AMD VEGA confirmed for 2017 H1

[nashathedog]

Not competing in the top end and not having a halo/flagship product is a sure fire way of continuing this trend though!

Hopefully that's all going to change this year, AMD's had their lean years, struggled through them and survived, Now comes the pay off, stock value is rising and they've got new product coming out in both the cpu & gpu markets, If the dev's have done well they'll have solid products with which show them in a positive light and going forward they can put their research budget into fine tuning and perfecting what they have. I just hope they have the product we all hope they have.

 

[ljt]

that's a very unrealistic expectation, based on the price of the 480 cards.

Well yes I agree that it is now the pound has tanked so much. But really the 8gb rx480 is a £200 card before brexit took its toll. So £270 isn't or wasn't at the time an unrealistic expectation.

However if the Vega family does indeed end up being released summer time it will be near enough 12 months after the gtx1070. So I would hope there is a card that gives 1070 performance for around the £270 price point by that time!

 

[akarypid]

Yep, they are managing fine still on good old GDDR5.

..not for much longer though. In SC15 their presentation dedicated a fair amount of time on Volta's memory bandwidth stated to be 1.2TB/s using HBM2.

In fact, they're looking beyond that, as they recon that the problem is really that memory access speed is not keeping up with GPU speed. They're already talking about next-gen memory (as in after HBM2).

 

[JediFragger]

Hopefully that's all going to change this year, AMD's had their lean years, struggled through them and survived, Now comes the pay off, stock value is rising and they've got new product coming out in both the cpu & gpu markets, If the dev's have done well they'll have solid products with which show them in a positive light and going forward they can put their research budget into fine tuning and perfecting what they have. I just hope they have the product we all hope they have.

AMD being AMD, they can still snatch defeat from the jaws of victory

 

[Firestar_3x]

AMD being AMD, they can still snatch defeat from the jaws of victory

I'm not holding out much hope tbh.

 

[Griffildur]

..not for much longer though. In SC15 their presentation dedicated a fair amount of time on Volta's memory bandwidth stated to be 1.2TB/s using HBM2.

In fact, they're looking beyond that, as they recon that the problem is really that memory access speed is not keeping up with GPU speed. They're already talking about next-gen memory (as in after HBM2).

Apparently Samsung is working on HBM3 already.

http://arstechnica.com/gadgets/2016/08/hbm3-details-price-bandwidth/

 

[TheRealDeal]

Apparently Samsung is working on HBM3 already.

http://arstechnica.com/gadgets/2016/08/hbm3-details-price-bandwidth/

Amd Navi on AMD's roadmap states Next Gen memory which suggests something different to hbm 2. Technology never stands still although with graphics it's certainly not moving as fast as we would like.

 

[Apache14]

Amd Navi on AMD's roadmap states Next Gen memory which suggests something different to hbm 2. Technology never stands still although with graphics it's certainly not moving as fast as we would like.

I was thinking by "next gen memory" they where indicating having some large storage (like ssd) on card to support the "normal" vram (be that gddr or hbm). I am pretty sure they have demoed something like that on one of their radeon pro cards (M.2 slots)

 

[Rroff]

I was thinking by "next gen memory" they where indicating having some large storage (like ssd) on card to support the "normal" vram (be that gddr or hbm). I am pretty sure they have demoed something like that on one of their radeon pro cards (M.2 slots)

Might just be for debugging - there have been engineering boards with storage devices mounted on and all sorts pictured before.

 

[akarypid]

I was thinking by "next gen memory" they where indicating having some large storage (like ssd) on card to support the "normal" vram (be that gddr or hbm). I am pretty sure they have demoed something like that on one of their radeon pro cards (M.2 slots)

Yes they did and they call it Solid Stage Graphics (SSG), but I think that has nothing to do with the so called 'next-gen' memory.

The problem there is bandwidth and how GDDR and HBM will reach their efficiency limits (albeit HBM will do that at higher data rates than GDDR).

Vega has an interesting memory controller that can manage 512TB of virtual address space, which makes it possible to integrate with system RAM, add-on SSDs, or even the network (!) storage via a unified memory model. They're actually calling the VRAM on Vega a memory cache. This all fits in with their ROCm toolchain that seamlessly manages memory using common pointers and their aim of building an HSA solution.

On one hand Vega will pack interesting tech on the memory front, on the other hand this all costs (in die space and efficiency) just lie ACEs do. So there is a handicap there.

In the end though, if they manage to get to Navi through this avenue (supposedly using multiple GPU chips on the same board working seamlessly as a single GPU) then it'll be worth it I guess. Once we hit 7nm we'll stay there for a while. We can't expect architectural improvements year-on-year so multi-chip is the apparent way to scale...

EDIT: If we're lucky, maybe dual-chip Vega will be a 'testbed' for Navi with just 2 chips that can work individually, OR combined as one. There's interesting stuff on the software side there: when the 2 chips are working as one, they may be using alternate frame rendering (which is common SLI/CrossFire) but I can't see how that would be 'transparent', OR they may be rendering different portions of the screen, as they're using the common memory controller to get to the VRAM that operates as a 'cache' and takes care of coherency issues. In the latter case whichever chip finishes rendering last (the two halves of the screen can have different complexity) would be responsible for signalling the ROP to dish out the frame and apply the adaptive sync protocol messages.

 

[lordfletch]

And even for the stragglers Nvidia can cut prices substantially and cut margins to nothing if need be. AMD won't win a price war being late, especially if Vega requires HBM2 to be competitive.

cut margins to nothing , you do realise they have Institutional Share Holders ?
that's business suicide.AMDs top cards wont be cheap,they cant be.

