AMD VEGA confirmed for 2017 H1

[Rroff]

Put simply, it is a classic case of (GIGO) "Garbage In Garbage Out". Now where are the leaks of what these things can actually do? If you came and read the last couple of pages of this thread you might be mistaken for thinking that it wasn't actually about Vega.

I read the last two pages and I can't get back that time now, its passed and I learned nothing about what I came in here to look at.

Click the go to last post button - if it isn't about Vega there probably isn't anything new about Vega.

 

[Vince]

Click the go to last post button - if it isn't about Vega there probably isn't anything new about Vega.

Clearly some strong tactics are being deployed by a seasoned graphics section pro.

 

[DarrenM343]

Quite insulting and not needed here really. If you are referring to someone (seems that way), then why not address that person in a more respectful way?. If one person is misinformed, that doesn't mean the rest are.

Thing is, there's enthusiastic fans of most things in life. People can take it too far but that passion can also be used as a good thing. What you're on about Darren is a personality flaw in some people which has absolutely nothing to do with any product. What is important is that we stop the "we/them" mentality. Both AMD and Nvidia create amazing cards but without either of the two, it's a progressional and financial nightmare. Look at Intel.

I did change my wording after my initial early morning post as I realised it might not have come across well. It wasn't targeted at anyone individually by the way as wouldn't do that anyway, just a general observation made but was also meant to be more tongue in cheek than how it came out . Anyway, I did change the wording before I thought anyone had replied .

Agree by the way and looking forward to seeing some competition in both areas (GPU, Intel).

 

[LoadsaMoney]

AMD’s Vega Will Double Your Usable Graphics Memory Capacity By Cutting Wasteful Memory Allocations

http://wccftech.com/amds-vega-doubles-usable-graphics-memory/

 

[TNA]

Only because a top on the line AMD card only being GTX 1080'ish means no cheaper than Pascal-Titan-X GTX 1080TI....

You wouldn't buy AMD if it was more Pascal-Titan-X than 1080.

Those who would buy one are actually quite happy as long as the price it right.

Bit confused by the post above, you saying I would not buy an AMD GPU?

What I want is AMD to bring out a card that is worth me upgrading my 1070 over and charge a reasonable amount for it like they usually do.

 

[flopper]

AMD’s Vega Will Double Your Usable Graphics Memory Capacity By Cutting Wasteful Memory Allocations

http://wccftech.com/amds-vega-doubles-usable-graphics-memory/

Raja Koduri – Chief Architect Radeon Technologies Group, AMD
With regards to the High Bandwidth Cache from a gaming perspective. We looked at all the modern games, the big games that push memory hard, and one of the things we noticed is the VRAM – graphics memory – utilization. We look at how much of the VRAM that the game allocates. So if the game say needs 4GB of memory when we looked at actually how much of that memory is actually used to render pixels we found that many games, actually most games, don’t use more than 50% of what they allocate.

That’s because the current/old GPU architecture doesn’t give you flexibility to move memory in fine granularity. So with Vega and with the High Bandwidth Cache and the HBC controller, for games it will utilize the amount of frame-buffer you have much more efficiently. So effectively you can think of it as Vega will be doubling your memory capacity for games.

Brad Chacos – Senior Editor,PC World
So basically a game that says it uses 4GB of VRAM right now, is in actuality using 2 and with Vega, you’re saying, it will actually allocate 2.

Raja Koduri – Chief Architect Radeon Technologies Group, AMD
Exactly

Anyone on a card with GDDR5x or such uses old tech now...

 

[old gamer]

AMD’s Vega Will Double Your Usable Graphics Memory Capacity By Cutting Wasteful Memory Allocations

http://wccftech.com/amds-vega-doubles-usable-graphics-memory/

I keep getting excited when there appears to be some news, only to be deflated when I relise it from wccftech and probably not worth the Internet bandwidth it was posted on

 

[FoxEye]

Anyone on a card with GDDR5x or such uses old tech now...

Listen I'm not nVidia fanboy, but I simply don't take any AMD claims as gospel. I'm pretty sure Raja would say Vega will increase the size of your girth if it would sell more cards. For balance, so would Jen-Hsun.

 

[Rroff]

Anyone on a card with GDDR5x or such uses old tech now...

GDDR5(X) has been old tech for awhile now - still does the job.

AMD’s Vega Will Double Your Usable Graphics Memory Capacity By Cutting Wasteful Memory Allocations

http://wccftech.com/amds-vega-doubles-usable-graphics-memory/

Problem with this is it takes a lot of effort to stay on top of - i.e. typically with id engines they "pre-cache" or "pre-touch" a lot of resources to prevent stutter, etc. later on when they are used for the first time - without carnal knowledge of that usage you can easily degrade the experience if you indiscriminately move that data out to a virtualised memory pool - so either you risk compatibility problems or you have the issue that until a per-application profile is created games won't see the VRAM saving.

To be fair with Fiji cards they pulled off much more aggressive memory management a lot better than I had expected but you can see some instances of stutter, etc. if you go looking for it but going to the extent of saving half your VRAM use is another matter entirely.

 

[LegendaryDread]

"For example say a game is built for 4GB and say you have a 4GB card it all plays well but when you swap in, for example you alt-tab out of the game and go into a browser or something or do something quick and you come back, it takes a long time."

I'm lost for words.

