Consumer 7nm GPUs from AMD are late 2019 to early 2020

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

prtskg

Senior member
Oct 26, 2015
261
94
101
Hi. Some BIG GUY said AMD is preparing 12nm Polaris refresh NEXT MONTH. I decided not post new thread about this news because it's not confirmed by any other site just by a BIG GUY at that forum, means it might not 100% accurate. Stay tuned.:p
https://translate.google.com/translate?sl=auto&tl=en&js=y&prev=_t&hl=en&ie=UTF-8&u=https://www.chiphell.com/thread-1911969-1-1.html&edit-text=
Whether this post came first or this linux patch -
https://www.phoronix.com/scan.php?page=news_item&px=AMD-Polaris-2019-0x6FDF
 
  • Like
Reactions: ryan20fun

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
AMD has cards that are competitive Up to GTX1080 territory, they are also not in a rush to launch anything new and spend resources they would other wise put for 7nm cards.
So both are sitting comfortable on their ass waiting to see how consumers will react to new high prices of the RTX cards , gathering data for the new 7nm cards that will follow the next 12-18 months.

Competitive from a performance perspective? Yes. Competitive from a profit standpoint, not even close. The Vega 60 and 56 FE cards, once fully packaged, probably cost AMD close to $200 more than GTX 1080, 1070 TI, and 1070. Nvidia is also sellingGP104 in products Vega cannot be in - laptops. (Big Vega dies are not sold in laptops).

Vega might be making a few sales in professional markets, but Nvidia has products that can beat Vega in nearly every metric, whether with GP104, GP102, GP100, GV100, and now TU104 and TU102. Vega is outgunned on every level it competes, is limited to it's scope of products, and is only exceptional in very limited use cases.

Vega alone isn't good enough to justify it's R&D. Simply having a product isn't good enough when it can't make much, if any, money. Hopefully the foundations for Vegas successor have been laid and AMD can pull a Kepler-like return to competitiveness. (I realize Kepler fizzled after 18 months but was great at launch).
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
A possible middle ground is to ignore the extreme top end and just concentrate on the low-end to high mid-range where most of the products are sold. Not the best, but certainly not the worst situation. They have done this and its profitable for them.

I guess, but in AMD's case they are laboring with hardware that has perf/watt and base cost problems vs. NV who has a perf/watt lead and has price flexibility due to significantly higher margins on their hardware. Also it's a little hard to say that they are concentrating on the low-end when all they did was rehash Polaris in the 500X series. Unless they really are releasing 12nm Polaris, then things could be a little interesting.

Ngg path was driver specific and ought to have worked without developer support.
Some of the same features as in maxwell btw.
So its not a small deal its not working. Look at the raw shader numbers to give you an idea of the potential.

Preachin to the choir there. NGG path is b0rked though, as AMD has come out and said in response to the guy trying to enable it in hardware (see other thread). I have no expectation that we'll see it working outside of a). 7nm Vega and b). Navi, since AMD has punted on 12nm Vega.
 

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
Irrelevant to the topic at hand. Zen's success does not preclude AMD from continuing to serve the consumer dGPU market. At. All. They maintained an entirely separate RTG division under Raja Koduri while developing Zen in the first place. Continued work on Zen+ and Zen2 could have continued while RTG refined Vega and moved it to GF's 12nm process, which according to many is not much more than a refined 14nm process anyway . . .

It has everything to do with the topic at hand, unless taken out of context like you did. The point was as i'm sure you know, that Zen's success proves that it was the right choice to balance their R&D in the way that they did. If that meant taking some resources from the GPU side even if their consumer GPUs weren't at the top of the charts then that's what they had to do. Focusing on 7nm compute and building on their strengths where they are still ahead of the nvidia's latest cards in many cases seems like the more important task at hand. 7nm brings major improvements to PPP.
 

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
Maybe AMD has been too busy WIPING THE FLOOR OF INTEL WITH RYZEN AND EPYC to bother with Nvidia.

Just my OPINION

Zen has been a great return to competition, and I expect Zen 2 to be even more so. I look forward to my first AMD CPU purchase (next year) in over a decade.

But, I don't think AMD's GPU and CPU timetables massively affect one another. After all, it's highly unlikely CPU and GPU engineers are regularly (if at all) swapping out with each other.

