The GTX 1080 is $550; Vega 56 is $400 and Vega 64 is $500. With decent cooling, the vega gpus can be overclocked to give significant boost. Also the freesync monitor is cheaper than gsync. Vega56 + FreeSync will save you around $350. That is ~ 25% saving if you are going for a 1.2-1.5k$ rig.
I don't think so. The article touched on the fact that the watercooled one is already seeing a disproportionate ramp-up of power in order to ramp up clocks. That says to me that it's beginning to eat in to the headroom of the card and will soon begin to run up against its voltage wall. That suggests to me that they're already pushing the card pretty hard to get the stock performance levels, which doesn't leave a lot on the table for overclockers.
Here, AMD is once again (see last year's Polaris) a victim of the inferior GlobalFoundries 14nm LPP process. TSMC 16nm would have been much better in perf/W, but sadly a very restrictive wafer supply agreement locks AMD to GloFo for the time being.
It is just sad AMD could not get both Graphics and CPU momentum at the same time.
Poor thermals means that this line of GPUs will find it's way into exactly 0% of the laptop market, which is unfortunate.
Alienware will find a way, even if it requires a water connection to cool it
And after the price parity, you're just stuck with a worse card.
"Same performance as 1080"
Vega 56 is faster than the 1080: 15% faster in SP GFLOPS, 130%(!) faster in DP GFLOPS, and 15000%(!!) faster in half-precision GFLOPS (Nvidia artificially cripples theirs). All comparisons made at base clocks.
"double the TDP"
Merely 17% higher: 210W (Vega 56) vs 180W (1080).
"Only 30% higher than the Fury"
And Nvidia's top 16nm GPU (Titan Xp) is only 39% faster than their top 28nm (TITAN Z) released... 3 years ago. The 28nm→14/16nm transition was hard for everyone.
Saw it on /r/AMD. This is supposedly from SIGGRAPH.
If AMD says you can't go higher than 1080 (in a cherrypicked setting, no doubt) then you cannot make the claim that it is faster.
Benchmarking software should do more to provide a clean and meaningful representation of the distribution of performance - people care more about a card's worst case than its best or average.
Another significant consideration - as alluded to in the TensorFlow comment - is the compute performance.
Although compute performance is not yet that important from a gaming perspective, I would be rather surprised not to see some games start to utilise it for engine functionality - especially after Vulcan and OpenCL merge.
Wait to see the performance of the card as a whole, and consider the implications for the next generation before you declare the card a non-starter and evolutionary dead-end.
2) I'm all for AMD to kick Nvidia's ass in compute and end cuds's reign but it won't sell gaming cards. Vega FE (A 1000$ card) is out for a month and I haven't seen anyone benchmark it and show that it's better, Even if games will use it there is a heck of a difference between building the model and running inference.
The next generation of games could potentially utilise compute functionality, especially if Microsoft and Sony get off their arses and push the capability on consoles.
(That said, porting such software to a PC environment would not be withouut significant difficulty, given the vastly different memory configuration.)
I'll be hugely disappointed in the industry if we don't see some decent Vulcan based game engines in the next few years.
You should check Tech Report's Frame Time Analysis:
http://techreport.com/review/31546/where-minimum-fps-figures...
It seems to be bringing some technological advancement too - using GPU memory as a cache seems like the way forward. Naive calculation gives me a couple of MB to store a single 1080x1920 texture. Current game usage is approx 1000 times that. It takes more than one texture to build a scene, but there seem to be gains if more textures are loaded on-demand (or slightly earlier).