AMD VEGA confirmed for 2017 H1

[Gregster]

Correct, but it's going to take a long time to get there as the vast majority of graphics assets target 1080p.

It is not just a matter of having a card that can render 4K@60, it's also a matter of games shipping with high-quality textures suitable for 4K.

New game engines or even current game engine add ons can possibly make good use of faster memory but that goes beyond my knowledge base but it is clear that HBM and GDDR6 will be the future.

so Vegas only for the 4K brigade, bit of a small market that

Anyone on lower than 4K, only has the 480 to get from AMD, or has to go Nvidia for more performance, looks like its a 1080 for me then

Don't you start.... It is bad enough with some of the others

 

[LoadsaMoney]

Don't you start.... It is bad enough with some of the others

 

[shankly1985]

You guys are missing a massive key part of 4k gaming the Raja picked up on in the PC gamer interview.

4k gaming atm isnt true 4k all we doing it upping the Resolution! the textures etc etc are mostly all still designed with old resolution in mind even 1080p graphics are still used in games.

So once we start seeing true 4k gaming this will up the pixel processing and HBM might start being more of a use!
Its one of them things again for the Future! but am sure some on here will say its pointless because we not seeing gains atm
But remember we need to start from somewhere the industry wont move forward with out the hardware becoming available.

Raja explains it better

 

[Rroff]

You guys are missing a massive key part of 4k gaming the Raja picked up on in the PC gamer interview.

4k gaming atm isnt true 4k all we doing it upping the Resolution! the textures etc etc are mostly all still designed with old resolution in mind even 1080p graphics are still used in games.

So once we start seeing true 4k gaming this will up the pixel processing and HBM might start being more of a use!
Its one of them things again for the Future! but am sure some on here will say its pointless because we not seeing gains atm
But remember we need to start from somewhere the industry wont move forward with out the hardware becoming available.

Raja explains it better

Depends a bit - he was saying PC game assets are in their shipping form adequate to "resolve" at 1080p - which TBH is being a bit generous - a lot of games you start to get significant texel scaling beyond about 720p :S but there is a limit - if you had enough detail in the assets used to properly take advantage of all the pixels on the screen at 4K some of them would be pretty much ignored at say 1440p anyhow and have little impact on performance - though the point that there is room for improvement of the quality of the shipped assets of a game and that they will have a performance impact isn't something that should be ignored - I think we are still some way from a point where that would utilise HBM in a useful way though even if developers got onto it tomorrow.

 

[shankly1985]

Depends a bit - he was saying PC game assets are in their shipping form adequate to "resolve" at 1080p - which TBH is being a bit generous - a lot of games you start to get significant texel scaling beyond about 720p :S but there is a limit - if you had enough detail in the assets used to properly take advantage of all the pixels on the screen at 4K some of them would be pretty much ignored at say 1440p anyhow and have little impact on performance - though the point that there is room for improvement of the quality of the shipped assets of a game and that they will have a performance impact isn't something that should be ignored - I think we are still some way from a point where that would utilise HBM in a useful way though even if developers got onto it tomorrow.

Correct.
But without HBM within GPus on the market we will never move forward. Least Devs have a framework to now build on.
So think of it has AMD helping the industry for the future of graphics

Something AMD and the one of the big reasons i back AMD is because of all this approach to PC gaming
"Open Standards" Will always play a place for me its the reason I use Android anything Open is the right way too go.

Locking things down only hurts others!

 

[nashathedog]

It just chance, just becausr you weren't asked to re-take the survey doesn't mean there is any bias.

That's why I started my response with "It's not biased"

They only have to ask a few thousand users to get very solid statistical data, whether you have never bean asked or get asked multiple times makes no difference.

They ask new members plus a random few that seem to get asked more than once.

its like rolling a dice 6 times. Some users will never see the number 6, others might see it 6 times but the dice itself is truly random. Anecdotes of someone rolling a dice 4 times in a roll doesn't change the statistical outcomes.

From what I can see it's only a small number of participants that are redoing the survey while new members seem to make up the bulk (I don't know if we can we get a number for the average monthly join up rate?), If the vast majority of people asked are newcomers and the best selling card doing the rounds is selling 4 times better than the competition it stands to reason that the results will lean towards that card. At the end of the day without definite numbers for who's asked to do the survey the whole points moot as neither point of view can be proven or disproven. I'm not claiming it has anything to do with bias.

There's not much of an argument here. I'm not saying I know the divide between users but I do know that I wouldn't be using Steam for a source of enlightenment.

+1

 

[supersuperjamie]

I must be one of the random few then, I've done it three times in the past five years. And had a different cpu, gpu and monitor each time!

Which is a large part of why I'm hoping ryzen, am4 itx and Vega all hit the marks we all hope. Fed up of chopping and changing all the time.

 

[nashathedog]

Working on a base clock of 500mhz for HBM1 on the Fury cards most can overclock to 545mhz and if you are lucking you can get 600mhz.

HBM1 overclocks in steps, 545mhz 600mhz and I think the next one is 660mhz. If you try and run it at any other speed like 550mhz for example it will just default down to 545mhz.

?

I stick 550 on mine (when I remember) for benching and it seems fine, I haven't tested anything else I just tried that and it worked and my software reports it at 550, On 3dmark 560 seems to be a popular HBM overclock too. What makes you think it works in steps?

 

[harney]

https://www.youtube.com/watch?v=UuchUscHWSw&feature=em-uploademail

ah right guys

 

[D.P.]

From what I can see it's only a small number of participants that are redoing the survey while new members seem to make up the bulk (I don't know if we can we get a number for the average monthly join up rate?), If the vast majority of people asked are newcomers and the best selling card doing the rounds is selling 4 times better than the competition it stands to reason that the results will lean towards that card. At the end of the day without definite numbers for who's asked to do the survey the whole points moot as neither point of view can be proven or disproven. I'm not claiming it has anything to do with bias.