All that said, while it's entirely conjecture on Happy Medium to say AMD cancelled a GPU, he is correct in pointing out the time between Navi and Polaris will be historically the longest gap between new generations of GPUs. I lay blame on the lack of cutting edge node availability for high performance processors (hence the reason for fat, power hog Turing GPUs).... But I digress.
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,248
136
I lay blame on the lack of cutting edge node availability for high performance processors

I'm sure that plays an important part in the grand scheme of things in the end. The market ignoring their products and using them for nothing more than pricing pressure in the past I'm sure played a larger role in the decision to sit it out. You guys are naive if you think AMD employees don't hang out on the web after hours viewing reviews, arguments, goal post moving, etc after they launch a product. I'd imagine they're somewhat tired of reading all the too little, too late, can barely match a XXXX and have somewhat decided is it even worth it to try. Put yourself in their shoes and look back at the launches and view the old threads and maybe you'll understand what I said....Not aimed at you tviceman but at all whom read.
 

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
I'm sure that plays an important part in the grand scheme of things in the end. The market ignoring their products and using them for nothing more than pricing pressure in the past I'm sure played a larger role in the decision to sit it out. You guys are naive if you think AMD employees don't hang out on the web after hours viewing reviews, arguments, goal post moving, etc after they launch a product. I'd imagine they're somewhat tired of reading all the too little, too late, can barely match a XXXX and have somewhat decided is it even worth it to try. Put yourself in their shoes and look back at the launches and view the old threads and maybe you'll understand what I said....Not aimed at you tviceman but at all whom read.

I don't believe that forum arguments drive AMD product cycles one bit. Not even in the slightest. Both sides shift goal posts all day long to fit their respective narratives. Happens all the time.
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
It has everything to do with the topic at hand, unless taken out of context like you did. The point was as i'm sure you know, that Zen's success proves that it was the right choice to balance their R&D in the way that they did. If that meant taking some resources from the GPU side even if their consumer GPUs weren't at the top of the charts then that's what they had to do. Focusing on 7nm compute and building on their strengths where they are still ahead of the nvidia's latest cards in many cases seems like the more important task at hand. 7nm brings major improvements to PPP.

I am not taking anything out of context. I'm saying that you're wrong. AMD had the R&D resources to do 12nm Vega in addition to everything else they are doing right now with Rome and Matisse. Zen/Zen+/Zen2 poached resources from the old CMT teams, not from RTG. AMD was able to develop Zen and Polaris/Vega simultaneously, so how are they now somehow unable to port Vega over to 12nm and tweak it a bit while working on Zen+ and Zen2? Impossible, especially with the enhanced revenues starting in 2017. It is fallacious to assume that AMD is that strapped for resources.

AMD has simply pulled out of the consumer market. They know they can sell all their cutting-edge GPUs (Vega, Vega 20) on the pro market, so there's no reason for them to sell those products to consumers. I have long suspected that the Vega shortages had less to do with HBM availability and miners than it did the pro market. So I understand why the pro market is getting Vega 20 and we aren't, I get it. But the pro market would never have wanted 12nm Vega refresh, certainly not with 7nm Vega 20 on the horizon. GF has the 12nm capacity to help AMD produce some more GPUs aimed squarely at the consumer market. It was all a matter of whether AMD had the desire to serve the market. Expanding production brings with it risks, and AMD simply did not want to take those risks.

If the rumors are true, then AMD is going to throw us a bone in the form of 12nm Polaris. I can understand why they might rehash Polaris since there is no need for them to find a new source of HBM2 for those cards (12nm Vega would still require HBM2). For people in the midrange, that might be just what the doctor ordered until Navi10 comes out. But for those of us looking for an upgrade from AMD to replace our Vega FEs and RX Vega 64s, there is nothing.
 
  • Like
Reactions: Headfoot

DooKey

Golden Member
Nov 9, 2005
1,811
458
136
I am not taking anything out of context. I'm saying that you're wrong. AMD had the R&D resources to do 12nm Vega in addition to everything else they are doing right now with Rome and Matisse. Zen/Zen+/Zen2 poached resources from the old CMT teams, not from RTG. AMD was able to develop Zen and Polaris/Vega simultaneously, so how are they now somehow unable to port Vega over to 12nm and tweak it a bit while working on Zen+ and Zen2? Impossible, especially with the enhanced revenues starting in 2017. It is fallacious to assume that AMD is that strapped for resources.

