AMD Ryzen Gen 2 Set For Q2 2018

[StinkyPinky]

That s 10% higher perf at same power, this amount to 20%+ perf/watt at same frequency....

https://www.computerbase.de/2018-01/amd-ryzen-threadripper-2000/



50% higher uplift than the current industry trend, wich mean 11-12% instead of 7-8%...

You're right. My bad.

Am very interested to see how they go with the Zen+. Has AMD confirmed the sockets from Zen to Zen 2 are compatible?

 

[antihelten]

These are good points. As of early last year Vega 20 was looking to be 14nm+, which we assumed would be ported over to 12nm. So the switch to 7nm is a little surprising.

I think you might be confusing Vega 20 with something else. Early last year (Sep 2016 actually) Vega 20 was leaked as being on 7nm not 14nm+, so it hardly comes as a surprise now, that it is indeed on 7nm.

 

[moinmoin]

Has AMD confirmed the sockets from Zen to Zen 2 are compatible?

Yes.

 

[maddie]

Maybe they think they need all of 12nm production for Ryzen desktop, Ryzen APUs and Threadripper. This meltdown exploit almost certainly will drive more sales if they are unaffected by it and only GF will fab 12nm unlike 14LPP where Samsung could have helped.

Reinforce success and let the the less competitive lines (GPUs), fend for themselves until a proper response can be attempted. Classic strategies.

 

[Topweasel]

I think you might be confusing Vega 20 with something else. Early last year (Sep 2016 actually) Vega 20 was leaked as being on 7nm not 14nm+, so it hardly comes as a surprise now, that it is indeed on 7nm.

There is a public road map from early last year that had Vega as 14nm and 14nm+ for it's refresh. I am not going by rumors I am going by official roadmaps.

 

[antihelten]

There is a public road map from early last year that had Vega as 14nm and 14nm+ for it's refresh. I am not going by rumors I am going by official roadmaps.

I know (although I'm not sure if I would categorize May as early last year, but that is just semantics), but that roadmap was clearly a very high level overview, and as we now know, clearly didn't cover all of Vega. Either way though, All I'm saying is that I don't see how anyone can really be surprised by Vega 20 on 7nm given the available leaks.

 

[Glo.]

There is a public road map from early last year that had Vega as 14nm and 14nm+ for it's refresh. I am not going by rumors I am going by official roadmaps.

AMD today announced one more GPU with Vega architecture. And nobody is talking about it.


Interesting...

 

[plopke]

AMD today announced one more GPU with Vega architecture. And nobody is talking about it.


Interesting...

which one?

And tbh vega was so overhyped , late , rare ,... i wonder if they should keep the name .

 

[Glo.]

which one?

And tbh vega was so overhyped , late , rare ,... i wonder if they should keep the name .

Vega Mobile with HBM2 stack.

 

[nehway0912]

So what's the difference between 300 and 400 series chipsets, PCIe 3.0 x4 vs. PCIe 2.0 x8? Sorry I don't quite understand AMD strategy here.

 

[turtile]

Maybe they think they need all of 12nm production for Ryzen desktop, Ryzen APUs and Threadripper. This meltdown exploit almost certainly will drive more sales if they are unaffected by it and only GF will fab 12nm unlike 14LPP where Samsung could have helped.

Reinforce success and let the the less competitive lines (GPUs), fend for themselves until a proper response can be attempted. Classic strategies.

They were talking about 7nm being more work than they thought and that they had to use extra resources to keep it on track. My guess is that they took all of the engineers away from 12nm to 7nm. I mean, it's not like they were going to sell a lot more Vega for a 10% speed bump. Or 12nm Vega won't be out until August or later so they didn't mention it. If 12nm Vega doesn't show up, it means that we won't see Zen+ APUs and they will skip to Zen 2 APUs.

 

[Topweasel]

I know (although I'm not sure if I would categorize May as early last year, but that is just semantics), but that roadmap was clearly a very high level overview, and as we now know, clearly didn't cover all of Vega. Either way though, All I'm saying is that I don't see how anyone can really be surprised by Vega 20 on 7nm given the available leaks.

