Vega Not Before 2017: AMD to Investors

[$ReaPeR$]

With new DX12 and Vulkan games AMD cards are performing a lot better than DX11 games.

Volta will be a completely new design with Async compute in mind. Thus explains the lack of new flagship product from AMD.

RTG can put 8192ALU to make a flagship, however without good DX12 applications it will still loose to Nvidia's Pascal in DX11 games.

thats not enough, games will not come out that quickly, also Nvidia will not sit with their hands on their asses.

 

[Captain_Tom]

This pretty much confirms my suspicion - AMD is just churning out cheap marketshare-capturing chips while it waits for three things:

1) Mature 14nm

2) Full HBM2 availability. I truly believe Vega will be HBM top to bottom (460 - Fury). I think they will re-release 465, 475, and 485 chips with HBM on-board in addition to their new 495 and Fury series.

3) DX12 to be the near standard. Look at how the Fury X matches the 1080 in DX12. AMD doesn't want to launch its high-end until their cards are fully utilized.


The only thing I hope is that a 12GB 384-bit Polaris cards will launch in October to face the 1070/1080. Otherwise AMD is really leaving its enthusiast fans high and dry.

 

[Audiophizile]

This pretty much confirms my suspicion - AMD is just churning out cheap marketshare-capturing chips while it waits for three things:

1) Mature 14nm

2) Full HBM2 availability. I truly believe Vega will be HBM top to bottom (460 - Fury). I think they will re-release 465, 475, and 485 chips with HBM on-board in addition to their new 495 and Fury series.

3) DX12 to be the near standard. Look at how the Fury X matches the 1080 in DX12. AMD doesn't want to launch its high-end until their cards are fully utilized.


The only thing I hope is that a 12GB 384-bit Polaris cards will launch in October to face the 1070/1080. Otherwise AMD is really leaving its enthusiast fans high and dry.

"Look at how the fury x matches 1080 in dx12" source? In the 1 vulkan game out it can come close. AMD waiting for vulkan to mainstream so it can compete is a joke. Vulkan to be even 20% of released games in the next 2 years is unlikely and that would still only make hand full of games. We should wait more than 2 years for AMD to compete? That sounds like a terrible idea.

 

[dyonoctis]

This pretty much confirms my suspicion - AMD is just churning out cheap marketshare-capturing chips while it waits for three things:

1) Mature 14nm

2) Full HBM2 availability. I truly believe Vega will be HBM top to bottom (460 - Fury). I think they will re-release 465, 475, and 485 chips with HBM on-board in addition to their new 495 and Fury series.

3) DX12 to be the near standard. Look at how the Fury X matches the 1080 in DX12. AMD doesn't want to launch its high-end until their cards are fully utilized.


The only thing I hope is that a 12GB 384-bit Polaris cards will launch in October to face the 1070/1080. Otherwise AMD is really leaving its enthusiast fans high and dry.

As far I know Vega isn't supposed to be a refresh of polaris, but the high performance chip (maybe 4096 stream for vega 10, and maybe 6144 stream for Vega 11).
I really wish that they could use HBM 2 on the whole gaming line-up. I really do. But it's unlikely. I don't think that HBM2 is cheap enough for that.
And just like btarunr said, Amd is planning to leave a huge gap between polaris chips and vega chips. It wouldn't make any sense for them to do a surprise launch of a "polaris 9".
If you look at how a rx 480 is as power hungry as a gtx 1070, I can understand that Amd will not want to lauch a bigger chip unless it's using HBM2.

 

[Prima.Vera]

So because of AMD incompetence, now we are stuck with 2 monopolies companies, Intel and nVidia.
Translation, this is bad mostly for the users who will be forced from now on to buy even more callously expensive products from the above 2, if they want top quality for their systems.
Why the hell in the world are we lacking a 3rd competitor for both CPU and GPU ???

 

[Captain_Tom]

As far I know Vega isn't supposed to be a refresh of polaris, but the high performance chip (maybe 4096 stream for vega 10, and maybe 6144 stream for Vega 11).
I really wish that they could use HBM 2 on the whole gaming line-up. I really do. But it's unlikely. I don't think that HBM2 is cheap enough for that.
And just like btarunr said, Amd is planning to leave a huge gap between polaris chips and vega chips. It wouldn't make any sense for them to do a surprise launch of a "polaris 9".
If you look at how a rx 480 is as power hungry as a gtx 1070, I can understand that Amd will not want to lauch a bigger chip unless it's using HBM2.

I am only basing my prediction on the roadmap they released that shows Vega occupying all of 2017. Usually they would put Polaris under Vega if it was going to continue to be sold at the same time.


