AMD VEGA confirmed for 2017 H1

[Seanspeed]

If a 490 does release in December it wouldn't surprise me if it was a dual Polaris chip and they market it as being specifically for VR. Then it would be priced so that it looks like a bargain compared to the Radeon Pro and don't forget AMD "wouldn't suggest you buy it for every day gaming" but many would buy it just to regret it.
It depends how forward thinking the Radeon group is, It may make them a few bucks in the short term but it would once again damage the brand name by focusing on mgpu which has a negative effect on the brand due to how unreliable it is.

Think about it, A few short months back loads of people were claiming DX12 was going to turn things around for mgpu yet it has had the opposite effect so far, It may eventually improve things but it's going to take years for that to become the case so for now another mgpu card will not do AMD any favours..

AMD wont be so stupid as to name a dual-GPU card as a mainstream lineup option. I know a lot of people dont have faith in their marketing skills, and there's a bit of merit to that, but this would be a step too far.

The time for doing such a thing has passed anyways unless Vega is like 8+ months away or something, which I doubt.

Pretty sure if there is a 490 in December it will be dual Polaris 10 chips on an interposer or similar structure. No secret that this is what AMD have intended to do for ages. Multi GPU but without the driver / engine headaches.

But I doubt we'll see anything before January, and this would be sooner than anyone expected the multiGPU, but singleGPU as far as the OS is concerned, product. Refreshed VEGA, NAVI or XB+ were the expected debut.

For one, how is this 'no secret'? I'd say it's *very* secret. Seems the only evidence we have of it is some Youtube ranter with some incredibly outlandish ideas. That seems to be what most people who say this base that off of. Other than that, there's been absolutely no solid rumors or suggestions this will happen.

I know you're going to talk about how stacked memory is some indicator this is where GPU's are going, but GPU's are not memory chips. It's a whole new world of complication doing what you're suggesting. And it would *not* suddenly solve any multi-GPU issues whatsoever. It would probably create its own set of unique challenges/headaches on top of many existing multi-GPU issues.

I'm still betting hard that 490 is going to be baby Vega and that we'll see a new 'named' top line chip for the top Vega GPU. Dont know if it'll be called Fury or whatever, but something outside the numbering scheme. I think strong evidence for this is that we've seen an official AMD slide where they planned to refresh Polaris with a revision '475/485/etc', meaning the 400 series will stick around for a bit.

 

[Rroff]

For one, how is this 'no secret'? I'd say it's *very* secret. Seems the only evidence we have of it is some Youtube ranter with some incredibly outlandish ideas. That seems to be what most people who say this base that off of. Other than that, there's been absolutely no solid rumors or suggestions this will happen.

The main engineer that was working on hardware level multi GPU solutions was made redundant so I can't imagine it is a direction they are still pursuing but who knows - it is certainly not something that was designed into Polaris from the ground up which it would really need to be - you can't really start rejigging your memory interface, etc. like would be required mid production.

EDIT: As an aside GF supposedly now has the revised performance fixes for 14LPP live so it wouldn't surprise me if we see a revision of Polaris released (I can only go on a random forum post by a supposed employee for that though).

 

[CAT-THE-FIFTH]

It will be interesting to see what happens once Zen is released - apparently most of their R and D is into Zen,so once that is released it should hopefully release more funds for GPU development.

 

[LoadsaMoney]

Couldn't agree anymore to what you've said.
Regarding 980 Ti owners 'massively downgrading' to a 480 if anyone has done that it'd be a stopgap I was about to sell my Ti and pick up a 780 non Ti for £80 though ended up keeping the Ti and buying another one because sod it I just hope upcoming resident evil 7 works well at 4K and dead rising 4 though I'm waiting for it to come out on steam which is 90 days from 6th Dec so 1080 Ti will be out by then

Fixed

 

[Seanspeed]

The main engineer that was working on hardware level multi GPU solutions was made redundant so I can't imagine it is a direction they are still pursuing but who knows

I think AMD are going to try and bet on DX12/Vulkan for multi-GPU support going forward. I think it's a losing bet, but it's a small enough market that it wont make much difference, at least.

It will be interesting to see what happens once Zen is released - apparently most of their R and D is into Zen,so once that is released it should hopefully release more funds for GPU development.