I saw it earlier or at least one that looked like it but even with the leaks and AMD's shift to 12nm. AMD's announcement specifically sounded more like it was going to be more Vega driven than Ryzen. I mean we can go back and forth on rumors, there enough sites, and sites that specifically throw idea's at the wall to see if stuff sticks that you can use them for any narrative you want it to do. Not only was Vega 20 a bit of late thing to find out about (lots of early rumors had AMD going from Vega right to Navi, with Vega having a life almost exactly like Fury).

But AMD specifically listed two process under Vega last year. Mid last year if you want to go by your May point. 14nm and 14nm+. That heavily implied a second Vega on 14nm. Then came the announcement that Goflo got some extra density work figured out and 14nm+ was becoming more like 12nm and AMD said it was a great opportunity for some of their products (again time table on Zen+ implied it was going to be part of that). So it makes sense a mid to late 2018 product like Vega 20 would move to that.

Moving Vega 20 all the way to 7nm is not something a reasonable person saw coming. Mainly because for whatever performance advantages they gain on Vega isn't worth affecting anything else in the pipeline because it isn't going to be good enough or fast enough to make up for the increased delay to market. The only thing Vega 20 on 7nm that would make sense is that 7nm might not be as close to read for AMD to use on Zen 2 as they would hope and using it on a lost cause, small volume release like Vega 20 might give them and GoFlo enough experience to make sure the process is running at its peak in time for Zen 2 and later Navi.

 

[antihelten]

Moving Vega 20 all the way to 7nm is not something a reasonable person saw coming. Mainly because for whatever performance advantages they gain on Vega isn't worth affecting anything else in the pipeline because it isn't going to be good enough or fast enough to make up for the increased delay to market. The only thing Vega 20 on 7nm that would make sense is that 7nm might not be as close to read for AMD to use on Zen 2 as they would hope and using it on a lost cause, small volume release like Vega 20 might give them and GoFlo enough experience to make sure the process is running at its peak in time for Zen 2 and later Navi.

Vega 20 was never moved to 7nm, it was only ever mentioned on 7nm and never ever mentioned in connection with 14nm, 14nm+ or 12nm (outside of misguided speculation), and as such it is obvious that it was never planned for anything other than 7nm, and given the available info this was very much something any reasonable person could see coming. Now I'm not saying that a reasonable person would have been able to predict it with 100% certainty, or even 50% certainty, but it surely shouldn't come as a surprise, given that 7nm is the only node even mentioned in connection to Vega 20.

 

[DrMrLordX]

If they can drop both 7nm Vega and Navi in Q1 2019 then it won't be so bad, but not having at least a refresh product for 2018 is frustrating.

I am still holding out some hope that AMD will find a way to sell me a video card this year. I do not want a 580, and I hesitate to get a Fury X. RX Vega 64 is a non-starter at current prices/availability.

What's funny is that when I look at Vega, I ask myself, "why the hell isn't AMD selling more of these to the consumer market?". HBM2 shortage conspiracies aside, I again and again realize that AMD would make more money selling every Vega they can produce as a professional or AI/deep learning card to someone that isn't me. Even $799 for a consumer card pales in comparison to what corporate customers will pay for essentially the same hardware with a different driver stack.

edit: okay, AMD found a way to sell me a card. Ironically NewEgg had a Frontier Edition for sale for "only" $749, which is less than launch price, and is actually less than I would pay for a normal Vega 64. The driver stack is becoming mature enough for those cards that I am willing to buy one. So for what I am getting, as an AMD fan and a crypto miner, I am not disappointed.

I think I would rather have gotten a consumer card for $499 but hey, at least I have a video card with as much RAM as my main rig. Ha! Now I need a better monitor . . .

 

[Topweasel]

Vega 20 was never moved to 7nm, it was only ever mentioned on 7nm and never ever mentioned in connection with 14nm, 14nm+ or 12nm (outside of misguided speculation), and as such it is obvious that it was never planned for anything other than 7nm, and given the available info this was very much something any reasonable person could see coming. Now I'm not saying that a reasonable person would have been able to predict it with 100% certainty, or even 50% certainty, but it surely shouldn't come as a surprise, given that 7nm is the only node even mentioned in connection to Vega 20.

Dude I just mentioned that AMD specifically showed a roadmap last year that had 2 entries for Vega. 14nm and 14nm+. Navi only had 1 and it was 7nm. So don't lie just to make it seem like AMD isn't changing their plans they obviously are.