 

[J.D]

I did change my wording after my initial early morning post as I realised it might not have come across well. It wasn't targeted at anyone individually by the way as wouldn't do that anyway, just a general observation made but was also meant to be more tongue in cheek than how it came out . Anyway, I did change the wording before I thought anyone had replied .

Agree by the way and looking forward to seeing some competition in both areas (GPU, Intel).

That's the horrible thing with text, it's easily misinterpreted. All good and thank you for explaining .

Yes, it is exciting times but in retrospect it's quite sad that we've waited so long for to be excited like this (competition stirring up the market).

 

[triss]

Anyone on a card with GDDR5x or such uses old tech now...

And anyone with HBM is using old tech.. oh and they are working on HBM3 so HBM2 is old tech thats how the cycle goes.

Latest GDDR or latest HBM form are both current tech really though . Pro's and cons of both and i hope they both carry on to progress

 

[Gregster]

Oddly enough, anything computer based is old tech after it has been out 6 months. It is and always will be the nature of PC building.

 

[Apache14]

Problem with this is it takes a lot of effort to stay on top of - i.e. typically with id engines they "pre-cache" or "pre-touch" a lot of resources to prevent stutter, etc. later on when they are used for the first time - without carnal knowledge of that usage you can easily degrade the experience if you indiscriminately move that data out to a virtualised memory pool - so either you risk compatibility problems or you have the issue that until a per-application profile is created games won't see the VRAM saving.

To be fair with Fiji cards they pulled off much more aggressive memory management a lot better than I had expected but you can see some instances of stutter, etc. if you go looking for it but going to the extent of saving half your VRAM use is another matter entirely.

I agree however even application profiles to "whitelist" data for exclusion would be useless as an application update may change how that application allocates memory objects.

I hope they are being cleaver and are somehow detecting things like textures / meta data on allocation and discarding the non needed data. Unfortunately detecting these data 'types' is difficult in practice. That being said I'm not an expert in GFX memory subsystems.

Here is hoping they have some magic that's beyond me

 

[nashathedog]

AMD’s Vega Will Double Your Usable Graphics Memory Capacity By Cutting Wasteful Memory Allocations

http://wccftech.com/amds-vega-doubles-usable-graphics-memory/

Haven't we heard this before?

"4gb's of HBM is the same as 6gb's of gddr5"

Er No!

 

[shankly1985]

It's all to do with the new memory cache and controller.

 

[Lokken86]

Haven't we heard this before?

"4gb's of HBM is the same as 6gb's of gddr5"

Er No!

Actually 4Gb of HBM on the Fury's is roughly equivalent to 5.5gb on a DDR5 card in terms of how much memory is used.

 

[MarcM]

I read this quite often but haven't gone through it for a few days any ideas mentioned above of an actual date of availability?

I really need a new graphics card I have overhauled my system entirely over the past few months and gotten right back into technology and pc gaming again except the newer games wont play on my AGED 8800gtx. Can only play wow and CS:GO atm. Was waiting for this to drop but considering a 480 from overclockers to tide me over unless someone has a release date?

 

[Rroff]

No one has a date - if its based on the revised 14nm process at GF then probably a minimum of 2 months away - as they have working samples most likely March - June.

 

[drunkenmaster]

Haven't we heard this before?

"4gb's of HBM is the same as 6gb's of gddr5"

Er No!

It's all to do with the new memory cache and controller.

Actually 4Gb of HBM on the Fury's is roughly equivalent to 5.5gb on a DDR5 card in terms of how much memory is used.

There is NOTHING special in the memory controller or HBM, 4GB is 4GB.

AMD are saying, here is a chart of how much memory is allocated throughout a game on a graph, that line below it is how much is actually used. Game devs are idiots and put no thought into what memory is allocated vs used, almost no optimisation at all. For most games it's a case of, ask system how much video memory there is, set to do garbage collection when < 500MB left or something, nothing more or less, the rest just sits in memory for no reason. With no sense put on which is allocated first if the game appears to need more memory than the card has for the given settings.

AMD are saying, much more aggressive unloading of memory will drastically cut down memory allocated without any real change in memory actually used. AMD is already significantly more memory efficient than Nvidia in this regards from the drivers side and has been for as long as I can remember now.

Watch Dogs 2 benchmarks on gamegpu (website is struggling to load for me but the images are available if you search watch dogs 2 gamegpu on google), a RX480 and a 1080 shows at 4k 7.5GB use for the 1080, 6.5GB for the RX480, 6.5 vs 5.5GB for 1440p and 6.3 vs 5.2GB for 1080p.

This is frequently seen in I'd say by far the majority of games. I think I've seen one ever where Nvidia had lower memory usage, a decent amount where it's similar and dozens/hundred where AMD is using significantly less memory.


HBM is no more efficient, and loading data isn't suddenly better because there is more bandwidth. If a request for data you need comes in and it's not in memory the request goes off the gpu, across the same buses, the same cpu, the same sata connection the same access time and finally gets back to the GPU maybe 20ms later and then it can be processed... loading into memory and then being used is a fraction of the time of actually going across the system to get it.

Comparing Fury to a 980ti and deciding HBM is more efficient because it's using less flies in the face of frankly a decade of evidence that AMD uses noticeably less memory. As shown with two gddr using cards, AMD is still using miles less memory, nothing to do with HBM. A byte is a byte is a byte.