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

[DrMrLordX]

So wccftech is claiming they have confirmed AMD will release a consumer GPU at CES.

https://wccftech.com/amd-launching-ryzen-3000-series-cpus-apus-radeon-gpu-ces/

It's wccftech so it's most likely clickbait but it is interesting. It's (supposedly) going to be a 7nm variant of Vega 20. It would be really nice and it could be plausible as it would act as a 7nm pipe cleaner for moving to Navi. But seeing how AMD has flat out said not to expect a 7nm consumer GPU until after a few other things come out, I just don't believe its for real. If AMD announces any GPU at CES it will be the Pro version and not for gamers/consumers.

It might well be Vega 20 FE. You are wise to be skeptical. Vega20 was already a 7nm pipecleaner for them regardless of which market they serve with it, so there's no financial reason for them to serve the consumer market except to keep up a presence (I guess). They have no signaled anything other than Navi for the consumer market in any of their releases.

 

[darkswordsman17]

It might well be Vega 20 FE. You are wise to be skeptical. Vega20 was already a 7nm pipecleaner for them regardless of which market they serve with it, so there's no financial reason for them to serve the consumer market except to keep up a presence (I guess). They have no signaled anything other than Navi for the consumer market in any of their releases.

I doubt it. I think that's already released (just not in big fan fare as its not aimed at markets that needs that, they'll probably announce more general availability for it at some point, but I don't see them using CES to launch Vega 20, it is called consumer electronics show after all). I think its Navi. Lisa Su specifically mentioned they would be pushing things forward for gamers with their CES announcements. Other rumors said AMD was pushing for a 1Q release of Navi. If its a GPU it basically has to be Navi.

I'll be surprised if they really launch them though. I think it'll be announcements and touting different aspects. I even doubt soft launch (as in specific models/pricing setup), but that's possible. I think real availability will be probably March. Just a guess (based on nothing but that).

They could do Ryzen in a tiered way, where the early chips are 8 core ones (with Intel level clocks and IPC improvements) and that would get plenty of people excited, and should be pretty easy to keep compatible with 400 series boards. Then at Computex they launch the higher core count ones along with the 500 series boards (as well as Threadripper, which I wonder if they would market the high core count Ryzen chips as Threadrippers as well).

 

[DrMrLordX]

Navi at CES would surprise me. Near as I can tell, it isn't ready. I also don't see them doing a midrange release (Navi10 or whatever) with the 590X having been released so recently. Something like Navi10 wouldn't even outperform existing Vega64 cards.

My expectation is nothing at all from RTG at CES. Feel free to prove me wrong, AMD.

 

[jpiniero]

Lisa Su specifically mentioned they would be pushing things forward for gamers with their CES announcements.

Probably talking about Pinacle Ridge.

 

[darkswordsman17]

So for people not wanting to click on the link:

The tweet claims that Navi is not GCN eg. a new CU design which probably also gets away of the ROP limits and increases efficiency. That at least makes the leaks more believable as the power figures stated are way too good to just be another shrunk GCN part.

I'm not sure that's a major bombshell as I believe AMD has outright said Navi is a new GPU architecture. I might be wrong, but could swear they have themselves said it is not GCN and is a new architecture.

That's probably a better piece of news (assuming its true) than anything else that's been leaked so far. GCN was getting a bit long in the tooth. Hopefully the success of Ryzen will given AMD more to work with for Radeon R&D.

While the non locked cores/CU count setup is good news, I'm not sure the rest means all that much. Honestly, considering that Polaris was still decent for its market, GCN has shown it had ridiculously longevity and gets a bad rap (have been seeing people for years claim GCN is awful yes, outright horrible, which is just stupid; I'm not saying you are, but this is something that's been said since almost the start of GCN by some people). Heck, if they hadn't been stuck to GF, Polaris and Vega might've been outright competitive with Nvidia this last go around (so perf/w would've been closer to Pascal, and performance probably would've been more in line with the 590 from the get go as they probably could've had higher clocks). A lot of the stuff will carry over some and I think there's a good bit of blurring (stuff that changed between versions of GCN would've been typically been part of new architectures, and a healthy bit that goes into "new architectures" is still based on older stuff).

But they could use some changes, especially if it makes it easier for developers to get the potential out of their stuff. If the next consoles are just using standard CPU/GPU chips, that could bode well for them (as it means even less work between console/PC versions for developers).

Heck yes. Still droves of GPUs hitting from miners finally giving up. And if ETH prices tick up again, Lavasnow and other advanced ASICs can be ramped up to make the idea extinction, just as it did to Bitcoin mining.

AMD better have a plan to compete not with RTX 20xx though. Nvidia is almost certainly sandbagging with this release, due to oversupply of 10 series and depressed pricing support (only extreme bleeding edge types buy 20xx, and 1080ti makes 2080 price look iffy, ditto 1080 vs 2070, 1070ti vs 2060 probably).

