Thursday, August 1st 2019

AMD Could Launch New Navi GPUs Soon

AMD's president and CEO Dr. Lisa Su was talking during AMD's Q2 earnings Q&A conference and got asked a very interesting question. When prompted about high end Navi GPUs, Dr. Su answered with "I would say they are coming. You should expect that our execution on those are on track and we have a rich 7 nm portfolio beyond the products that we have already announced in the upcoming quarters."

This answer gives us hope to see more powerful Navi GPUs possibly by the end of the year, meaning that AMD's answer to Turing is almost ready. As we saw earlier in the rumors, we might get additional higher end GPU models in form of alleged RX 5800 and RX 5900, with XT variants available for both of those models. The RX 5800 is supposed to utilize a new GPU core called Navi 12, while the core for RX 5900 is still unknown.
Source: VideoCardz
Add your own comment

92 Comments on AMD Could Launch New Navi GPUs Soon

#26
Mephis
Patriot, post: 4090788, member: 77367"
Big Navi was finished, then it got respun on 7nm+ /euv. each spin is 3mo. It could arrive in Jan if the first spin was successful...otherwise April+
Source please?

Every thing I have seen says that TSMC expected N7 (7NM DUV) customers, which Zen 2 and Navi are, to transition to 6NM (N6) when they move to EUV. N7 and N6 use the same design rules, while N7+ (7NM EUV) and N7P (7NM High Performance DUV) use different design rules.

As far a I understand, it would take more than a respin to go from 7NM DUV to 7NM EUV, due to the different design rules.
Posted on Reply
#27
Patriot
Mephis, post: 4090823, member: 186806"
Source please?

Every thing I have seen says that TSMC expected N7 (7NM DUV) customers, which Zen 2 and Navi are, to transition to 6NM (N6) when they move to EUV. N7 and N6 use the same design rules, while N7+ (7NM EUV) and N7P (7NM High Performance DUV) use different design rules.

As far a I understand, it would take more than a respin to go from 7NM DUV to 7NM EUV, due to the different design rules.
It has been a couple of months since that news came out, 7nm+/euv is mass production march currently so not going to be a Jan launch but might be a demo. AMD slides have 7nm euv on roadmap but if 6nm is easier and sooner nothing prevents that given it would be an easier conversion. What is known for is the power reduction or density increase with euv. And navi needs it to grow bigger.
Posted on Reply
#28
Mephis
Patriot, post: 4090835, member: 77367"
It has been a couple of months since that news came out, 7nm+/euv is mass production march currently so not going to be a Jan launch but might be a demo. AMD slides have 7nm euv on roadmap but if 6nm is easier and sooner nothing prevents that given it would be an easier conversion. What is known for is the power reduction or density increase with euv. And navi needs it to grow bigger.
So you are guessing, based on available information. I thought that maybe you saw something I didn't, or that maybe you had an inside source.

My guess is that big Navi will be 7nm DUV, just like current Navi. The delay is due to trying to work the kinks out of a new process on larger die and accumulating enough chips that meet specs. I don't think we will see 7nm EUV until Zen 2+ (or whatever the refresh is called). I think and I hope, considering how expensive new nodes are, that they wait till a new GPU to move nodes.
Posted on Reply
#29
Casecutter
"Quarters" = a couple... so probably a month or so into 2020.

For me it's the small Navi's, what can RTG provide if they stay on GloFlo 12nm using to 1200-1500 Cores, 80-90 TMU's, 32 ROP's with 4/8Gb DDR6? I think they don't need 7nm to show what a Navi can offer in 75-120W packages. Holding to a very cost effectively node, could we see high side performance bettering RX 590 for <$150 MSRP? That still leaves a 7nm variants to com in and bust the 1660 and 1660Ti at MSRP's that are like $200 and $250.
Posted on Reply
#30
efikkan
Patriot, post: 4090835, member: 77367"
What is known for is the power reduction or density increase with euv. And navi needs it to grow bigger.
EUV is going to help yields, but density will be held back by energy density long before transistor density.