 

[raghu78]

If they can drop both 7nm Vega and Navi in Q1 2019 then it won't be so bad, but not having at least a refresh product for 2018 is frustrating.

I think AMD is going to get thrashed by Nvidia once Volta or Ampere arrives on the desktop in Q2 2018. I don't know why AMD scrapped Vega refresh plans when 7nm is more than a year away. Apple is going to eat up all the TSMC N7 capacity this year. Navi on TSMC N7 will not arrive before Q1 2019 (best case) or more realistically Q2 2019. AMD is going to struggle to sell Vega against GV106 which should bring GTX 1070+ perf at 120w-130w. GV104 and GV102 will have no competition. This is going to be worse than Fiji / R9 300 vs Maxwell gen. Atleast there Fiji was just 15-20% slower than custom GTX 980 Ti. We are going to see the worst possible performance gap between Nvidia and AMD ever in GPU history.

 

[Yakk]

I wonder if AMD's change of GPU schedule is more indicative of a much bigger general shift in their business model.

By concentrating on being an OEM GPU supplier to the biggest manufacturers they can keep growing their GPU business ecosystem with themselves at the center HUB. Selling consumer grade GPU for PC market which mostly gets gobbled up by miners does not grow their business. And it looks to be the case for the next couple years at least.

So while it would be sad to see AMD possibly slowly dropping out of the consumer PC add in GPU card space, business wise it's probably the best use of their resources IMO. I just hope they are properly taking care of their AIB partners which might have to change their business models entirely.

 

[Lawrence Yan]

I think AMD is going to get thrashed by Nvidia once Volta or Ampere arrives on the desktop in Q2 2018. I don't know why AMD scrapped Vega refresh plans when 7nm is more than a year away. Apple is going to eat up all the TSMC N7 capacity this year. Navi on TSMC N7 will not arrive before Q1 2019 (best case) or more realistically Q2 2019. AMD is going to struggle to sell Vega against GV106 which should bring GTX 1070+ perf at 120w-130w. GV104 and GV102 will have no competition. This is going to be worse than Fiji / R9 300 vs Maxwell gen. Atleast there Fiji was just 15-20% slower than custom GTX 980 Ti. We are going to see the worst possible performance gap between Nvidia and AMD ever in GPU history.

AMD's GPUs are what ensured that bankruptcy didn't happen between 2012 - 2016. Zen is what will bring AMD back as a genuine competitor in the space. Right now, however, AMD is still close enough to the edge that I don't blame them for continuing to focus mainly on their CPUs. They threw a hail mary and it paid off, and especially now with things Meltdown, AMD wants to make very sure that they nail down EPYC. Once they're back to full stability, I think there will be a fresh source of heavy investment back into GPUs. 2019 should be the year of the GPU assuming all goes well in the year of 2018 for the CPU.

 

[eek2121]

I think AMD is going to get thrashed by Nvidia once Volta or Ampere arrives on the desktop in Q2 2018. I don't know why AMD scrapped Vega refresh plans when 7nm is more than a year away. Apple is going to eat up all the TSMC N7 capacity this year. Navi on TSMC N7 will not arrive before Q1 2019 (best case) or more realistically Q2 2019. AMD is going to struggle to sell Vega against GV106 which should bring GTX 1070+ perf at 120w-130w. GV104 and GV102 will have no competition. This is going to be worse than Fiji / R9 300 vs Maxwell gen. Atleast there Fiji was just 15-20% slower than custom GTX 980 Ti. We are going to see the worst possible performance gap between Nvidia and AMD ever in GPU history.

AMD does not (and will not) have any issues selling GPUs in the near future. Demand is simply too high and will be for the foreseeable future.

 

[coercitiv]

Can we please move on from the dGPU discussion? This is not the place for it.

 

[DrMrLordX]

I think AMD is going to get thrashed by Nvidia once Volta or Ampere arrives on the desktop in Q2 2018.

I don't know that I agree. The demand for AMD dGPUs is insane . . . even older products. Let's all count our blessings that AMD isn't selling CPUs like that.

I wonder if AMD's change of GPU schedule is more indicative of a much bigger general shift in their business model.

Yup. It makes me wonder how much longer AMD is going to sell CPUs to the consumer market. If things go the way I think they could go in large datacenters over the next few years, demand for Zen/Zen 2 dice on the server side may crush the consumer market. It hasn't happened yet, and Ryzen at least was cheap enough to produce that AMD has kept the channel well-supplied with their products.

