AMD VEGA confirmed for 2017 H1

[FoxEye]

If it's a real game-changer, then factoring in everything that's happened, you can see £600 - £800 being the launch price for big Vega.

Hopefully not more

 

[eddyr]

, but each generation has to be in the same time frame, surely? If AMD beat the pascal 1080 or even TX in July that's beating technology that's already been around for a year. That's not exactly winning anything? One could argue they're almost a generation behind.

You are always beating tech that's been out for a 1 year, 2, 3 with every consecutive generational release. There is also no generational lock step with competitors in that way.

Worth noting that what is chosen to be released is not necessarily an indication of the limit of a generation of tech. Had AMD chosen years back to build a bigger Polaris GPU, say 350mm2 (RX480 is ~230mm2) then that would have been a 1080 competing/beating chip. (1080 is around 320mm2). Polaris still had 100+ watts to play with. They chose to save all that upfront development cost (questionable whether it would even generate a positive return depending on manufacture costs/yields/sales particularly with Vega due to surplant that performance bracket so soon) and focus resources on future products.

 

[JediFragger]

Anandtech write-up on uarch changes:

http://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser

They say it is the biggest uarch change for AMD graphics since GCN 1.0 and looking at some pictures of the GPU it looks like it is over 500mm2. It looks more like AMD is targeting Titan X level performance I suspect.

First time that Anandtech have had a gfx article up on time since Anand left

 

[Gerard]

Quite, but it is easy to see how people feel mislead I for one was hoping to learn something new. See some FACTS so to speak.

But you got a fact, the chip exists, and has no wood screws

 

[Steampunk]

I think this shows Polaris/Fury to be pipe cleaner and proof-of-concept products. Market-wise, Polaris aimed for low-hanging fruit, while waiting for the proper 14nm on a more mature process, with HBM, interposer, etc. AMD probably decided that there's no point in building a high end version of a stop-gap product.

 

[Killer7xx]

Anandtech write-up on uarch changes:

http://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser

They say it is the biggest uarch change for AMD graphics since GCN 1.0 and looking at some pictures of the GPU it looks like it is over 500mm2. It looks more like AMD is targeting Titan X level performance I suspect.

A Vega chip with that die size should be over Titan X level performance. The changes sound like AMD have increased IPC and efficiency by a lot compared to previous revisions of GCN. On paper at least, NCU seems superior to Pascal, it's like a super GCN.

 

[Davedree]

I think this shows Polaris/Fury to be pipe cleaner and proof-of-concept products. Market-wise, Polaris aimed for low-hanging fruit, while waiting for the proper 14nm on a more mature process, with HBM, interposer, etc. AMD probably decided that there's no point in building a high end version of a stop-gap product.

Agreed but amd failed to get into mobile and left a massive market for nvidia to imflate prices on. Polaris was an execution that was late to replace their ancient hawaii chips and compete with 2 year old gm204 maxwell. Polaris would have been around if the 20nm process was good enough. However Nvidia are still untouchable with gp106 and gp104 in that they can sell the dies on desktop and mobile.

Fiji was just gcn 1.2 Tonga scaled up by x2 with a prototype hbm, and a lot that was wrong with fury will be straightened out for vega aka fury2.

Vega from the slides is looking very attractive, but i'll hold judgement when it appears.

 

[JediFragger]

AMD have been ****-poor for too long (esp on mobile, it's embarrassing!!). Hoping that RTG can start to turn that around from this Summer

 

[Davedree]

Well polaris was meant to do that, amd were banging the ppw drums,
But look at the lineup of current mobile gpu's it's a disgrace.

 

[Panos]

A Vega chip with that die size should be over Titan X level performance. The changes sound like AMD have increased IPC and efficiency by a lot compared to previous revisions of GCN. On paper at least, NCU seems superior to Pascal, it's like a super GCN.

Actually if you take the tech for what it is, Pascal is just a shrink of Maxwel, with higher clocks.
Vega is a brand new chip designed from scratch with new GCN, bells & whistles. As is Volta, but that's delayed for 2018.

So it wouldn't surprise me Vega is faster than TXP which is few years old tech at it's core. I will be surprised if it is not.