AMD has simply pulled out of the consumer market. They know they can sell all their cutting-edge GPUs (Vega, Vega 20) on the pro market, so there's no reason for them to sell those products to consumers. I have long suspected that the Vega shortages had less to do with HBM availability and miners than it did the pro market. So I understand why the pro market is getting Vega 20 and we aren't, I get it. But the pro market would never have wanted 12nm Vega refresh, certainly not with 7nm Vega 20 on the horizon. GF has the 12nm capacity to help AMD produce some more GPUs aimed squarely at the consumer market. It was all a matter of whether AMD had the desire to serve the market. Expanding production brings with it risks, and AMD simply did not want to take those risks.

If the rumors are true, then AMD is going to throw us a bone in the form of 12nm Polaris. I can understand why they might rehash Polaris since there is no need for them to find a new source of HBM2 for those cards (12nm Vega would still require HBM2). For people in the midrange, that might be just what the doctor ordered until Navi10 comes out. But for those of us looking for an upgrade from AMD to replace our Vega FEs and RX Vega 64s, there is nothing.

Interesting take on the situation at AMD. If true this might explain what we're seeing with RTX prices. I'm sure NV knows what AMD is up to and vice versa. I'm going to have to chew on this a bit.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
It is fallacious to assume that AMD is that strapped for resources. I can understand why they might rehash Polaris since there is no need for them to find a new source of HBM2 for those cards

Being strapped and making a profit are 2 different things. I'm skeptical about the polaris on 12 nm rumor.Simply not worth it to do a shrink, RX 580 is running at the limit already, especially also memory bottleneck. Higher clocks wouldn't add much. It would need to be a completely new die. A bigger die and gddr6 would be needed. And now you are making essentially a new chip so why use old polaris uarch and not vega?

Interesting take on the situation at AMD. If true this might explain what we're seeing with RTX prices. I'm sure NV knows what AMD is up to and vice versa. I'm going to have to chew on this a bit.

As you say. Vice-versa. If AMD new NV doesn't have a new mainstream product in the GTX 1060 / RX 580 bracket, AMD also doesn't need anything new and better to invest money in zen 2.
 
Last edited:

SirDinadan

Member
Jul 11, 2016
108
64
71
boostclock.com
AMD will not bother with a Polaris rehash unless the new card can match a GTX 1070 at a much reduced price, at least -15% cheaper. Of course I don't have data, but if you look up a "Which GPU should I buy" forum, many clearly state that the person is not interested in AMD solutions. So to have a fighting chance, AMD needs to offer the same sort of performance but with a hefty price-cut.

On a different note, AMD failed to create a Tonga like performer. There's a relatively huge gap between the 1050Ti - 1060 and the the RX560 and 570/580. According to my benchmarks the RX560 would perform much better with more BW so GDDR6 could bring some easy gains to that chip. AMD should focus on mainstream and entry level as the high-end is pretty much lost, don't think that the current tech can scale-up well, just look at Fiji and Vega and how efficiency went out of the window with Polaris as well.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Being strapped and making a profit are 2 different things.

Then let me clarify: AMD would make a profit selling 12nm Vega parts as a stopgap between RX Vega and Navi20 or Navi30 (or whichever Navi is meant to finally replace RX Vega). There would be risks involved, but the odds are fairly strong in AMD's favor. They would still make a profit, even if they had to put together a deal to move a limited production run of HBM to GF (assuming Hynix and Samsung are strapped for capacity).

They had the means, the motive, and the opportunity. They didn't take it.

I'm skeptical about the polaris on 12 nm rumor.Simply not worth it to do a shrink, RX 580 is running at the limit already, especially also memory bottleneck. Higher clocks wouldn't add much. It would need to be a completely new die. A bigger die and gddr6 would be needed. And now you are making essentially a new chip so why use old polaris uarch and not vega?

RX580 may be bottlenecked on the RAM side, but moving to 12nm should up clocks for both GPU and memory. It doesn't need to be a completely new die. But then I'm not the one promulgating the rumor, so I don't know all the details (if any).

AMD also doesn't need anything new and better to invest money in zen 2.

Why is anyone assuming that AMD needs to/wants to move funding between RTG and the Zen design team? Zen 2 development effort is already fully-funded. They're hitting their targets on Rome, and personally I expect to see Matisse in April-June (hopefully April). Adding more cash does nothing for them, why keep bringing it up? AMD is not limited on the design-side anyway. Months ago, people started getting nervous about 7nm Zen2 from GF hitting the market on time due to delays at GF. Now that GF is out of the leading edge fab business, the fear is that certain 7nm CPU/APU products might be delayed due to the transition to TSMC. In both cases, potential delays for Zen 2 were related to external factors - namely, fabs. And the real breadwinner - Rome - was never delayed.