But late 2019-2020ish? If RTX 20xx is a stopgap release to be followed by a major leap, as I suspect, then it will take an FX > Zen type of leap to get to relevance. I really hope it works out for them, because that helps everyone. Linus made the point that Intel during their long defacto monopoly didn't raise prices beyond inflation (actually got cheaper year by year considering $ value), yet Nvidia has been steadily cranking prices up gen upon gen.

I'm not sure what they'll do, the rumors are saying the pricing should be like Polaris. But if the performance is there, and they should have efficiency improvements over Pascal, so they could charge the same prices that Nvidia does for that stuff or even higher (so like $400 and below). But they seem to be looking to get marketshare (not a terrible idea). But that likely means lots of Polaris and Vega chips that practically can't sell once Navi hits. But maybe they're fine with that so that since a ton of those were miners, and figures hey if they dump them for $100 and less then great, boost for very strict budget gamers.

Their plan is likely "we've got Xbox and Playstation" and sounds like they'll be pushing APUs that are chips with a CPU and GPU module on them, which should help budget gamers a lot. I don't think AMD is planning on all the RTX like stuff Nvidia is doing, at least not in the consumer space. But they are planning for stuff beyond Navi. I really don't know how much direct competition Nvidia and AMD will be doing in the GPU space though if Nvidia keeps trying to specialize their hardware and AMD looks at mainstream value segments and for software to dictate how that gets used. Or maybe they'll develop some dedicated ray-tracing chip (or partner with someone else that makes one) and it'll be another module they can offer.

Nvidia's issue is they're designing their GPUs for non consumer stuff, and then trying to fit them back down to consumers. I'm not sure that's gonna work if it means prices keep going up like that. Er, actually I want to clarify. I think that is a smart move for the future, as I think gaming is headed to the cloud in a big way. But there will still be years before that happens, and I think that's where AMD's plans are right now (I think they've outright said that they'd do different chips for consumer and HPC markets as they just don't feel like their attempt at targeting both with one design like they had been doing).

I think if I were Nvidia, I'd not mind ceding the lowend/mainstream and let AMD and Intel duke it out where they'll be leveraging their ability to pair it with their CPUs. Keep pushing in HPC (and market the heck out of RTX for that as I think that's where it would work best), and then on the opposite end, push out a new Tegra with the modern standard ARM cores and keep at making your Shield consoles to prep for when simpler more "terminal" style becomes the dominant consumer gaming hardware. Make their own VR headset (with like a belt that would house some tracking stuff, the processor, and then battery for extended play time) I think that's what Nvidia sold Nintendo on. Where they control all the software (so no worry about piracy, and easier to manage the online system), plus cheaper, less complex hardware (with higher profit margins) for consumers. I think that's where Microsoft is headed. And will just be a matter of time before others (including the big publishers) as well.

 

[stuff_me_good]

If AMD would be smart they would introduce some AVX like instruction set for their new GPU design or at least some hairworks like black box type effect packet that is locket to the new core and mandatory to new consoles. This seems to be the only way they could compete nvidia and beat them in their own game.

But they are still playing the good guys card and getting stumped on. Not a viable business practice when you are the underdog.

 

[DrMrLordX]

If AMD would be smart they would introduce some AVX like instruction set for their new GPU design or at least some hairworks like black box type effect packet that is locket to the new core and mandatory to new consoles. This seems to be the only way they could compete nvidia and beat them in their own game.

But they are still playing the good guys card and getting stumped on. Not a viable business practice when you are the underdog.

Radeon Technologies Group has been a good corporate citizen in the past few years. They've provided valuable APIs as open source (think Vulkan, contributions to OpenCL/OpenCL 2.0) and have even open-sourced part of their driver stack (AMDGPU on Linux). Many of their HSA software projects were open-sourced and maintained by AMD employees on company time. And let's not forget TressFX (versus Hairworks). I would be profoundly disappointed were AMD to change their policies on open sourcing APIs and drivers.

 

[darkswordsman17]

Navi at CES would surprise me. Near as I can tell, it isn't ready. I also don't see them doing a midrange release (Navi10 or whatever) with the 590X having been released so recently. Something like Navi10 wouldn't even outperform existing Vega64 cards.

My expectation is nothing at all from RTG at CES. Feel free to prove me wrong, AMD.

I just realized you were meaning a Frontier Edition type card based on Vega 20. I don't think that is going to happen. Vega 20 is almost entirely for enterprise/compute/deep learning stuff, its not even aimed at like the Quado/FirePro CAD type of cards (AMD is leaving Vega 64 based stuff for that market til I think 2020 when the bigger version of Navi is due). That's the impression I got based on them classifying it as an Instinct card (that might just be rumor though, but think they've openly said that too, and said its already being used by their partners for that, hence I don't see them launching it at CES, as its already out).

It would surprise a lot of people, but based on what AMD has said I don't know what else it would be. They have repeatedly said Vega 20 is not for gamers/consumers, it is for enterprise market. I could be mistaken, but I believe they were asked point blank in one of their investor calls and they said Vega 20 is enterprise and that Navi is their 7nm gaming chip, so I don't see them releasing a consumer Vega 20 after telling their investors they won't be doing that.

