CPR W3 on Hairworks(Nvidia Game Works)-'the code of this feature cannot be optimized for AMD'

humbug · 27 May 2015 at 17:31

Silent_Scone said:
Talk about missing the point lol. Honestly, this is what you get for taking one statement I said and trying to turn it around with an outdated benchmark.

Both you and mmj need to read his link properly, they state very clearly that the 285 has 2x the tessellation throughput of the 290X despite only being half the card.

It seems you only read one paragraph, the 285 trails the 290 in low tessellation factors

Ofcourse it does, the 2816 Shaders are not bottlenecked by the 290 much narrower pipelines at low tessellation, its only when you turn the tessellation up that the 1796 GCN 1.2 front end shaders overtake the then bottlecked 2816 shaders of GCN 1.1.

GCN 1.2 has 4x the throughput of GCN 1.1

Silent_Scone · 27 May 2015 at 17:39

Don't talk at me I wasn't the one that quibbled GCN performance. You were at cross purposes with that graph.

4x the performance you say?...

humbug · 27 May 2015 at 17:44

Silent_Scone said:
Don't talk at me I wasn't the one that quibbled GCN performance. You were at cross purposes with that graph.

4x the performance you say?...

Don't stick your oar in then ^^^ lol

mmj_uk · 27 May 2015 at 17:44

humbug said:
Both you and mmj need to read his link properly, they state very clearly that the 285 has 2x the tessellation throughput of the 290X despite only being half the card.

Because with GCN1.2 they've once again doubled the number of geometry engines (including the tessellation unit) compared to GCN1.1/290X, just as they did with GCN1.0/280X.

http://www.anandtech.com/show/7457/the-radeon-r9-290x-review/3

With Hawaii AMD has doubled the number of geometry engines from 2 to 4

Look at the diagram, each geometry engine contains a single tessellation unit.

XFX Launches its Radeon R9 290X Graphics Card

XFX announced its Radeon R9 290X graphics card, the R9-290X-ENFC. The card is every bit identical to AMD's reference design, and includes an Origin key to Battlefield 4. Based on the 28 nm "Hawaii" silicon, it features 2,816 stream processors, 176 TMUs, 64 ROPs, 4 tessellation units, and a 512-bit wide GDDR5 memory interface, holding 4 GB of memory. It features reference clock speeds of 1000 MHz core, and 5.00 GHz memory. Available now, the XFX Radeon R9 290X is priced at $570.

There is no link between number of stream processors and tessellation units with AMD GPU's because they have fixed units.

humbug · 27 May 2015 at 17:50

mmj_uk said:
Because with GCN1.2 they've once again doubled the number of geometry engines (including the tessellation unit) compared to GCN1.1/290X, just as they did with GCN1.0/280X.

http://www.anandtech.com/show/7457/the-radeon-r9-290x-review/3

Look at the diagram, each geometry engine contains a single tessellation unit.

Yes, now increase the shader count and do the maths.

Gregster · 27 May 2015 at 18:09

@humbug. No I wasn't being deliberately (ly) dumb lol. You posted the graph and I took it for face value and gave my comments on it. AMD are pretty rubbish at tessellation, so even when AMD users finally get some Witcher 3 drivers, I still feel they will struggle and will need to use the slider for tessellation in the CCC.

It isn't rocket science lol.

Gerard · 27 May 2015 at 18:31

Gregster said:
@Humbug. No I wasn't being deliberately (ly) dumb lol. You posted the graph and I took it for face value and gave my comments on it. AMD are pretty rubbish at tessellation, so even when AMD users finally get some Witcher 3 drivers, I still feel they will struggle and will need to use the slider for tessellation in the CCC.

It isn't rocket science lol.

Or they're pretty rubbish at doing over tessellation? There's plenty of pics showing that visually there's literally bugger all difference between the amd optomised implementation and nvidia's totally over the top implementation. Especially for something like hair ffs.

Gregster · 27 May 2015 at 18:35

Gerard said:
Or they're pretty rubbish at doing over tessellation? There's plenty of pics showing that visually there's literally bugger all difference between the amd optomised implementation and nvidia's totally over the top implementation. Especially for something like hair ffs.

No need to get all Davey and I was just going on Humbug's graph. I have everything maxed out and it runs 'sweet as' for me, so no need to worry if people do use lots of tess

Silent_Scone · 27 May 2015 at 18:37

For the record I just tried lowering HairWorksAA to 4x and the difference in framerate was literally nothing, not 1 frame.

Nexus18 · 27 May 2015 at 18:42

Hold on, why are we comparing the 980 and 290, the 290 is over a year older :confused:

Of course it is going to be worse in pretty much everything...

But yes I won't disagree that the "current" (excluding the 285) AMD cards are super crap for tessellation compared to nvidia cards, even the 780 (released same year as the 290) is far better than the 290 for tess.

But humbug has a point regarding the 285, it competes with a nvidia 760 in terms of overall performance in games and its tess. performance is a decent bit higher so I think it is safe to say that AMD's new lot of cards will be a lot better for tessellation, at least matching or even surpassing the 980/970 for tess., we will just have to wait and see....

Silent_Scone said:
For the record I just tried lowering HairWorksAA to 4x and the difference in framerate was literally nothing, not 1 frame.

Are you sure that is tessellation and not anti-aliasing?

Silent_Scone · 27 May 2015 at 18:47

It is what it says it is, the level of anti aliasing applied to HairWorks, lowering it to 2x offers 2 or so FPS but it looks noticeably worse.

Gerard · 27 May 2015 at 18:51

Gregster said:
No need to get all Davey

Mei · 27 May 2015 at 18:51

if you talking about the ini file setting i believe thats AA levels and doesnt effect the tessellation

kind of in the name "HairWorksAA"

Nexus18 · 27 May 2015 at 18:52

Silent_Scone said:
It is what it says it is, the level of anti aliasing applied to HairWorks, lowering it to 2x offers 2 or so FPS but it looks noticeably worse.

I thought it was the tessellation of hairworks that kills FPS, not the anti-aliasing level?

Phixsator · 27 May 2015 at 18:56

I think it would be healthy for some of you gents to take a step back and breath in some fresh air before you murder eachother

. Now let us all sit peacefully in a circle and sing some Kumbaya.

humbug · 27 May 2015 at 19:00

Gregster said:
No need to get all Davey and I was just going on Humbug's graph. I have everything maxed out and it runs 'sweet as' for me, so no need to worry if people do use lots of tess

Clearly you are not going off my Graph as you are ignoring the comparable Tessellation performance to Nvidia on it, don't tell me you didn't see it.

Mei · 27 May 2015 at 19:03

why was there a kangaroo in that vid

Phixsator · 27 May 2015 at 19:09

Mei said:
why was there a kangaroo in that vid

i have no idea lol i just picked the first i found on youtube

Gregster · 27 May 2015 at 19:10

Gerard said:

Auto correct on my phone (at work) and should have said "Ragey".

We had all this all the way back with Crysis 2. AMD had problems with tessellation back then.

Edit:

Thread is getting very ill tempered and I will drop out of it for now.

Silent_Scone · 27 May 2015 at 19:12

Nexus18 said:
I thought it was the tessellation of hairworks that kills FPS, not the anti-aliasing level?

Depends who you ask. Some people have been blurting that 8XMSAA was excessive but it doesn't really have a huge performance impact as it's only applied to HairWorks. The amount of tessellation is indeed the main performance culprit, but lowering it also has quite an impact on rendering quality as can be seen in the AMD screenshot example