A much larger Navi will run into problems with energy consumption and have to sacrifice some clock speed to stay within reasonable power consumption.
And it seems that everyone knows there is a "big navi" coming, and it may very well be coming, but do we actually have any solid indicators about what a bigger navi would offer? It seems like everyone assumes it will offer more shader processors. I just want to remind people that many had the same assumption about Vega 20.

Casecutter, post: 4090839, member: 94772"
"Quarters" = a couple... so probably a month or so into 2020.
Oh, the mathematician in me cringes when people uses equality that way. :(

These words by Lisa Su is probably very intentionally chosen to leave any ambiguity they would need in case plans changes, and makes it sounds like she is saying something, when she is actually revealing absolutely nothing. There is no reasoning that would lead this to meaning exactly two quarters. What she really said is that they will refresh the lineup within the next ~1.5 years or so, which we would expect anyway, so there is nothing new here.

I don't doubt there may be something coming soon, but I would expect a smaller Navi first.
Posted on Reply
#31
kapone32
efikkan, post: 4090855, member: 150226"
EUV is going to help yields, but density will be held back by energy density long before transistor density.

A much larger Navi will run into problems with energy consumption and have to sacrifice some clock speed to stay within reasonable power consumption.
And it seems that everyone knows there is a "big navi" coming, and it may very well be coming, but do we actually have any solid indicators about what a bigger navi would offer? It seems like everyone assumes it will offer more shader processors. I just want to remind people that many had the same assumption about Vega 20.


Oh, the mathematician in me cringes when people uses equality that way. :(

These words by Lisa Su is probably very intentionally chosen to leave any ambiguity they would need in case plans changes, and makes it sounds like she is saying something, when she is actually revealing absolutely nothing. There is no reasoning that would lead this to meaning exactly two quarters. What she really said is that they will refresh the lineup within the next ~1.5 years or so, which we would expect anyway, so there is nothing new here.

I don't doubt there may be something coming soon, but I would expect a smaller Navi first.
I did comment on another post that for me they want to exhaust the supply of Polaris and Vega cards before giving us anything new on the Navi side.
Posted on Reply
#32
Vayra86
dj-electric, post: 4090558, member: 87186"
Please. Prepare. Your. Partners. Well. Ahead.

Seriously, AMD. Missing advanced designs are a forfeit to sales. Don't so it again.
You just know they will do it again anyway :p

In other news, Intel could get 10nm on track soon. :D See what I did there
Posted on Reply
#33
bug
efikkan, post: 4090781, member: 150226"
So there may be new models in the "upcoming quarters", there will also be a new generation from Nvidia in the "upcoming quarters". If "big navi" arrives close to or after Nvidia's next generation, it will be too little too late.

I also wonder where people got the idea that "big navi" is right around the corner.
Probably wishful thinking more than anything.
Posted on Reply
#34
Assimilator
Vayra86, post: 4090877, member: 152404"
You just know they will do it again anyway :p

In other news, Intel could get 10nm on track soon. :D See what I did there
10nm is never going to be on track for Intel. It is a dead node that they are using only for learning from, to ensure that they don't make the same half-decade misstep with their next node shrink.
Posted on Reply
#35
Darmok N Jalad
Mephis, post: 4090823, member: 186806"
Source please?

Every thing I have seen says that TSMC expected N7 (7NM DUV) customers, which Zen 2 and Navi are, to transition to 6NM (N6) when they move to EUV. N7 and N6 use the same design rules, while N7+ (7NM EUV) and N7P (7NM High Performance DUV) use different design rules.

As far a I understand, it would take more than a respin to go from 7NM DUV to 7NM EUV, due to the different design rules.
It would depend. If it is a standard transition, then you are correct and the design needs to be adjusted for the new node—it’s not simply a shrink of an existing design. That said, AMDs 12nm transition was quite different. They shrunk the transistors but not the design, resulting in unused die space between transistors. They simply capitalized on the energy savings and reduced thermal density. I’d never seen that approach before, so I’m not sure how common the latter is.
Posted on Reply
#36
efikkan
Assimilator, post: 4090920, member: 7058"
10nm is never going to be on track for Intel. It is a dead node that they are using only for learning from, to ensure that they don't make the same half-decade misstep with their next node shrink.
There is no reason why Intel can't make their 10nm node perform similarly to TSMC's "7nm" on DUV, it's just a question of iterations and adjustments. And to my knowledge, Intel seem to plan to launch their first dedicated graphics chips on 10nm.

