Media slides AMD Radeon Pro Duo leaked

	Radeon Pro Duo	Radeon R9 Fury X	Radeon R9 Nano	Radeon R9 Fury
GPU	2x Fiji XT	Fiji XT	Fij XT	Fiji Pro
Stream-processors	2x 4.096 (8192)	4.096	4.096	3.584
TMU's	2x 256 (512)	256	256	224
ROP's	2x 64 (128)	64	64	64
Perf	16,4 TFLOPS	8,6 TFLOPS	8,2 TFLOPS	7,2 TFLOPS
Core clock	1.000 MHz	1.050 MHz	1.000 MHz	1.000 MHz
mem speed	500 MHz	500 MHz	500 MHz	500 MHz
mem bus	2x 4.096 (8192)-bit	4.096-bit	4.096-bit	4.096-bit
Mem	2x 4 GB (8 GB) HBM1	4 GB HBM1	4 GB HBM1	4 GB HBM1
Bandwidth	2x 512 (1024) GB/s	512 GB/s	512 GB/s	512 GB/s
TDP	350W	275W	175W	275W

Ryu5uzaku
Senior Member

Posts: 7037
Joined: 2006-09-24

#5263163 Posted on: 04/26/2016 11:50 AM

what i meant is:
if you enable nvidia-hairworks the fps difference between the radeon pro and the 980ti would maybe be at 0.
so turning it off for a benchmark is a way to make it obvious that the setting is distorting realistic/significant results.

The difference would be near 0 if the 980 ti would be oc'd when that hairworks is enabled. Since the hit is not THAT bad anymore even if it is rather bad. It is in essence two furies in the end.

Fox2232
Senior Member

Posts: 11803
Joined: 2012-07-20

#5263183 Posted on: 04/26/2016 12:31 PM

Then why benchmark the game, it's a 1500 dollar card, a halo product. Hairworks is eyecandy, eyecandy that people who spend 1500 dollars on a card probably wants. Sorta the reason to include 4k.

TressFX is eye candy. Hairworks even with its modern iteration in Witcher 3 looks worse than standard pre-scripted hair.
There is some difference in movement of hair, but visually, it looks like Witcher has low amount of pretty thick hair or a lot of thin pigtails (but then it does not have proper texture).

Witcher 3 is good game, but not thanks to Hairworks. On wild animals it even looks damn bad. But what do you expect it to look like on animals?
Geralt has few hairs on him and performance impact is very noticeable. What if you had to deal with just one wolf with body properly covered in fur on top of that? Or with entire pack? (As that's usual encounter.)

vase
Senior Member

Posts: 1653
Joined: 2015-03-20

#5263190 Posted on: 04/26/2016 12:44 PM

Then why benchmark the game, it's a 1500 dollar card, a halo product. Hairworks is eyecandy, eyecandy that people who spend 1500 dollars on a card probably wants. Sorta the reason to include 4k.

My point is. 4K is a standardized non-proprietary requirement of a GPU. Resolution scaling up to 4K and more.

Gameworks Hairworks™ is a proprietary piece of closed source code which is designed to cripple performance on non-nvidia cards.

So how can you call it a benchmark under standard conditions if you use proprietary extras that bias/distort a result that much in framerate/-time as not even 16x/8x MSAA vs AA disabled would manage to do?

You cant put a Porsche and a Lamborghini on a test track which has been optimized especially for Porsche tyre material only and then call it a fair comparison.
You have to test them on test tracks with no manipulation of either side to find out what car performs better.
Yes the track can be 1 mile long (1080p) it can be 2 miles long (1440p) it can be 3 miles long (2160p) - it can have almost no curves (AA disabled) to many curves (AA x8) ... that is all a sports car has to be good and deal with.
But special road surface preparation (Nvidia Hairworks™) for one car vendor only is not a thing you want to have in a benchmark/comparison.

And to close the circle... EXACTLY THAT is the reason why they disabled hairworks in this benchmark. Because it wouldn't show the true performance of both cards.
To be clear: I wouldnt want a benchmark comparison with AMD features enabled either (apart form knowing that AMD doesnt have many closed source standards to begin with which would try to bully their competitors out)

Keesberenburg
Senior Member

Posts: 852
Joined: 2013-01-11

#5263204 Posted on: 04/26/2016 01:15 PM

600 dollar vs 2500 dollar, better buy 3 gtx 980 ti's for 1800 dollar, much faster

Denial
Senior Member

Posts: 13320
Joined: 2004-05-16

#5263205 Posted on: 04/26/2016 01:16 PM

My point is. 4K is a standardized non-proprietary requirement of a GPU. Resolution scaling up to 4K and more.

Gameworks Hairworks™ is a proprietary piece of closed source code which is designed to cripple performance on non-nvidia cards.

So how can you call it a benchmark under standard conditions if you use proprietary extras that bias/distort a result that much in framerate/-time as not even 16x/8x MSAA vs AA disabled would manage to do?

You cant put a Porsche and a Lamborghini on a test track which has been optimized especially for Porsche tyre material only and then call it a fair comparison.
You have to test them on test tracks with no manipulation of either side to find out what car performs better.
Yes the track can be 1 mile long (1080p) it can be 2 miles long (1440p) it can be 3 miles long (2160p) - it can have almost no curves (AA disabled) to many curves (AA x8) ... that is all a sports car has to be good and deal with.
But special road surface preparation (Nvidia Hairworks™) for one car vendor only is not a thing you want to have in a benchmark/comparison.

And to close the circle... EXACTLY THAT is the reason why they disabled hairworks in this benchmark. Because it wouldn't show the true performance of both cards.
To be clear: I wouldnt want a benchmark comparison with AMD features enabled either (apart form knowing that AMD doesnt have many closed source standards to begin with which would try to bully their competitors out)

The source is actually available to anyone now and while it may not be the optimal method of hair simulation, it's definitely not "designed to cripple performance". The developer of the game can set both the tessellation level and AA level on the hair itself. It's all configurable variables in the SDK.

Click here to post a comment for this news story on the message forum.