AMD already put all their eggs in the Zen basket. That basket has enough eggs. It does not need more.
 
  • Like
Reactions: neblogai

Glo.

Diamond Member
Apr 25, 2015
5,661
4,419
136
Why is anyone assuming that AMD needs to/wants to move funding between RTG and the Zen design team?
Because in order for AMD to design and make Zen architecture they HAD to pull resources from a lot of RTG projects. Now its good time it will pay dividends both ways.

Funniest part: a lot of Zeppelin designers are in this very moment at RTG. For example Suzanne Plummer went to RTG.
 
Mar 11, 2004
23,031
5,495
146
I am not taking anything out of context. I'm saying that you're wrong. AMD had the R&D resources to do 12nm Vega in addition to everything else they are doing right now with Rome and Matisse. Zen/Zen+/Zen2 poached resources from the old CMT teams, not from RTG. AMD was able to develop Zen and Polaris/Vega simultaneously, so how are they now somehow unable to port Vega over to 12nm and tweak it a bit while working on Zen+ and Zen2? Impossible, especially with the enhanced revenues starting in 2017. It is fallacious to assume that AMD is that strapped for resources.

AMD has simply pulled out of the consumer market. They know they can sell all their cutting-edge GPUs (Vega, Vega 20) on the pro market, so there's no reason for them to sell those products to consumers. I have long suspected that the Vega shortages had less to do with HBM availability and miners than it did the pro market. So I understand why the pro market is getting Vega 20 and we aren't, I get it. But the pro market would never have wanted 12nm Vega refresh, certainly not with 7nm Vega 20 on the horizon. GF has the 12nm capacity to help AMD produce some more GPUs aimed squarely at the consumer market. It was all a matter of whether AMD had the desire to serve the market. Expanding production brings with it risks, and AMD simply did not want to take those risks.

If the rumors are true, then AMD is going to throw us a bone in the form of 12nm Polaris. I can understand why they might rehash Polaris since there is no need for them to find a new source of HBM2 for those cards (12nm Vega would still require HBM2). For people in the midrange, that might be just what the doctor ordered until Navi10 comes out. But for those of us looking for an upgrade from AMD to replace our Vega FEs and RX Vega 64s, there is nothing.

They might have had the resources but the return for doing that likely was not enough to warrant it and they'd be better off putting those resources towards other products. I think it is as simple as that. Consumer gaming cards don't make enough money for AMD for them to chase it right now, instead focus on making sure they capitalize when their product stack is better. Its why their consumer APUs are also languishing. They just don't make enough money, so its better to save those resources (namely money) towards next gen products, where they should be more competitive. AMD could certainly stand to be prepared for a GPU launch for a change, and they stand to gain a lot more by getting a better situation right than trying to sort out some attempt at salvaging consumer Vega for gamers.

No they haven't. If they have then Nvidia has almost as much, as the RTX cards are Nvidia's pro chips, and they never even released a consumer Volta and might not have new mainstream cards before Navi. They rode Pascal for over 2 years and possibly 3 for the mainstream chips. You're ignoring that the entire industry has slowed because of declining advancements in production processes. So low margin markets are going to slow down.

Well, duh. Vega 20 is not a gaming GPU, it'd be stupid to try and sell it as one, especially if you have a market where you can make more money with the same chip. It probably would be better than Vega 64, but it'll still be stuck, limited by the basic setup of the GPU (and that its locked at the same number of shaders/CUs as Vega 64), plus it'll have a significant portion of the chip wasted on HPC/compute focused pieces (that if they could use for ray-tracing or inferencing, would be tacked on and probably not be impressive), so it could also be somewhat unimpressive in perf/W for gaming.

