Poll: The Vega Review Thread.

[RavenXXX2]

This shows the nvidia drivers are buggy? They complete ignore the refresh rate settings.

Or lets say, if I have a 4k monitor with 60Hz refresh rate, I expect the GPU to downclock once reached. The same reviewer made also BF1 or Witcher3 where the Vega 64 looks much better AC seems the worst of the benchmarks so far.

Nope that's vsync which is disabled.

 

[LoadsaMoney]

so everything on Vega works, and they got no improvement at all over Fiji, so its basically just a higher clocked Fiji, with 8GB vram, instead of 4GB ?

 

[spikerules]

thats correct

So theoretically PS4 Pro could potentially be as powerful as the Xbox one X? Why did I order an X

 

[Shin1234]

So theoretically PS4 Pro could potentially be as powerful as the Xbox one X? Why did I order an X

its going tobe swings and roundabouts, xbox one x and ps4 pro will trade blows.

 

[Rroff]

This shows the nvidia drivers are buggy? They complete ignore the refresh rate settings.

Or lets say, if I have a 4k monitor with 60Hz refresh rate, I expect the GPU to downclock once reached. The same reviewer made also BF1 or Witcher3 where the Vega 64 looks much better AC seems the worst of the benchmarks so far.

What do you mean?

 

[CAT-THE-FIFTH]

I don't believe the Xbox one X supports rapid pack maths. There is no fp16 for Xbox one x

It apparently has some features like the shaders which are derived from Vega.

 

[Gregster]

This shows the nvidia drivers are buggy? They complete ignore the refresh rate settings.

Or lets say, if I have a 4k monitor with 60Hz refresh rate, I expect the GPU to downclock once reached. The same reviewer made also BF1 or Witcher3 where the Vega 64 looks much better AC seems the worst of the benchmarks so far.

Not sure what you mean but NVidia drivers are certainly not buggy in AC:U, it is a game I loved and completed twice.

 

[shankly1985]

It apparently has some features like the shaders which are derived from Vega.

Interesting read here
http://wccftech.com/rapid-packed-math-ps4-pro-gap-xb1x/

 

[spikerules]

It apparently has some features like the shaders which are derived from Vega.

So fp16 is fundamentally akin to a compression method? Where fp32 isn't required fp16 would be used?

Do we know what scenarios would allow for fp16 over fp32?

 

[Gerard]

Show us.

On the worst case scenario map, fps goes to a minimum of 60, on other maps like sinai or monte grappa the framerate is in the 80's + most of the time. Upped in 4k so should be processing to that soon.

 

[CAT-THE-FIFTH]

So fp16 is fundamentally akin to a compression method? Where fp32 isn't required fp16 would be used?

Do we know what scenarios would allow for fp16 over fp32?

I think its more about being able to render certain parts of the scene using lower precision to boost framerates?

Interesting read here
http://wccftech.com/rapid-packed-math-ps4-pro-gap-xb1x/

Thanks - I will have a read.

 

[spikerules]

I think its more about being able to render certain parts of the scene using lower precision to boost framerates?

Does that effect visual quality at all?

 

[CAT-THE-FIFTH]

Does that effect visual quality at all?

Looking at the video in the link shankly1985 posted,it could be because a lot of stuff which is rendered is not actually seem fully(like the hair stuff),or simply does not need that level of precision for a game.

I expect Nvidia will support it with Volta,as they are talking about it supporting mixed FP16/FP32 precision too.

 

[Shin1234]

I think its more about being able to render certain parts of the scene using lower precision to boost framerates?

The way i had it described to me its more a case that a job is 64, that job can be split down, so 64/32/16/8. The more you split it down the easier the job is todo in parralel aslong as you have the ability todo it in paralel. So running 32 you do the job of 64 in parralel, then again at 16 you do the job of 32 in parralel. So fp16 is basically hyperthreading of 32.

 

[CAT-THE-FIFTH]

The way i had it described to me its more a case that a job is 64, that job can be split down, so 64/32/16/8. The more you split it down the easier the job is todo in parralel aslong as you have the ability todo it in paralel. So running 32 you do the job of 64 in parralel, then again at 16 you do the job of 32 in parralel. So fp16 is basically hyperthreading of 32.

Well TBH,anything that will enable things to be done more efficiently in games is OK by me. I just feel the PC is hamstrung by trying to brute force everything.

 

[Shin1234]

Well TBH,anything that will enable things to be done more efficiently in games is OK by me. I just feel the PC is hamstrung by trying to brute force everything.

The pc is brute forcing everything, its that the games are built in such a way that brute force is the way todo cos its the easiest way. These things should stop it.

 

[pmc25]

Technically it seems many of these features are "enabled" but not actually working optimally or actually being utilised - some can only function in compatibility "deferred" modes without further developer support or need to be utilised by developers.

Those hoping for a miracle from these features are likely going to be sorely disappointed especially as the bigger gains will likely be power saving when a developer properly utilises them :s

Most of them are not exposed in public drivers. See discussion at Beyond3D.

 

[chrcoluk]

http://wccftech.com/radeon-rx-vega-64-msrp-pricing-update/

That is a much better statement, why they took so long I dont know.

One simple thing AMD could do is sell cards directly from their own website at MSRP.

There is still a free market in that situation.

 

[Shin1234]

That is a much better statement, why they took so long I dont know.

One simple thing AMD could do is sell cards directly from their own website at MSRP.

There is still a free market in that situation.

Well thats something, do amd make there own cards, even there ref cards are from custom board partners.

 

[Rroff]

Most of them are not exposed in public drivers. See discussion at Beyond3D.

I've seen some of those discussions its a ****** joke - you've got people from like pcper and anandtech, etc. saying AMD has 100% told them one thing in one thread and then 100% the opposite in another thread while saying something different again on their sites, while AMD's ISAs etc. are often showing something different again.

Fact is most of what people think they are talking about are actually technically "enabled", at least according to AMD's whitepapers - actual utilisation or accessibility is another matter and some stuff seems to be all behind closed doors working with select developers only and will probably be all like *tada* exposed when those products are launched. Its a really really bad approach regardless of intentions.