And how much insight and internal knowledge of steam surveys do you have that the public don't know? What secret information do you have available that no one else has access to?




The fact is the steam survey is an attempt to create an unbiased and accurate estimate of installed user base configurations that is widely accepted by industry and who's results highly correlate with other sources and information.


It is frankly ridiculous the people trying to disregard the steam survey results because of their personal anecdotal experiences and applying their zero knowledge of statistics and sampling theory.



The data is no better or worse than the JPR market share etc., which is simply an estimate of supply chain volume and not personal sales.

 

[shankly1985]

Let's just agree too disagree with survey. We have said are own stance lets just leave it has that and get back on to topic.

 

[D.P.]

Depends a bit - he was saying PC game assets are in their shipping form adequate to "resolve" at 1080p - which TBH is being a bit generous - a lot of games you start to get significant texel scaling beyond about 720p :S but there is a limit - if you had enough detail in the assets used to properly take advantage of all the pixels on the screen at 4K some of them would be pretty much ignored at say 1440p anyhow and have little impact on performance - though the point that there is room for improvement of the quality of the shipped assets of a game and that they will have a performance impact isn't something that should be ignored - I think we are still some way from a point where that would utilise HBM in a useful way though even if developers got onto it tomorrow.

An easy test to see if more bandwidth is useful is to overclock memory vs core and see the performance gains. Memory bandwidth is still not a limiting factor. Of course you can artificially make it so.


HBM2 and future iterations will be fantastic and a will be a great step forwards but as with most things, there is a time and place. Being a step ahead of technology doesn't always help you. The FuryX is a prime example, nvidia could create a faster, cheaper, lower power GPU without HBM.

Nvidia's approach with bringing HBM2 out in the professional compute market first makes a lot of sense. Volumes are lower, profit margins are far higher (like 100x higher), and compute is one area that loves bandwidth. Nvidia can learn a lot without the worries of create mainstream volume. If they destroy a $150 chip and $60 of memory it matters much less when it is sold for $2000-15,000 etc.


I do hope AMD can do something special with HBM2 but if it is something around 1080 performance +-10% then its really another waste and FuryX all over again where AMD get lower profit margins and lower sales due to higher prices and the need to wait in H12017 to get decent HBM supply.

 

[D.P.]

Let's just agree too disagree with survey. We have said are own stance lets just leave it has that and get back on to topic.

Agreed, i wont say anymore.

 

[beany_bot]

Adored TV just put up a video that explains something why he thinks Big Vega is only 10-15% faster than a 1080. (on current drivers). He says it's not good enough.

 

[Rroff]

I do hope AMD can do something special with HBM2 but if it is something around 1080 performance +-10% then its really another waste and FuryX all over again where AMD get lower profit margins and lower sales due to higher prices and the need to wait in H12017 to get decent HBM supply.

An interesting one to me is the development process - it could be a necessary expense to get to a future point. However there are companies now that will help you design and build a product on a custom interposer with HBM "relatively" inexpensively so I can't imagine it could be entirely down to that aspect.

 

[Kaapstad]

?

I stick 550 on mine (when I remember) for benching and it seems fine, I haven't tested anything else I just tried that and it worked and my software reports it at 550, On 3dmark 560 seems to be a popular HBM overclock too. What makes you think it works in steps?

You can set whatever you want but the memory will still run at the nearest step.

AMDMatt can give you more detail of how the above works.

Lots of people set 550 or 560 in their OC software but only actual get 545mhz in practice.

 

[Kaapstad]

Can you link that please, both the P6000 and P5000 both use GDDR5x.

Neither use HBM.

The P6000 also runs faster than the Titan Xp in games, albeit with just under 300 more CUDA cores. Then again an increase of 10fps is huge, and it has better frame times.

http://hothardware.com/reviews/nvidia-quadro-p6000-and-p5000-workstation-gpu-reviews?page=1

NVidia Tesla P100

Same core count as the Pascal Titan but using HBM2 instead of GDDR5X

https://www.overclockers.co.uk/pny-nvidia-tesla-p100-gpu-accelerator-16gb-hbm2-3584-cuda-core-gx-052-pn.html

 

[humbug]

https://www.youtube.com/watch?v=UuchUscHWSw&feature=em-uploademail

ah right guys

Its "Alreet Guys"

Anyway, this guy has been accused of being an AMD fanboy, especially around here, i don't think so...

Once again its just good solid info, GTX 1080+ 10% is fine by me, it means the cut down Vega will probably be GTX 1070 + 10% and not hugely expensive.

Bring it

 

[D.P.]

An interesting one to me is the development process - it could be a necessary expense to get to a future point. However there are companies now that will help you design and build a product on a custom interposer with HBM "relatively" inexpensively so I can't imagine it could be entirely down to that aspect.

That is the thing. They have already pushed FuryX out as a pipe cleaner and learning process. A second version just doesn't make sense. Of course if they have a really high end product with high margins and low volumes like Titan XP then it may make sense but none of the rumours seem to indicate that.

 

[N19h7m4r3]

NVidia Tesla P100

Same core count as the Pascal Titan but using HBM2 instead of GDDR5X

https://www.overclockers.co.uk/pny-nvidia-tesla-p100-gpu-accelerator-16gb-hbm2-3584-cuda-core-gx-052-pn.html

That's not a graphics card though, and cannot be bested in games or benchmarks such as Timespy against the Titan X even.

As it stands the P6000 Quadro that uses GDDR5X is faster than the Titan X, while only having a small amount of extra CUDA cores.

Would you then say if it used HBM2 it would be slower than it is right now? So then being closer to the Titan X; instead of being faster?