As for power usage - it truly comes down to both DX12 and HBM. I got a Fury for $310 the other day, and according to Hardwarebot the memory portion only uses ~10-20w as opposed to the 80-100w it said my 480's GDDR5 memory controller was using. Thus the 480 would be a 75w card if it had HBM. But don't underestimate DX12. In Vulkan the 480 is ~80% as strong as the 1070, meaning it has almost the same efficiency when fully utilized.

AMD wants both sources of improved efficiency.

So because of AMD incompetence, now we are stuck with 2 monopolies companies, Intel and nVidia.
Translation, this is bad mostly for the users who will be forced from now on to buy even more callously expensive products from the above 2, if they want top quality for their systems.
Why the hell in the world are we lacking a 3rd competitor for both CPU and GPU ???

I gotta say that at least Intel continues to massively improve efficiency and performance. Sure their consumer i7's aren't much better, but they have 28-Core Xeons. On the otherhand Nvidia is already milking with cut dies that still use GDDR5. They could release a $1000 3840-SP 16GB HBM2 monster RIGHT NOW, but instead they would prefer to just sell a 1080 Ti for $1200.

 

[dwade]

Vega vs GTX 2080 in 2017. Things will get uglier than a Master P's sneaker.

 

[Audiophizile]

I am only basing my prediction on the roadmap they released that shows Vega occupying all of 2017. Usually they would put Polaris under Vega if it was going to continue to be sold at the same time.


As for power usage - it truly comes down to both DX12 and HBM. I got a Fury for $310 the other day, and according to Hardwarebot the memory portion only uses ~10-20w as opposed to the 80-100w it said my 480's GDDR5 memory controller was using. Thus the 480 would be a 75w card if it had HBM. But don't underestimate DX12. In Vulkan the 480 is ~80% as strong as the 1070, meaning it has almost the same efficiency when fully utilized.

AMD wants both sources of improved efficiency.

Except a 1070 uses less power than 480 and is still 20% faster meaning the 480 is still pretty far behind in efficiency.

 

[NC37]

I was Team AMD but after so much time hyping so much their hardware and fail(Fury OC) after fail(RX480 OC/ PCI-Sig issue) I lost confidence on them.

Then you should probably write off any other team too. Just about every team has had issues or times when they've screwed over customers. nVidia's 970 lawsuits are the latest.

Fact is any team is going to screw you. They are out to make money. When they flop they are more interested in making their failure seem like a good thing. It's just how the world works.

Best thing to do is play the field. Don't be constrained.

 

[dyonoctis]

I am only basing my prediction on the roadmap they released that shows Vega occupying all of 2017. Usually they would put Polaris under Vega if it was going to continue to be sold at the same time.


As for power usage - it truly comes down to both DX12 and HBM. I got a Fury for $310 the other day, and according to Hardwarebot the memory portion only uses ~10-20w as opposed to the 80-100w it said my 480's GDDR5 memory controller was using. Thus the 480 would be a 75w card if it had HBM. But don't underestimate DX12. In Vulkan the 480 is ~80% as strong as the 1070, meaning it has almost the same efficiency when fully utilized.

AMD wants both sources of improved efficiency.

I Think that this graph is more or less about launching date rather than acurate life-span. The 28 nm goes back to 2012, yet they put it slightly behind 2015. Polaris is in between 2016 and 2017, while vega sit above 2017. (since it will launch in the first half of 2017.) The second clue is the way amd use the codename. Just think as polaris 10 as being Tahiti, and polaris 11 pitcairn. Vega10 is the new hawaii, and Vega 11 is just the new Fiji XT. It's already been esthablished that Vega will be the high-end chip, and not the refresh, if that was the case we would have heard of something a crazy as "HBM2 on a 179$ Gpu". We are like 3 years too soon for something like that to happen.

 

[Captain_Tom]

I Think that this graph is more or less about launching date rather than acurate life-span. The 28 nm goes back to 2012, yet they put it slightly behind 2015. Polaris is in between 2016 and 2017, while vega sit above 2017. (since it will launch in the first half of 2017.) The second clue is the way amd use the codename. Just think as polaris 10 as being Tahiti, and polaris 11 pitcairn. Vega10 is the new hawaii, and Vega 11 is just the new Fiji XT. It's already been esthablished that Vega will be the high-end chip, and not the refresh, if that was the case we would have heard of something a crazy as "HBM2 on a 179$ Gpu". We are like 3 years too soon for something like that to happen.

No I mean this:

http://images.techtimes.com/data/images/full/234571/amd-gpu-roadmap-2016-2018.jpg

Notice how both the Fury and 300 series were shown, and also how in their 2017 CPU roadmaps they show Zen above Bristol-Ridge APU's (Even though Bristol Ridge is launching in 2016). Supposedly Samsung will begin manufacturing cheap HBM2 chips soon. They will offer ~720 GB/s instead of the 1 TB/s ones meant for the top-end chipsets (Vega 11 / GP100), and supposedly cost decently less than HBM1. If you think about it they could put 4GB of 2048-bit HBM2-cheapo on the 480 core and get 360 GB/s of bandwidth in addition to massive power savings.