Intel's 10nm DUV node isn't dead or broken, and the yield issues are solvable, but the "bigger" problem is achieving high enough production volume. This is the reason why Intel probably will not transition completely to 10nm, assuming 7nm isn't delayed of course.
Posted on Reply
#37
bug
efikkan, post: 4090943, member: 150226"
There is no reason why Intel can't make their 10nm node perform similarly to TSMC's "7nm" on DUV, it's just a question of iterations and adjustments. And to my knowledge, Intel seem to plan to launch their first dedicated graphics chips on 10nm.

Intel's 10nm DUV node isn't dead or broken, and the yield issues are solvable, but the "bigger" problem is achieving high enough production volume. This is the reason why Intel probably will not transition completely to 10nm, assuming 7nm isn't delayed of course.
I've heard two theories about Intel's 10nm. One is that it's going to be short-lived, replaced in 2021 by 7nm. The other is that 7nm is going to go towards their GPU first and we're in for 10nm+ and 10nm++.
My money's on the latter, but it's just a hunch.
Posted on Reply
#38
Manu_PT
ratirt, post: 4090652, member: 165024"
Are you saying that used cards are louder than new cards even if it's the exact same kind of a card? Cause it does sounds like it and I can't see any difference aside for the used cards' warranty is shortened.


NV sure does lead, mostly in the pricing department. :) Unfortunately this ain't my box of candy though. :)
Be grateful if AMD comes up with something new to stand against NV's 2080 and TI. New release from NV to "smash AMD" will surely give you more arguments :)
:)
Posted on Reply
#39
efikkan
bug, post: 4090945, member: 157434"
I've heard two theories about Intel's 10nm. One is that it's going to be short-lived, replaced in 2021 by 7nm. The other is that 7nm is going to go towards their GPU first and we're in for 10nm+ and 10nm++.
My money's on the latter, but it's just a hunch.
Intel have clearly stated in May this year that they plan both 10nm++ and 7nm for 2021. Many people are forgetting that Intel's current Ice Lake-U/-Y are still on the first generation 10nm node. If we for a moment think back in time about the iterations from early 14nm to 14nm+ and 14nm++, it should be clear that we can't judge the future iterations of the node yet. Intel wouldn't plan a 10nm++ for 2021 if it was pointless.
Posted on Reply
#40
bug
efikkan, post: 4090950, member: 150226"
Intel have clearly stated in May this year that they plan both 10nm++ and 7nm for 2021. Many people are forgetting that Intel's current Ice Lake-U/-Y are still on the first generation 10nm node. If we for a moment think back in time about the iterations from early 14nm to 14nm+ and 14nm++, it should be clear that we can't judge the future iterations of the node yet. Intel wouldn't plan a 10nm++ for 2021 if it was pointless.
Well, calling 10nm "first generation" is a stretch, but what you said jives well with the version I had my money on ;)
Posted on Reply
#41
danbert2000
A launch in November could peel off quite a few holiday sales from Nvidia, especially as stocks dry up. I'm curious how high up they can go with the TDP though, as 300 W is probably the most people will accept at this point and the 5700 XT is already pushed up the V/F curve to try to max out performance and still can't keep up with a 2070 Super with the same power consumption. Presumably wider chips will be able to stick to the sweet spot and maybe AMD will finally beat the 1080 Ti in a consumer product (not counting Radeon VII, which was never really that available and still struggled with some games).

Really, AMD can try all they are able to and they're still not going to break 30% market share for years. Navi is looking good for the new consoles, though. That should make everyone happy as Polaris ended up holding up game design a bit in the last few years.
Posted on Reply
#42
bug
danbert2000, post: 4090963, member: 165365"
A launch in November could peel off quite a few holiday sales from Nvidia, especially as stocks dry up. I'm curious how high up they can go with the TDP though, as 300 W is probably the most people will accept at this point and the 5700 XT is already pushed up the V/F curve to try to max out performance and still can't keep up with a 2070 Super with the same power consumption. Presumably wider chips will be able to stick to the sweet spot and maybe AMD will finally beat the 1080 Ti in a consumer product (not counting Radeon VII, which was never really that available and still struggled with some games).