 

[vapor matt]

Vega might exist as a chip but its no where near production or even available for vendors. late Q2 or Q3 we might see it. When we see hardware we can buy in reviews I will again take notice. the hype train hit the buffer's today with AMD. yes potentially they have a very good chip but the speculation of its performance is nothing with out hardware to back it up.


By the time vega hits the shelves to purchase it will be volta its against not pascal. A custom AMD card with a Vega GPU beating a 1080 production card is not a fair comparison. production card V production card is, and that's what everyone wants to see, but not in 6 months time.

I don't back any side red or green, but paper launches and some custom Vega chipped cards is not what's been hyped up for the last 3 months. I thought we were going to see vega production hardware on display.


AMD hardware to purchase NOW is what opens the wallets, past history should tell you this.

 

[JediFragger]

Actually if you take the tech for what it is, Pascal is just a shrink of Maxwel, with higher clocks.

Tweaked for better DX12 performance too.

Am looking forward to pairing Vega with my 38UC99


edit - My only concern is that the Doom benchmarks for Vega that are floating around only points towards a 40%ish performance increase over FuryX. Hopefully they're v immature drivers.

 

[LoadsaMoney]

edit - My only concern is that the Doom benchmarks for Vega that are floating around only points towards a 40%ish performance increase over FuryX. Hopefully they're v immature drivers.

and to match the 1080, Vega had to be running under Vulkan

 

[JediFragger]

and to match the 1080, Vega had to be running under Vulkan

Yeah, tbf their OGL drivers are/were a bit potato! Vulkan just evens the playing field with people who can actually write OGL drivers

 

[Killer7xx]

Tweaked for better DX12 performance too.

Am looking forward to pairing Vega with my 38UC99


edit - My only concern is that the Doom benchmarks for Vega that are floating around only points towards a 40%ish performance increase over FuryX. Hopefully they're v immature drivers.

They're not immature drivers, they were Fiji drivers with extra debugging layers.

 

[LoadsaMoney]

Vega' as fast as the 1080 in a plethora of games, from DOOM, to DOOM, and not forgetting, DOOM.

 

[Yaayuh!]

Vega' as fast as the 1080 in a plethora of games, from DOOM, to DOOM, and not forgetting, DOOM.

Don't forget Doom.

 

[Ian Evey]

Vega' as fast as the 1080 in a plethora of games, from DOOM, to DOOM, and not forgetting, DOOM.

And by the time we see vega don't forget Doom 2

 

[DarrenM343]

You are always beating tech that's been out for a 1 year, 2, 3 with every consecutive generational release. There is also no generational lock step with competitors in that way.

Worth noting that what is chosen to be released is not necessarily an indication of the limit of a generation of tech. Had AMD chosen years back to build a bigger Polaris GPU, say 350mm2 (RX480 is ~230mm2) then that would have been a 1080 competing/beating chip. (1080 is around 320mm2). Polaris still had 100+ watts to play with. They chose to save all that upfront development cost (questionable whether it would even generate a positive return depending on manufacture costs/yields/sales particularly with Vega due to surplant that performance bracket so soon) and focus resources on future products.

That's kind of my point, but beating a competitor 1 year after releasing a product is not the same as beating the competitor the same day of release. Some don't seem to understand that and seem to want to call a winner regardless - even if they're ahead for a short time only before the competitor they might have only just beaten release their new range.
Anyway, time will tell. I'd like to see where we sit one year from now. I know the answer to that as R&D £££ makes a difference but will be good to see regardless. Will AMD move away from their past? That's what they need to do but already seemingly delayed and no top end products for a long time which is not good.

Of course AMD and Nvidia are in dfferent cycles it seems so it's kind of hard to compare in a way but if you had to declare a winner of this gen this moment then that goes firmly to Nvidia and nobody can deny that - been ahead for a very long time.AMD might even jump ahead when they do release but it probably won't be for long. The staggered cycles means they could have course release something faster before NVidia again launch their next range. Maybe we should judge but who has the performance crown the longest ! .

I like to think AMD will be able to join the party and the new architecture will work well but we need to be also aware R&D costs money and those with the deeper pockets will likely develop better products....

 

[LoadsaMoney]

Don't forget Doom.

Yup, forgot that one, oops.