Overclock.net banner

[Various] Exclusive: AMD VEGA Presentation

92K views 1K replies 218 participants last post by  }SkOrPn--' 
#1 ·
AMD VEGA Presentation : Source









An unnamed Vega chip ( According to some Source , it's Vega 10) : Source





Device ID's unnamed VEGA= 687f:C1 : Source



Radeon Vega



AMD Vega & Ryzen - Star Wars: Battlefront 4K Ultra test



Hands LITERALLY On AMD Vega!



AMD's Raja Koduri "The P in PC should stand for performance" - The Full Nerd from CES 2017



Interview with Scott Wasson (AMD) at CES 2017



The AMD Vega Memory Architecture Q&A With Jeffrey Cheng



AMD Radeon RX Vega brand reveal video



Vega: Of Primitives & Pixels. A Look at the GCN Revolution.



Radeon Vega Frontier Spec



Radeon Vega Frontier Edition



Radeon Vega Frontier VS Radeon Fury X

 
See less See more
7 11
#5 ·
:thinking:


Geometry throughput slide : Data base on AMD engineering design of Vega.Radeon R9 Fury X has 4 geometry engines and a peak of 4 polygons per clock.Vega is designed to handle up to 11 [polygons per clock with 4 geometry engines.This represents an increase of 2.6x.
 
#6 ·
Quote:
Originally Posted by Xuper View Post

thinking.gif
Quote:
Geometry throughput slide : Data base on AMD engineering design of Vega.Radeon R9 Fury X has 4 geometry engines and a peak of 4 polygons per clock.Vega is designed to handle up to 11 [polygons per clock with 4 geometry engines.This represents an increase of 2.6x.
11 is a weird number, isn't it ?
 
#8 ·
Quote:
Originally Posted by EightDee8D View Post

11 is a weird number, isn't it ?
to me yeah.but I'm not expert.so I don't know why 11.
 
#10 ·
Looks very promising. I really want to see actual hard facts benchmarks, a release date and info on pricing.
 
#13 ·
Quote:
Originally Posted by prjindigo View Post

It looks like they've installed independent scheduling in the GPU.

Frankly I see nVidia not bringing the 1080ti out yet as a sure sign that the info they've managed to scavenge about Vega is scary to them.
If it scared them they would have release the 1080 TI now, before vega hits the market, and work on getting volta out as soon as possible to combat this.

Also all those HBM memory stacks, I about we will see 16GB on a desktop GPU. It will cost too much. Most likely it is for the vega firepro and the desktop will get 2 stacks of 8GB total. And the problem I think about it, is that they will be similar performing to the 4xHBM1 stacks, just more memory. Compared to GDDR5X, at least memory wise, I don't think nvidia will have much to worry about.
So it will all depend on the internal GPU architecture and how well their scheduler is.
 
#16 ·
Quote:
Originally Posted by Defoler View Post

If it scared them they would have release the 1080 TI now, before vega hits the market, and work on getting volta out as soon as possible to combat this.

Also all those HBM memory stacks, I about we will see 16GB on a desktop GPU. It will cost too much. Most likely it is for the vega firepro and the desktop will get 2 stacks of 8GB total. And the problem I think about it, is that they will be similar performing to the 4xHBM1 stacks, just more memory. Compared to GDDR5X, at least memory wise, I don't think nvidia will have much to worry about.
So it will all depend on the internal GPU architecture and how well their scheduler is.
They are probably waiting on AMD releasing first so they can build a card to beat it in the event of the vega being faster than the 1080/titan
 
#18 ·
Quote:
Originally Posted by Defoler View Post

If it scared them they would have release the 1080 TI now, before vega hits the market, and work on getting volta out as soon as possible to combat this.
Or Vega is so fast it crush the TXP and therefore pointless to bring out the 1080 Ti as they probably can't compete.

They pull back the 1080 Ti, replace it with the full 3840 GP102 and release it again 1-2 months later. All the current supposed 1080 Ti card will have their shroud changed and labelled to GTX 1170 for Q3.
thumb.gif
 
#21 ·
I am not sure how fast Vega can be, but if they improved IPC, Clock Speed as well as count cores over P10... it going to crush all pascal of this gen.