Really, AMD can try all they are able to and they're still not going to break 30% market share for years. Navi is looking good for the new consoles, though. That should make everyone happy as Polaris ended up holding up game design a bit in the last few years.
Ironically, 5700 is a better buy than 2060, but $350 is still far from mainstream. So that won't gain them market share.
Posted on Reply
#43
efikkan
bug, post: 4090952, member: 157434"
Well, calling 10nm "first generation" is a stretch…
That's how Intel describes it. They usually call it a new "generation" when they change the major parameters of the node, like gate pitch etc. Intel has called their 10nm plans "too aggressive", so I wouldn't be surprised if they grow some parts in 10nm+/++, like they have done in 14nm+/++. This doesn't mean the effective density will increase though, but it will help yields and leakage.

What they have done so far between the "disastrous" Cannon Lake and the "okay" Ice Lake-U/-Y is probably mostly tweaking the design to fit the characteristics of the node. These experiences has been used to develop the 10nm+ node.
Posted on Reply
#45
Frick
Fishfaced Nincompoop
ratirt, post: 4090652, member: 165024"
Are you saying that used cards are louder than new cards even if it's the exact same kind of a card? Cause it does sounds like it and I can't see any difference aside for the used cards' warranty is shortened.
What? No. What? I'm saying loud cards are universally bad. The 290X reference could be had for little money (used duh) but I wasn't interested in it because it's a loud card. The same goes for RX480/580.
ZoneDymo, post: 4090696, member: 66089"
Ermm you might want to put on less rose coloured glasses.
As far as I can recall both have had perfectly fine stock coolers for a long time, with some outliers on both sides.
"Perfectly fine" is a lot more inclusive than "excellent" and "outstanding piece of equipment". I don't have a decades worth of data in my head but the phrase "quiet cooling" as applied to AMD reference cards is not something I have often seen.
Xaled, post: 4090721, member: 158027"
Reference cards that are more expensive than custom cooled cards?!
FEs are not reference cards, they are more expensive even than many custom cooling cards.
AMD made these coolers just to let the new cards have as competitive price as possible. And Nivida on the other way used the "excellent cooling" just to make their cards as expensive as possible
Is the first sentence a statement or a question? Doesn't matter, the performance/currency ratio have to be insane for me to be fine with a loud card. And a "competitive price" would be like 20% cheaper than the AIB cards, but often those are cheaper than the reference cards, so vOv.

I just dislike loud cards because they clutter up the market. I really don't understand why that is so difficult to fathom, and also I don't understand why AMD has such a hard time designing properly good cooling. Sometimes it happens, but it seems to me they for quite a while now has used exactly the same design for many of their cards and all of those are loud.
Posted on Reply
#46
lexluthermiester
bug, post: 4090657, member: 157434"
AMD's answer to Turing is Navi.
I don't think so. Many, including me thought it was, but it seems Navi is just an answer to a new generation of GPU's. AMD is still working on their RTRT offerings. My guess is that they want to do a better launch than the RTX launch. While I think the RTX launch was successful, no one can deny that it could have been better. AMD is not done, as Ryzen has proven, they just getting started.
Posted on Reply
#47
Mephis
lexluthermiester, post: 4091059, member: 134537"
I don't think so. Many, including me thought it was, but it seems Navi is just an answer to a new generation of GPU's. AMD is still working on their RTRT offerings. My guess is that they want to do a better launch than the RTX launch. While I think the RTX launch was successful, no one can deny that it could have been better. AMD is not done, as Ryzen has proven, they just getting started.
AMD isn't done, but neither is Nvidia. Navi is their response to Turing. It doesn't have RTRT yet, but that doesn't mean it isn't their response. We have heard that both consoles have RTRT and are both based of Navi, so that means that there is a version of Navi with it. The problem AMD faces is that Nvidia isn't standing still and is a node behind, yet still ahead in performance and features. AMD releases big Navi with RTRT next year, possibly pulling even with Nvidia and Nvidia releases their new architecture (Ampere?) with 2nd gen RTRT and a node shrink and pulls ahead again. At this point until AMD pulls of something huge, it will be a vicious cycle.
Posted on Reply
#48
ratirt
lexluthermiester, post: 4091059, member: 134537"
I don't think so. Many, including me thought it was, but it seems Navi is just an answer to a new generation of GPU's. AMD is still working on their RTRT offerings. My guess is that they want to do a better launch than the RTX launch. While I think the RTX launch was successful, no one can deny that it could have been better. AMD is not done, as Ryzen has proven, they just getting started.
I think same. AMD is just starting. Ryzen has taken a lot from AMD and we have to keep in mind that AMD is one company that has CPUs and GPUs competing with other two giants on gpu and cpu front. The Ryzen is successful this means (maybe) AMD will now focus more on the NAVI front. I surely hope so. I need a new card and I want Navi :D

