AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread

Look at the picture under the graph, they have the card.

 

Maybe but the author claims to have actual numbers. Just matching the 980Ti will be great for competition and prices. Assuming AMD follows its recent pattern of offering similar performance for much less $$.

 

I was expecting a lot (my expectations about AMD are pretty low these days), but I definitely expected at least 10-15% performance over a Titan X. The minimum performance improvement over Hawaii is 45% due to pure CU increase. Add 5% in clocks, and we're at 52%. And then the big one: this should hardly ever be constrained by memory BW. We don't know how often that really happens on a Titan X, but 30% is definitely a conservative estimate. So add another 15% with that bottleneck removed? Apparently not...
It's not R600 territory, but it's close.

Edit: I meant to say "I wasn't expecting a lot..." Duh.

 

early driver.
what about dx12 and win 10 as that is by far more interesting than dx11.
seems to beat the competition and seems as predicted.

 

Are we looking at the same thing? Fiji is 56% faster than 290X in that 3DMark bench. That surely puts it above TX at least 10%?

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_980_Ti/31.html

 

This even I do NOT expect. Your expectations are definitely out of this world.

Realistically, if it is on par with the competition, this will already be a huge success.

 

The 980Ti OC numbers look ominous for AMD, but this is nowhere close to being R600 territory.

 

https://www.reddit.com/r/Amd/comments/399775/new_picture_of_fury_nano_w_backplate_tubing_and/

That's one pretty powerful fan there -- Scythe?

 

They need the most advanced and expensive memory in existence to match the performance of a mainstream solution? My standards for great success (tm) are higher than that...

 

I don't know: I'm looking at the VC link that shows Fiji and Titan X performing identically.

 

Why ? Fiji will also get over clocked.

 

Since the memory also comes with 4 GB limit, it would be too wasteful to design faster chip.

 

Hope there will be card reviews of apples to apples comparison, "air to air" or "hybrid to hybrid". Anything else is embarrassing ....

 

Yeah I see now. I was just looking at performance relative to 290X. So if we forget about the clock difference thats
54 and 51% faster respectively.

FuryX 7873
FuryX 3960

290X 5091
290X 2617

 

How does an apple to apple review look like? Reviewers always test a whole bunch of reference and aftermarket cards
and then you can make your judgment based on price/performance/noise etc.

 

There are tons of games that work with 4GB just fine. The 4GB is more a marketing problem than a performance one.

 

HBM is officially not even out yet and you already transformed GDDR5 into a mainstream solution. Sorry but it will be only when all high end graphics cards ship with either HBM or HMC.
HBM is AMD's own development, so it is understandable they will use it and not anything else.

The memory type itself is not the decisive factor... maybe the Fiji chip is slower overall and no matter what memory you throw at it, there will be no much if any gain.

And still, I wouldn't be so convinced that HBM is the most expensive solution.

 

That's exactly my gripe: it looks like they wasted a fantastic piece of enabling technology on something that doesn't deserve it.

Do you honestly believe that?

 

Not and argument for this performance class - there are also tons of games that work with 4MB just fine.

 

290X is slowest.