Can we please move on from the dGPU discussion? This is not the place for it.

Sorry if things got off track, but you have to admit, it was a pretty stunning announcement in what was an otherwise-positive roadmap launch for AMD. Imagine what it would have been like if AMD had come out and said, "Oh sorry, we cancelled Ryzen 2, but have a look at our Vega refresh instead!". The only difference there is that Ryzen 1 chips aren't selling for above MSRP, and they have good supply.

 

[antihelten]

Dude I just mentioned that AMD specifically showed a roadmap last year that had 2 entries for Vega. 14nm and 14nm+. Navi only had 1 and it was 7nm. So don't lie just to make it seem like AMD isn't changing their plans they obviously are.

Please stop accusing me of lying, when I clearly explained to you how said roadmap was obviously just a high level overview of AMD products, and as such didn't account for every product.

You're the one pretending that the roadmap mentioned Vega 20 specifically, when it obviously didn't (it didn't mention any specific products, only the Vega architecture as a whole). The only time Vega 20 has ever been mentioned specifically is in connection to 7nm.

AMD making high level slides like this that don't account for every single product within a series is nothing new. For instance when they introduced RX 580, they had a number of slides about the new 500 series, which among other things said that the series was made on 14nm, but of course we know that the 500 series also contains 28nm products. All this means is that these kind of slides shouldn't be interpreted all that strictly.

 

[Topweasel]

Please stop accusing me of lying, when I clearly explained to you how said roadmap was obviously just a high level overview of AMD products, and as such didn't account for every product.

You're the one pretending that the roadmap mentioned Vega 20 specifically, when it obviously didn't (it didn't mention any specific products, only the Vega architecture as a whole). The only time Vega 20 has ever been mentioned specifically is in connection to 7nm.

AMD making high level slides like this that don't account for every single product within a series is nothing new. For instance when they introduced RX 580, they had a number of slides about the new 500 series, which among other things said that the series was made on 14nm, but of course we know that the 500 series also contains 28nm products. All this means is that these kind of slides shouldn't be interpreted all that strictly.

Okay not lying just being Obtuse. AMD last year had two generations of Vega both on 14nm on of the two being 14nm+. One for Navi and it was the first generation to include 7nm.

If that second Vega generation isn't Vega 20 than what was it?

 

[antihelten]

Okay not lying just being Obtuse.

Pot meet kettle.

AMD last year had two generations of Vega both on 14nm on of the two being 14nm+. One for Navi and it was the first generation to include 7nm.

Navi was the first generation to officially include 7nm, but we had leaks long before that about Vega 20 on 7nm.

If that second Vega generation isn't Vega 20 than what was it?

We obviously don't know exactly, because as I already mentioned twice before, the roadmap was just a high level overview. And we might never know since Vega on 14nm+/12nm is now no longer part of the new roadmap. But if we once again assume that this new roadmap is also just a high-level overview that doesn't necessarily cover every nook and cranny, then I would say that we might at some point see Vega on 14nm+/12nm as part of a Zen+ APU.

 

[Topweasel]

Pot meet kettle.



Navi was the first generation to officially include 7nm, but we had leaks long before that about Vega 20 on 7nm.



We obviously don't know exactly, because as I already mentioned twice before, the roadmap was just a high level overview. And we might never know since Vega on 14nm+/12nm is now no longer part of the new roadmap. But if we once again assume that this new roadmap is also just a high-level overview that doesn't necessarily cover every nook and cranny, then I would say that we might at some point see Vega on 14nm+/12nm as part of a Zen+ APU.

It's pointless to keep this up my point was AMD had a public outline that included 14nm+ generation of Vega nothing more nothing less. And now it's straight to 7nm Vega as publicly announced. I have been really clear on that. It is a sign that AMD has changed it's public planning of Vega. That doesn't mean they haven't been working on the option to switch to 7nm for Vega for awhile now. But treating me as being wrong because some throw spegatti at the wall rumor sight suggested it was possible doesn't mean that AMD who has public stock owners to speak to hasn't changed it's public plans.

Referring to AMD public materials doesn't make me Obtuse and not actually using the Code Name Vega 20 doesn't mean the second generation of Vega on 14nm+ wasn't referring specifically to that.