I'm not sure about that. I think it simply is that AMD had little margin on Vega in the consumer space, so it just wasn't worth it. They just plain were not going to make a ton of money from consumer gaming Vega (until crypto-mining, but then them upping production was likely more throwing a bone to AIBs and their channel partners so that they wouldn't abandon them as they go through this down period). And because it wasn't going to make much money management chose to stop trying to get its broken features working, and instead move on. Absolutely they'd be better off selling every Vega they could to the pro market, as its the same chip, HBM would be more expensive some but at a much higher price, its a no brainer. The thing is, I don't think Vega should have ever been a consumer gaming card (as it was released, if they could have made a GDDR5X version, or maybe gone with 3 stacks of HBM or something, then maybe, but the 2 stacks/2048 HBM was a step back and a costly one). They lucked out that crypto-mining took off and they sold all of them they could, as otherwise they were going to have to do something about prices (pushing their margins even lower). Now, we'll probably never know what went wrong with Vega. Was the design inherently broken? Was it the chip as produced (and AMD couldn't afford to rework it more, so it was produce it as it was or kill it)? Was it the process? Was/is it the software (I have a hunch Raja was trying to get their software stack reworked like Nvidia had done, where they could implement some features like the geometry discard via software, and possibly even backport them to older GCN like Polaris, which could have pretty big benefits)? I'm pretty sure I've posted this whole line of thought before. My guess is its a confluence of things. Hopefully we see things swing back the other way and Navi somehow is a modern 4000 series level shift.

If there is a 12nm Polaris, its likely more about appeasing OEMs than offering a real worthwhile improvement for gamers. If Polaris is pretty bandwidth starved, going to maybe GDDR5X (I think GDDR6 would be too expensive for that level of card right now) might help a bit as well. So if they get an extra 10% performance and 10-15% perf/W it'd be pretty nice. Because 12nm Vega wouldn't have offered much for them to care about. So there was no point to making it. It wouldn't offer gamers much either. It might be a bit more efficient but it might end up like the 580 and be worse but with performance might go up a bit (possibly with higher speed HBM to get back to Fury bandwidth).

I doubt 12nm Polaris offers enough to get Polaris owners to buy. Maybe the bulk of those goes to OEMs, but they bin a fair amount of good chips for AIBs to put on their custom cards, and they can maybe offer something closer to 20% performance uptick overclocked, but I'm doubtful that its going to be much above RX580.

I get it, people are frustrated. There isn't much of anything really worthwhile to buy (people likely upgraded years ago, be it to Polaris or Pascal and there just isn't much for them to upgrade to yet). And the few ones that are, are expensive. And Navi might not even fix that for people on Vega. And who knows what Nvidia is doing/planning.

Being strapped and making a profit are 2 different things. I'm skeptical about the polaris on 12 nm rumor.Simply not worth it to do a shrink, RX 580 is running at the limit already, especially also memory bottleneck. Higher clocks wouldn't add much. It would need to be a completely new die. A bigger die and gddr6 would be needed. And now you are making essentially a new chip so why use old polaris uarch and not vega?



As you say. Vice-versa. If AMD new NV doesn't have a new mainstream product in the GTX 1060 / RX 580 bracket, AMD also doesn't need anything new and better to invest money in zen 2.

I believe that it doesn't require much reworking to go from 14nm->12nm. I think there was even some comments that they could directly port it and it'd come out ok, with a 5-10% improvement in perf/W. If they tweak it at all they could possibly eke out more like with Ryzen. Which, maybe there is more room and they go with GDDR6 and get a 25% uptick over RX580. I'm pretty skeptical that we see that. I have a strong hunch that its more about just simply supplying OEMs. I'd guess that 12nm is just GF's new 14nm (meaning, there isn't 14nm as it was when Polaris came out, so continuing production of Polaris chips means it has to be 12nm). So people reading that this is some significant reworking and will be a market improvement I think are going to be disappointed. It should be a bit better, but I'm not expecting more than 10%. I would love to be surprised though.

AMD will not bother with a Polaris rehash unless the new card can match a GTX 1070 at a much reduced price, at least -15% cheaper. Of course I don't have data, but if you look up a "Which GPU should I buy" forum, many clearly state that the person is not interested in AMD solutions. So to have a fighting chance, AMD needs to offer the same sort of performance but with a hefty price-cut.

On a different note, AMD failed to create a Tonga like performer. There's a relatively huge gap between the 1050Ti - 1060 and the the RX560 and 570/580. According to my benchmarks the RX560 would perform much better with more BW so GDDR6 could bring some easy gains to that chip. AMD should focus on mainstream and entry level as the high-end is pretty much lost, don't think that the current tech can scale-up well, just look at Fiji and Vega and how efficiency went out of the window with Polaris as well.