AMD cant just release Zen and then sit back and smoke a cigar. They'll need to keep investment into development or else they'll just fall behind very quickly again with how relentless Intel is on this front.

 

[Rroff]

I think AMD are going to try and bet on DX12/Vulkan for multi-GPU support going forward. I think it's a losing bet, but it's a small enough market that it wont make much difference, at least.

I can just see it TBH - jumping on that 2 years prematurely - maybe even be instrumental in pushing it into prime time use but then when it is widely adopted and actually practically useful to engine developers AMD's hardware solution for that generation will be nowhere to be seen while nVidia meanwhile capitalise on the new tech.

 

[Griffildur]

Perhaps keep it on topic? Is it hard? The premise I mean.

If you want to worship at the altar of Jen, do it somewhere else. We don't flood your threads with nVidia cards catching fire spam, after all.

If you must respond do it on - topic, once it's released then start comparing it. Anything else is just a PITA, but up to you how you behave until a mod decides otherwise.

you might want to check who is actually starting these side discussions before you go ballistic. you might be surprised me thinks ...

 

[CAT-THE-FIFTH]

AMD cant just release Zen and then sit back and smoke a cigar. They'll need to keep investment into development or else they'll just fall behind very quickly again with how relentless Intel is on this front.

The initial R and D push will cost them more - the next improvements will minor compared to pushing out a whole brand new core.

 

[Seanspeed]

The initial R and D push will cost them more - the next improvements will minor compared to pushing out a whole brand new core.

Tell Intel that. Their R&D budget would suggest otherwise.

 

[nightrider1470]

Are these cards out Jan?

 

[CAT-THE-FIFTH]

Tell Intel that. Their R&D budget would suggest otherwise.

Most of their R and D budget is orientated around process node development and their graphics R and D too. AMD graphics R and D is already covered by their dGPU R and D.

Remember,has managed to go from K10 to BD to PD to SR to Exc and Exc mk2 in the last five years. They also developed Bobcat,Jaguar and Puma which were not far off the Intel Atom chips of the day(except Intel had better process tech). The major costs were going from K10 to BD and developing the Bobcat core. The main issue is BD sucked so was a poor starting point. But look at the Athlon and Athlon 64 - models after that had iterative improvements.

The vast majority of AMD R and D is being spent on developing Zen from the ground up - once they have the basic building blocks for Zen,it will be much cheaper iterative improvements like what Intel has done for years. I imagine the testing required for a totally new uarch must cost a lot of money,since most of the errata will never have been documented before.

With less R and D spend on the CPU side,it will be refocused towards dGPU. If you look at the last few generations of GCN - they have been iterative improvements. Nvidia had one major uarch change with Maxwell. According to certain people on Beyond3D,Vega is a much bigger change than from GC2 to GCN3,or GCN3 to GCN4,but like with Volta I expect Navi in 2018 will be where the major uarch changes are.

Nvidia seems to be moving towards tick-tock. So,a more conservative change at first on a new node,followed by a bigger change when the node is more cost effective. I suspect we might see the same with AMD.

 

[kration]

Are these cards out Jan?

I hope something happens soon as I'm looking to replace my 290 and I haven't got the faintest idea about what I'm going to replace it with and when...

 

[humbug]

Tell Intel that. Their R&D budget would suggest otherwise.

Well, means nothing, Intel may well have spent $100bn or whatever it is over 10 years to get where they are with CPU's today and AMD may well do it in this one generation for £1.50.

 

[AthlonXP1800]

http://www.fudzilla.com/news/graphics/42233-vega-10-launching-in-1h-2017

Look like we may have to wait a long time until June for Vega 10 due to issues with HBM2 memory delays, shortages and high costs. Vega 10 will have similar performance as Radeon Pro Duo.

Guess RTG messed up twice again that would see GTX 1080 Ti have no competitor for 6 months. Look like 2017 will be very good year for Nvidia again.

 

[Steampunk]

http://www.fudzilla.com/news/graphics/42233-vega-10-launching-in-1h-2017

Look like we may have to wait a long time until June for Vega 10 due to issues with HBM2 memory delays, shortages and high costs. Vega 10 will have similar performance as Radeon Pro Duo.

Guess RTG messed up twice again that would see GTX 1080 Ti have no competitor for 6 months. Look like 2017 will be very good year for Nvidia again.