Just imagine its has 20% better IPC, 20% higher clock and 78% more cores (4096 vs 2304). You do the maths, its 256% the performance.
 
#22 ·
Quote:
Originally Posted by Dotachin View Post

All I ask for is a good old DVI port in at least one aftermarket model and I´ll happily ditch Nvidia. Don't pull another Fiji AMD, I use a Qnix.
You know there are adapters from almost any connector to any other? DVI is dead and very limited. Qnix is a hack of a monitor that has nothing but the basic electronics, even the newer shouldn't have DVI only anymore. Just get a DP or HDMI to DVI adapter and use any GPU.

---

Just imagine it will be sold in 2017. Keep on dreamin' and waitin'.

I don't expect anything drastic until Navi. Vega still should be just an iteration of the same old.
Seeing what a flop Fiji was especially with it's pricing and power hunger, we can only hope that AMD managed to finally update the arch. at least a little for better power efficiency and higher clock speeds. P10 is like Maxwell and can't keep up with Pascal.

I bet the pricing especially if it's gonna have 16GB HBM2 will be outrageous. Sure it may force NV to drop the price on it's 1080 and faster cards if those will be challenged in the perf/price but other than that I bet it's gonna quite a big jump in price over the "promissed" $239 of a 480, like 3 times as much.

Sadly Fiji was priced so badly especially in EU that it made almost no sense for anyone to even consider it.

On top of that are they going to finally release an API to use the H265 HW encoder? What about better/faster support for neural networks? I swap between brands freely and right now these two things NV has a monopoly pretty much because they did put the effort/time/money into developing the necessary software part.

Vega can be awesome but it all depends on how it is priced. Overall I do not expect it to be affordable and as such it may follow the same bad fate as Fiji except this time the VRAM may actually be of usable amount.
Go AMD go, but do something about your terrible performance/price ratio finally otherwise it makes no sense to buy your products
frown.gif
 
#23 ·
Quote:
Originally Posted by guttheslayer View Post

Or Vega is so fast it crush the TXP and therefore pointless to bring out the 1080 Ti as they probably can't compete.

They pull back the 1080 Ti, replace it with the full 3840 GP102 and release it again 1-2 months later. All the current supposed 1080 Ti card will have their shroud changed and labelled to GTX 1170 for Q3.
thumb.gif
Lets hope those labels say 1160
smile.gif
 
#24 ·
Quote:
Originally Posted by seabiscuit68 View Post

"Just Imagine"

AMDs new tagline. "Just Imagine we are a competitive company"

So sad what they've become :-(
If that is what they promise.

Its not hard to jump 20% IPC from their 4 years old GCN architecture. Its not hard to boost their clock by 20% since their top speed is only 1.2GHz. They can do that and scale the cores count to which they please. They win.
 
#25 ·
Quote:
Originally Posted by guttheslayer View Post

Or Vega is so fast it crush the TXP and therefore pointless to bring out the 1080 Ti as they probably can't compete.

They pull back the 1080 Ti, replace it with the full 3840 GP102 and release it again 1-2 months later. All the current supposed 1080 Ti card will have their shroud changed and labelled to GTX 1170 for Q3.
thumb.gif
Hell yeah, if vega can make Nvidia do another 780Ti (just like how Hawaii did) it would be awesome XD
 
#26 ·
Quote:
Originally Posted by JackCY View Post

You know there are adapters from almost any connector to any other? DVI is dead and very limited. Qnix is a hack of a monitor that has nothing but the basic electronics, even the newer shouldn't have DVI only anymore. Just get a DP or HDMI to DVI adapter and use any GPU.
Passive adapters don't work for this case, and active adapters cost upwards of 100 bucks, add lag and limit overclocks, my Qnix is a perfectly capable 1440p 96hz ips monitor thank you, it just requires a 2 cents dvi port.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top