Strongly disagree with your first point. You're ignoring that they need to offer something to OEMs. They don't even have to offer something better than the performance those cards already offer, they simply need to have chips/cards available for OEMs. If they can make some improvements great, and certainly they'd likely be wanting to lower their cost so as to maintain or improve margins, but there's almost no chance they're targeting 1070 performance. They aren't going to hit that without making a pretty significant change to Polaris, which they aren't going to do right now as they're doing that for Navi.

Meh, we've been seeing that type of nonsense for over a decade. Granted yes, mindshare does matter, but plenty of the people saying that type of stuff know jack about anything so anecdotes like that aren't worth anything. Well unless you're paid by NVidia to post it (not that's not an accusation towards you, merely referencing that type of thinking came from them doing that, and that they are still doing that).

I find your argument to be quite bizarre. After you just said they need to massively increase the performance while substantially reducing cost you're talking about them putting GDDR6 on a budget video card? Unless GDDR6 let them go with like half the chips (and thus could reduce the packaging and/or other components on the card), it'd probably be too expensive for such a card. I'm not even sure GDDR5X would be worth it (maybe for the top end card).

Which is what they were/are doing. Vega was them selling what was definitely more of a professional/compute chip to gamers and it suffered badly for it (because the compute people don't need as strong of graphics processing so AMD basically went with what they already had while enabling reduced precision capability and then trying to implement a substantial geometry discard capability to make up for not addressing the base blocks of the GPU; and then HBM made pricing difficult for a consumer card, and going with 2 stacks caused bandwidth to decrease compared to what Fury offered). Its why they knew it'd be smarter to lead with Polaris (as they simply needed the revenue and its where they make their money in GPUs). They aren't doing any scaling of current tech, they've moved on. Vega 20 is HPC focused (holding the graphics part the same as Vega 64) and Navi is a Polaris level chip, being based off consoles (which helps AMD, as it helps developers in transitioning between console and PC).

How did efficiency go out the window with Polaris? It wasn't spectacular, but it was more efficient (and more performance per shader/CU count went up compared to previous GCN). Certainly Nvidia saw a bigger improvement, but that raises the question about foundry process and how much that played a role. Or are you talking about AMD pushing clocks out of the efficient range in order to try and offer a bit more performance? But again, that brings in how much of that was the process versus AMD's actual design.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
This is just a speculative post:

Vega 20 w/ GFX90x => 4096 ALU w/ 32 GB HBM2 => Q1 2019
Vega 21 w/ GFX95x => ~3000 ALU w/ anything less than 24 GB GDDR6 => Q2/Q3 2019

From Polaris 20 w/ approx $200 to Vega 21 w/ approx $700

Navi is delayed and getting potentially fused with Next-gen.

The revised Vega w/ RX nomenclature can implement Pathtracing in the same way as PowerVR GPUs can do for Raytracing.
>100 Gigarays?
 
Last edited:

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Why is anyone assuming that AMD needs to/wants to move funding between RTG and the Zen design team? Zen 2 development effort is already fully-funded. They're hitting their targets on Rome, and personally I expect to see Matisse in April-June (hopefully April). Adding more cash does nothing for them, why keep bringing it up?

Well obviously I meant that when this decision was made like 2-3 years ago. Now we can see the opposite is happening, CPU people moving to RTG.

I believe that it doesn't require much reworking to go from 14nm->12nm. I think there was even some comments that they could directly port it and it'd come out ok, with a 5-10% improvement in perf/W.

That is exactly what I question. Save power? Yes that would be possible. Higher performance? Not really. RX 580 is already at it's limits. Higher clocks will be bottlenecked my memory BD in many scenarios and memory clocks are already sky high. Changing memory type means new controller means new chip means new masks which means >$50 mio $ investment. Not going to happen.

I'd guess that 12nm is just GF's new 14nm (meaning, there isn't 14nm as it was when Polaris came out, so continuing production of Polaris chips means it has to be 12nm)

That is not how it works.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Because in order for AMD to design and make Zen architecture they HAD to pull resources from a lot of RTG projects.

Such as?

Furthermore, that happened (as @beginner99 said) years ago, in a time period when they still had enough resources to work on Vega at the same time that they worked on Zen. The decision to move Vega to 12nm for the consumer market would probably had happened in 2017, if it were to happen at all.

Just because they killed some other project at RTG in 2015 or so, does not mean that they'd be so over a barrel in 2017 that a simple move of Vega to 12nm would have been impossible. With or without fixed ngg path.

