega 64.

>hurr durr I'm a brainlet who half-remembers one physics lesson, die shrinks should result in less efficient and hotter chips!

A smaller node means lower voltage required for transistors to work. That means less current, resulting in less waste heat and lower power consumption at the same frequency.
The only detrimental effect is that with a smaller chip there is less surface area that can be used to transfer the heat away.

Weren't people saying the HBM cost $350 alone?

Actually making a Radeon VII 8GB is a tiny fraction in terms of cost and energy saving.

HBM2 is low powered RAM.

I had the Vega FE which is 16GB of VRAM just like my current Radeon VII, not much of a difference in power consumption between the two.

ACtually the Radeon VII is partially better at power handling since it's a smaller die and higher clocks.

Yeah, add another $50 for the cooler/shroud, then another $50~ for the die and you get a Radeon VII.

Most likely the cost of the whole card costs AMD $450-$500.

Selling it at $699, for a $200 profit is a very low margin since there's R&D involved. That's the reason why there are no custom AIBs available. AMD knows GAYMERS will buy GTX2080 at least anyway...

But I do have the Radeon VII it's a very nice card especially water cooled.

> Shouldn't the shrinkage to 7nm make this pic related thing run much cooler and faster?
Sure, if you go back to original Vega clocks and voltages.
> muh gains
Shrinks aren't supposed to have any perf gains.

pipeline bubbles and bottlenecks that make the architecture shit out past about 2560 shaders

Attached: 1537606375579.jpg (684x626, 230K)