 

[triss]

Yes they did and they call it Solid Stage Graphics (SSG), but I think that has nothing to do with the so called 'next-gen' memory.

The problem there is bandwidth and how GDDR and HBM will reach their efficiency limits (albeit HBM will do that at higher data rates than GDDR).

Vega has an interesting memory controller that can manage 512TB of virtual address space, which makes it possible to integrate with system RAM, add-on SSDs, or even the network (!) storage via a unified memory model. They're actually calling the VRAM on Vega a memory cache. This all fits in with their ROCm toolchain that seamlessly manages memory using common pointers and their aim of building an HSA solution.

On one hand Vega will pack interesting tech on the memory front, on the other hand this all costs (in die space and efficiency) just lie ACEs do. So there is a handicap there.

In the end though, if they manage to get to Navi through this avenue (supposedly using multiple GPU chips on the same board working seamlessly as a single GPU) then it'll be worth it I guess. Once we hit 7nm we'll stay there for a while. We can't expect architectural improvements year-on-year so multi-chip is the apparent way to scale...

EDIT: If we're lucky, maybe dual-chip Vega will be a 'testbed' for Navi with just 2 chips that can work individually, OR combined as one. There's interesting stuff on the software side there: when the 2 chips are working as one, they may be using alternate frame rendering (which is common SLI/CrossFire) but I can't see how that would be 'transparent', OR they may be rendering different portions of the screen, as they're using the common memory controller to get to the VRAM that operates as a 'cache' and takes care of coherency issues. In the latter case whichever chip finishes rendering last (the two halves of the screen can have different complexity) would be responsible for signalling the ROP to dish out the frame and apply the adaptive sync protocol messages.

Didnt the old gpu's do that to a lessor extent aswell ?IIRC my old gtx 8200 or something had memory shared with Ram to give it more of a buffer. Still accessing that slow not on gpu memory had disadvantages i thought?

 

[TNA]

So much salt in this thread!

+1

I want Vega to do well so unltimatly I can get better price for performance. Will probably get a new monitor in 2017, so going Freesync 2 alone is a huge saving if AMD deliver

am I the only one who finds that ad really poor taste? and yes I would say the same if it was used by NVidia.

I think you are in the minority on this one yes.

What nvidia did only a few years ago with the 970 was what I would call poor taste. If you managed like me to get over that, then you will be fine

 

[tommybhoy]

Be going Nvidia when i upgrade.

Think your having troubles now.

 

[Doom112]

Some AMD ** Please do not start using that term - EVH ** are still in a denial that even on Vulkan VEGA is only on par with GTX 1080. It is really a matter of concern for AMD when VEGA performs in 95% of games ,which are mostly based on DX11.

This is exactly what happens when you are short of R&D, customers and logic. You are launching a product nearly after 1 year and still it is on par against competitions second best card.

 

[shankly1985]

Some AMD ** Please do not start using that term - EVH ** are still in a denial that even on Vulkan VEGA is only on par with GTX 1080. It is really a matter of concern for AMD when VEGA performs in 95% of games ,which are mostly based on DX11.

This is exactly what happens when you are short of R&D, customers and logic. You are launching a product nearly after 1 year and still it is on par against competitions second best card.

If AMD was really short on R&D how can they keep bring new tech into the industry? LOL Stop talking Rubbish Doom! None of us on here know AMDs R&D

If they was super "Doomed" like most on here keep saying dont you think they would just go the cheap/quick root of releasing GPUs?
Why would they spend the extra time adding HBM 1/2 New Memory tech cache
Pushing Low APIs the list goes on and on!

Fair enough Vulkan is only in one game! But more hopefully will follow this year, plus DX12 should start dong well this year, DX11 isn't all bad either if we go from the work AMD has done with the 480 Driver and Hardware level too impressive performance just look at recent benchmarks alone 480 vs 1060
If the 480 is doing this good against the 1060 what with VEGA do to the 1080??

AMD dont need too take on Nvidia, they can go at allthis alone, AMDs recent share price is up $11.30 speaks for itself!

 

[Panos]

Some AMD ** Please do not start using that term - EVH ** are still in a denial that even on Vulkan VEGA is only on par with GTX 1080. It is really a matter of concern for AMD when VEGA performs in 95% of games ,which are mostly based on DX11.

This is exactly what happens when you are short of R&D, customers and logic. You are launching a product nearly after 1 year and still it is on par against competitions second best card.

To early for that-schizo

 

[Diagro]

Some AMD ** Please do not start using that term - EVH ** are still in a denial that even on Vulkan VEGA is only on par with GTX 1080. It is really a matter of concern for AMD when VEGA performs in 95% of games ,which are mostly based on DX11.

This is exactly what happens when you are short of R&D, customers and logic. You are launching a product nearly after 1 year and still it is on par against competitions second best card.

Way to early to start spouting this nonsense.

Seriously why troll?

 

[KillBoY_UK]

Way to early to start spouting this nonsense.

Seriously why troll?

But AMD R&D spending is about 69% of what it was just 5 years ago, $375m vs $259m with a low point of $229m while Nvidia and Intel's R&D budgets have been increasing this is all per quarter btw. So not quite nonsense.

That being said AMD made the money go far but a smaller budget can only so so much.

 

[N19h7m4r3]

I absolutely love that he's now calling people fanboys, and claiming with certainty that Vega is only on par with the GTX 1080.

I didn't know he had a completed retail Vega card in his possession.

If not falling for his trolling, and being nothing but doom and gloom about a A0 engineering sample makes me an AMD fanboy; then I'll happily be one, and wait for actual information and reviews of a finished product.