Remember their 485 update they hinted at? Not saying this WILL happen, but it does seem like a thing they might go for.

 

[Camm]

I've seen some odd shit in this thread...

Firstly, Nvidia has stated multiple times Volta is a 2018 part. Whatever happens with Vega in 2017 will be against Pascal, not Volta.

As for HBM2, AMD has priority order out of multiple fabs, I fully expect HBM2 to be AMD's game for 2017.

As for efficiency, its been well known for quite sometime that Pascal and previous architectures down to Kepler that they are DX11\OpenGL optimised chips, sacrificing compute performance to get better efficiency (both power and performance per nm) in most tasks. This means that if Volta has to re-architect to take advantage of GPU compute, those efficiencies will vanish (or at least what currently looks like a chasm of efficiency anyway).

 

[Assimilator]

Firstly, Nvidia has stated multiple times Volta is a 2018 part. Whatever happens with Vega in 2017 will be against Pascal, not Volta.

https://www.techpowerup.com/224413/nvidia-accelerates-volta-to-may-2017

This means that if Volta has to re-architect to take advantage of GPU compute, those efficiencies will vanish (or at least what currently looks like a chasm of efficiency anyway).

There's no evidence that nVIDIA's perf/watt advantage is due to their prioritising of DX11 functionality over compute, just as there's no evidence that AMD's disadvantage in the same department is due to them concentrating on compute. And given that nVIDIA's been on a consistent and successful push to increase perf/watt every generation since the Fermi debacle, I see every reason to expect the same of Volta.

 

[Camm]

https://www.techpowerup.com/224413/nvidia-accelerates-volta-to-may-2017



There's no evidence that nVIDIA's perf/watt advantage is due to their prioritising of DX11 functionality over compute, just as there's no evidence that AMD's disadvantage in the same department is due to them concentrating on compute. And given that nVIDIA's been on a consistent and successful push to increase perf/watt every generation since the Fermi debacle, I see every reason to expect the same of Volta.

There's plenty of evidence.

http://www.theinquirer.net/inquirer...hed-amds-mid-range-radeon-hd-7870-gpu-compute

You remove compute off your die, you get more die space for things that help you in other tasks. This wasn't necessarily the wrong decision at all in 2011, but with the nature of DX12 and Vulkan, its a decision Nvidia is in the process of reversing.

Also, the article you linked uses WCCFtech as its source with no relevant product maps. Try again.

 

[ZoneDymo]

easy, just drop the price on the fury cards and voila, competition

 

[FordGT90Concept]

The problem is that Fury costs a lot more to produce than the Pascal chips. AMD has to drop the price on them some but I'm curious how low the floor is.

 

[xkm1948]

There's plenty of evidence.

http://www.theinquirer.net/inquirer...hed-amds-mid-range-radeon-hd-7870-gpu-compute

You remove compute off your die, you get more die space for things that help you in other tasks. This wasn't necessarily the wrong decision at all in 2011, but with the nature of DX12 and Vulkan, its a decision Nvidia is in the process of reversing.

Also, the article you linked uses WCCFtech as its source with no relevant product maps. Try again.

Pretty much this. Real DX12/Vulkan applications will demolish Maxwell/Pascal based GPU completely. However by that time the new Volta structure with Async compute design from the start will take over. It is Nvidia after all, once Volta rolls out it will be bad news for current gen Nvidia GPUs. AMD's GCN will suffer as well from Nvidia's "optimization", just not as bad as Maxwell/Pascal.

 

[dyonoctis]

No I mean this:

http://images.techtimes.com/data/images/full/234571/amd-gpu-roadmap-2016-2018.jpg

Notice how both the Fury and 300 series were shown, and also how in their 2017 CPU roadmaps they show Zen above Bristol-Ridge APU's (Even though Bristol Ridge is launching in 2016). Supposedly Samsung will begin manufacturing cheap HBM2 chips soon. They will offer ~720 GB/s instead of the 1 TB/s ones meant for the top-end chipsets (Vega 11 / GP100), and supposedly cost decently less than HBM1. If you think about it they could put 4GB of 2048-bit HBM2-cheapo on the 480 core and get 360 GB/s of bandwidth in addition to massive power savings.

Remember their 485 update they hinted at? Not saying this WILL happen, but it does seem like a thing they might go for.