Mephis, post: 4091075, member: 186806"
AMD isn't done, but neither is Nvidia. Navi is their response to Turing. It doesn't have RTRT yet, but that doesn't mean it isn't their response. We have heard that both consoles have RTRT and are both based of Navi, so that means that there is a version of Navi with it. The problem AMD faces is that Nvidia isn't standing still and is a node behind, yet still ahead in performance and features. AMD releases big Navi with RTRT next year, possibly pulling even with Nvidia and Nvidia releases their new architecture (Ampere?) with 2nd gen RTRT and a node shrink and pulls ahead again. At this point until AMD pulls of something huge, it will be a vicious cycle.
I wouldn't compare console world to PC. They are similar but the difference is big here. RTRT navi for consoles can work but for PC not necessarily. IT is obvious that for console gaming, less resources is required and all of the games are made exactly for the console and what it has inside exclusively. With PC it is different.
I'm not quite sure that AMD's main goal is to release something to counter NV. Well of course they will have to counter it somehow but AMD is bringing new GPUs to the market cause it is also GPU company not CPUs only. They need to move forward. I think they are more concerned about customer expectations more and new release cause they are GPU producers. They are obligated to release something and if it counters NV that's a bonus. :)
We will have to see what AMD will come up with. I dont think they are releasing Ray Tracing card with NAVI. Maybe a refresh later on with RT based on NAVI but this NAVI launch will be without RT for sure. Console first with RT to see how things go and then focus on PC. Besides they still need to catch NV with the performance and adding RT feature, is not going to accomplish this.
Posted on Reply
#49
Xaled
Frick, post: 4091016, member: 23907"
Is the first sentence a statement or a question?
A declarative quistion or "quistioning statement"


Frick, post: 4091016, member: 23907"
Doesn't matter, the performance/currency ratio have to be insane for me to be fine with a loud card. And a "competitive price" would be like 20% cheaper than the AIB cards, but often those are cheaper than the reference cards, so vOv.

I just dislike loud cards because they clutter up the market. I really don't understand why that is so difficult to fathom, and also I don't understand why AMD has such a hard time designing properly good cooling. Sometimes it happens, but it seems to me they for quite a while now has used exactly the same design for many of their cards and all of those are loud.
Yes but if custom cooled version have the same price then it is a whole another story. Nvidia released their new "super" cards just based on referenced 5700s. AMD didn't want to play all cards initially or just faked a punch and probably waited for Nvidias answer (Super) then hit the first punch with a price drop and now the second punch would be custom cooled cards that start at the same price .

Amazing fake and 1-2 punch by AMD imo.
Posted on Reply
#50
Mephis
ratirt, post: 4091130, member: 165024"
...

I dont think they are releasing Ray Tracing card with NAVI. Maybe a refresh later on with RT based on NAVI but this NAVI launch will be without RT for sure. Console first with RT to see how things go and then focus on PC. Besides they still need to catch NV with the performance and adding RT feature, is not going to accomplish this.
You might be right about this. My only point is that Navi with RTRT exists. The only downside that I see to this strategy is that consoles won't be out until holidays of 2020. That would mean, no AMD RTRT for pc until 2021 at the earliest. That gives Nvidia basically almost 2+ years to establish a dominant standard that AMD would have to play catch up to. We are seeing how hard that is with Cuda.
Posted on Reply
Add your own comment