No they haven't. If they have then Nvidia has almost as much, as the RTX cards are Nvidia's pro chips, and they never even released a consumer Volta and might not have new mainstream cards before Navi. They rode Pascal for over 2 years and possibly 3 for the mainstream chips.

I see NVidia as being in a precarious position. They're trying to milk the hell out of their robust pro/AI market sales and constantly testing the waters to see how much they can get consumers to keep paying for cards. If NVidia has abandoned any market, it's the market currently served by the 1060.

You're ignoring that the entire industry has slowed because of declining advancements in production processes. So low margin markets are going to slow down.

I sort of expected things to slow down, and the long ugly slog through 28nm territory was not pretty. Once we finally hit 16nm/14nm things started looking up. The new processes are there, and there's plenty of room for both companies to innovate in the 12-16nm range. I expected more slowdown with a number of products on those processes until 7nm was finally ready. NV seems to have been locked out of early runs of 7nm wafers from TSMC, and AMD just doesn't want to bother with anything but pro 7nm cards, so there you have it. At least NVidia is still releasing some "consumer" cards on 12nm, even though they are horribly overpriced. AMD did not show up to the fight.

Well, duh. Vega 20 is not a gaming GPU, it'd be stupid to try and sell it as one, especially if you have a market where you can make more money with the same chip. It probably would be better than Vega 64, but it'll still be stuck, limited by the basic setup of the GPU (and that its locked at the same number of shaders/CUs as Vega 64), plus it'll have a significant portion of the chip wasted on HPC/compute focused pieces (that if they could use for ray-tracing or inferencing, would be tacked on and probably not be impressive), so it could also be somewhat unimpressive in perf/W for gaming.

It's still 7nm Vega. Lower power, higher clocks, the whole shebang. They could just disable the HPC bits or spin a cut-down die and throw that bone to the market. Hell an optical shrink of Vega with fixed ngg would have been fine. Even on 12nm. So I think it would be no more stupid than AMD trying to sell RX Vega in the first place. If you are going to do one, may as well do the other.
 
Last edited:
  • Like
Reactions: coercitiv

SirDinadan

Member
Jul 11, 2016
108
64
71
boostclock.com
You're ignoring that they need to offer something to OEMs. They don't even have to offer something better than the performance those cards already offer, they simply need to have chips/cards available for OEMs.
OEMs are usually fine with a rebadge, up the card designation number and be done with it.

Meh, we've been seeing that type of nonsense for over a decade. Granted yes, mindshare does matter, but plenty of the people saying that type of stuff know jack about anything so anecdotes like that aren't worth anything.
I can only speak about what I can experience. Let me highlight the mindshare thing, just check out the very first response the guy gets. And BTW, it was me how called out the guy on the matter and therefore he edited his post - so yeah, sure, I'm on NV's payroll.

AMD pushing clocks out of the efficient range in order to try and offer a bit more performance?
Yep, exactly. IMHO they couldn't believe their eyes when they saw the performance of the Pascal cards and had to act to get the RX cards into the same ballpark performance-wise.

All in all, I think it's time to go back to the drawing board for AMD to come up with a heavily redesigned GCN or something completely new. And I nearly forgot: ECOSYSTEM, Radeon is so behind, it is unbelievable. At least their open-source driver stack is shaping up nicely, although there are still lot's of issues according to Phoronix's recent benchmarks.
 

badb0y

Diamond Member
Feb 22, 2010
4,015
30
91
The day AMD takes the performance crown is the day I set my G-Sync monitor on fire and dance around it.
 

Headfoot

Diamond Member
Feb 28, 2008
4,444
641
126
I feel like if AMD had anything remotely competitive, remotely soon to the Turing release we would have seen intentional "leaks" to take a little wind out of the sails, as we have seen many times before. That we see no credible rumors or even really any rumors at all about a Turing competitive AMD GPU tells me they have nothing close to release.
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,744
3,077
136
I feel like if AMD had anything remotely competitive, remotely soon to the Turing release we would have seen intentional "leaks" to take a little wind out of the sails, as we have seen many times before. That we see no credible rumors or even really any rumors at all about a Turing competitive AMD GPU tells me they have nothing close to release.
Just like all the rome leaks :)
 

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
Amd have to use gf capacity or pay a fine.
Gfx seems like the least painfull way all options considered.