If you're going to quote from the Daily Mail of tech news, at least read the article you've linked to. It doesn't say any of what you claimed, so your conclusions are also invalid.

 

[Dave2150]

I hope something happens soon as I'm looking to replace my 290 and I haven't got the faintest idea about what I'm going to replace it with and when...

1070 is a solid upgrade from a 290 if you can't wait for Vega.

Vega might be released as late as June, plus it might be £800+, since it uses very expensive HBM2, 16GB of it if rumours are to be believed.

 

[kration]

1070 is a solid upgrade from a 290 if you can't wait for Vega.

Vega might be released as late as June, plus it might be £800+, since it uses very expensive HBM2, 16GB of it if rumours are to be believed.

To be honest, I'll be looking at a 1080 early in the new year if there's no certainty re. new releases, as I can afford to drop a bit of cash. Pity AMD doesn't seem to have anything to compete at the top end at the mo...

 

[Seanspeed]

Most of their R and D budget is orientated around process node development and their graphics R and D too. AMD graphics R and D is already covered by their dGPU R and D.

Remember,has managed to go from K10 to BD to PD to SR to Exc and Exc mk2 in the last five years. They also developed Bobcat,Jaguar and Puma which were not far off the Intel Atom chips of the day(except Intel had better process tech). The major costs were going from K10 to BD and developing the Bobcat core. The main issue is BD sucked so was a poor starting point. But look at the Athlon and Athlon 64 - models after that had iterative improvements.

The vast majority of AMD R and D is being spent on developing Zen from the ground up - once they have the basic building blocks for Zen,it will be much cheaper iterative improvements like what Intel has done for years. I imagine the testing required for a totally new uarch must cost a lot of money,since most of the errata will never have been documented before.

With less R and D spend on the CPU side,it will be refocused towards dGPU. If you look at the last few generations of GCN - they have been iterative improvements. Nvidia had one major uarch change with Maxwell. According to certain people on Beyond3D,Vega is a much bigger change than from GC2 to GCN3,or GCN3 to GCN4,but like with Volta I expect Navi in 2018 will be where the major uarch changes are.

Nvidia seems to be moving towards tick-tock. So,a more conservative change at first on a new node,followed by a bigger change when the node is more cost effective. I suspect we might see the same with AMD.

If you've got a quote on their R&D budget being prioritized at GPU development, by all means share it. Otherwise I'm just going to assume you're speculating.

And yes, Intel spends boatloads on process node developments, which has been a very big reason why they've been ahead of the game for quite a while now. That will continue to be the case if AMD dont invest similarly.

You're seriously kidding yourself if you think AMD are in a 'good spot' and can just back off after releasing Zen. Especially if it's already still a generation or two behind in comparative capability when it releases, which it likely will be. If they want to keep in the CPU game, they need to run with Intel and that *will not* be cheap. They'll get outrun very quickly if they dont do so.

And as far as Vega being a bigger architectural shift than before, we'll see. I swear we hear the same exact thing each and every new cycle with AMD GPU's and the reality is more often than not far from meeting those expectations. I'm super hopeful though, dont get me wrong. In fact, I've been holding off a GPU upgrade to a 980Ti/1070 because I'd like to see how Vega stacks up. But only a fool will not have learned to temper expectations.

Well, means nothing, Intel may well have spent $100bn or whatever it is over 10 years to get where they are with CPU's today and AMD may well do it in this one generation for £1.50.

Ya know, you used to be worth discussing things with, even if I disagreed with you on just about everything and thought your arguments were usually pretty lackluster. Now? Feels like you're not even trying anymore.

 

[humbug]

Ya know, you used to be worth discussing things with, even if I disagreed with you on just about everything and thought your arguments were usually pretty lackluster. Now? Feels like you're not even trying anymore.

The fault is with you, say something interesting.

 

[bru]

On the future development front, it is quite easy to forget that GPU's and CPU's take years to develop. Anything AMD do to the R&D budget after the Zen and Vega launches will not be felt for several years to come, so it is the results of the cutting of the budget, over the last few years that we have to look forward too over the next couple.

Hopefully they can come out of the block with a good base line with Zen and indeed Vega, so the recent cutbacks wont effect them too much.