Because Fury and R9 300 were both launched in 2015. Meanwhile Polaris was launched in 2016 , and vega will be lauched at some point in 2017. That's how I'm reading this roadmap. I don't want to go into far-fetched hype. The facts that I have right now : vega will use HBM2, as it is written in the slides/roadmap. Vega will be launched in H1 2017 for the enthusiast market, as it is written in the slide. The 460,470 and the 480 are not included in that market.

AMD did hinted at a refresh , but if by refresh they meant HBM 2 for everyone, I doubt that they do be so secretive about that. Especially when that refresh would do wonders on the perf/watt ratio. As long as I'm not seeing an official Amd document saying "RX 485 with HBM2" I'm sticking with vega only being an enthusiast chip.

 

[FordGT90Concept]

Pretty much this. Real DX12/Vulkan applications will demolish Maxwell/Pascal based GPU completely. However by that time the new Volta structure with Async compute design from the start will take over. It is Nvidia after all, once Volta rolls out it will be bad news for current gen Nvidia GPUs. AMD's GCN will suffer as well from Nvidia's "optimization", just not as bad as Maxwell/Pascal.

Volta and Vega are expected to launch next year and both are expected to have HBM2. Volta/Vega cards should be in the $400+ range right now, not Pascal. the only reason why Pascal is at $400+ is because the jump between 28nm and 16nm literally turned what should be a mid-range card into high end.

 

[R-T-B]

So you base your opinion on the perspective performance of overclocking? That's disappointment waiting to happen.

I guess you didn't get the memo that many Nvidia cards also exceeded the PCIe slot's 75w spec by much more grievous amounts. Please, you are just here to bash one side.

Not really outside dual-GPU and Fermi... unless I missed something.

The RX 480 is a good chip but so are the pascal video cards. The RX 480 brought AMD much closer to Nvidia in power consumption and with similar performance. Considering that they are already setup for DX 12 and Vulkan, I would not call it a fail.

Agreed.

 

[RejZoR]

It's possible they'll use R9 Fury X foundation, shrink it down to 14nm and update the GCN to Polaris version if not even newer (there were speculations about GCN 4.0). Considering how well Fury X is holding up even against GTX 1080 in Vulkan and DX12, AMD isn't really worrying too hard about it. I mean, I'm sure Fury X owners have a big grin on their face, knowing their cards will hold up quite decently even in the future despite initial slight disappointment. Something NVIDIA owners of GTX 900 can just forget about...

 

[the54thvoid]

It's possible they'll use R9 Fury X foundation, shrink it down to 14nm and update the GCN to Polaris version if not even newer (there were speculations about GCN 4.0). Considering how well Fury X is holding up even against GTX 1080 in Vulkan and DX12, AMD isn't really worrying too hard about it. I mean, I'm sure Fury X owners have a big grin on their face, knowing their cards will hold up quite decently even in the future despite initial slight disappointment. Something NVIDIA owners of GTX 900 can just forget about...

Speak for yourself. In one of the most AMD favourable games in recent history my 980ti (thanks to Maxwell's OC room) gets ~60fps on Deus Ex MD at 1440p on a setting of very high and using ultra textures. I don;t have W10 so DX12 is irrelevant to me for now. The irony is I'll move to a clean W10 install on my next system upgrade - which keeps on getting put back because I'd like to see Zen. But.... my patience, like my hair, is wearing thin.

FTR - to give you a clue as to how AMD centric Deus Ex HR is, look at what PC he uses (found in the heroes bedroom)

Well, at least someone got one. Maybe there's a Zen inside it?

 

[Camm]

Not really outside dual-GPU and Fermi... unless I missed something.

950's spiked over quite badly.

 

[RejZoR]

Speak for yourself. In one of the most AMD favourable games in recent history my 980ti (thanks to Maxwell's OC room) gets ~60fps on Deus Ex MD at 1440p on a setting of very high and using ultra textures. I don;t have W10 so DX12 is irrelevant to me for now. The irony is I'll move to a clean W10 install on my next system upgrade - which keeps on getting put back because I'd like to see Zen. But.... my patience, like my hair, is wearing thin.

FTR - to give you a clue as to how AMD centric Deus Ex HR is, look at what PC he uses (found in the heroes bedroom)

Well, at least someone got one. Maybe there's a Zen inside it?

Because it's a DX11 game for now. When it gets DX12, Maxwell 2 users can expect pretty much zero benefits. Where AMD's cards will get huge boost across the range of GCN powered cards.

 

[the54thvoid]

Because it's a DX11 game for now. When it gets DX12, Maxwell 2 users can expect pretty much zero benefits. Where AMD's cards will get huge boost across the range of GCN powered cards.

That's the point, (well, my point), I have a 60hz refresh monitor and if I had DX12, I don't need the game to play faster. DX12 benefits GCN because GCN 'needs' to be better.
AMD ought to be better than they are. DX12 gets them there.