Where are you getting that from? (That Navi isn't ready.) As far as I know, AMD hasn't said anything more than that its 7nm and is coming, but haven't given any more details than that, but they've talked about how well 7nm has been going so I don't know why they wouldn't be able to have it going quite quickly (seems like 7nm is actually doing so well that they're ahead of where they expected they'd be). I don't think they've even actually confirmed that its in the next Playstation (but they probably won't confirm that until Sony announces it so that's not surprising). I actually think though that is partly why it'll come out so early, so they can put it in dev kits to prepare for the PS5 (which I believe will launch late next year).

Its a mainstream level card. That level of performance for $250 is a significant improvement over what's been available at that price. The rumors I've seen put Navi 10 allegedly at GTX1080/Vega 64+ (I think I've seen some say 15-20% + over) level performance for the top Navi 10 card. They could easily match the SP count of Vega 64, plus it'll likely clock higher. And memory bandwidth shouldn't be too far off. GDDR6 at 14Gbps would be 80% more bandwidth than Polaris 256GB/s, putting it at around 400GB/s if it keeps the 256bit bus. If they use 16Gbps, then it'd be a full doubling and they'd be getting 512GB/s. And if they change the bus, it could be even higher. If anything, Vega 20 for gamers makes no sense as it'd be stupid expensive and very possibly not offer Navi 10 level gaming performance (other than in very bandwidth/memory limited games). AMD has said its rendering capabilities (CU counts, etc) are the same as Vega 64, as they kept that all the same as Vega 64 and that all the stuff they added was for deep learning stuff. And while it probably can clock higher, and would have much more memory bandwidth, it'll also cost a LOT more money so that's not for consumers.

Something that caught my attention was it saying it would be announced at CES but availability would be 1H. That seems to align with the rumors from a couple of months back about Navi (where the source - allegedly AIB partners - said AMD was aiming for Q1 for Navi but that 1H was more likely; which I personally believe that Navi was the product that was supposed to start production on GF's 7nm in December, and so it likely got delayed a bit moving it to TSMC).

Oh, and I don't think the 590 will matter. AMD doesn't seem to care about their old products (from the rumors about Zen 2, they're going to destroy the earlier Ryzen chips in value too).

Yeah, we'll just have to wait and see. CES should be interesting!

Probably talking about Pinacle Ridge.

I think that will be part of it, but I think its more than that.

Here's a quote from their press release about Lisa Su CES Keynote:

In 2019, AMD will catapult computing, gaming, and visualization technologies forward with the world’s first 7nm high-performance CPUs and GPUs, providing the power required to reach technology’s next horizon. During her CES keynote, Dr. Su and guests will provide a view into the diverse applications for new computing technologies ranging from solving some of the world’s toughest challenges to the future of gaming, entertainment and virtual reality with the potential to redefine modern life.

And here's a quote from the head of the org that puts on CES:

“AMD is transforming the future of computing in our ever-expanding digital world and revolutionizing the $35 billion gaming industry,” said Gary Shapiro, president and CEO, CTA. “We look forward to Dr. Su’s keynote as she paints a picture of the next-generation of computing that will help redefine the future of gaming and virtual entertainment.”

They've outright said 7nm CPU and GPU is the focus of it. They've outright said gaming is one of the key areas (Vega 20's market is not gaming, entertainment, or virtual reality) and that Navi is their 7nm gaming GPU. I'm just using deductive reasoning based on what they've saying. Its possible that gaming part is for Zen 2, and then Vega 20 is for the "solving some of the world's toughest challenges", but something about how they keep talking about gaming makes me think a gaming GPU is almost certainly part of it.

If AMD would be smart they would introduce some AVX like instruction set for their new GPU design or at least some hairworks like black box type effect packet that is locket to the new core and mandatory to new consoles. This seems to be the only way they could compete nvidia and beat them in their own game.

But they are still playing the good guys card and getting stumped on. Not a viable business practice when you are the underdog.

That might be interesting but I'm not sure its feasible or would be worth it. I think Intel already tried that too (they put AVX into one of their chips that had a bunch of Atom cores) and well they're doing their own GPUs now instead.

I really don't think doing that type of situation would work (the black box type of Gameworks stuff for AMD). For one, I don't think it'd fly with the console companies, because developers tell them they want more common stuff so that their development resources goes further. That was why they switched to x86. And AMD doesn't have the clout, even with consoles, to be able to do that. That's why they let others take what Mantle was and make it into Vulkan and DX12, so that it would be part of a standard that others could be compatible with and gain support, otherwise few were going to support it.

I disagree, I think going the open route is the best for AMD. They really can't very well lock people in if they are the underdog as that makes it that much easier to just not use them (so instead of people considering switching to AMD, they'd just go, well I'm already in with their competitor, might as well stay if I'll have to switchover everything and lockin to AMD).

 

[DrMrLordX]

@darkswordsman17 :

If Navi10 can outperform Vega64, great. I'm too jaded at this point to expect it, though. And I realize that Vega20 has features that would essentially be useless for consumers (and I guess some professionals, too) so I remain skeptical of any rumours about Vega20 showing up in a Frontier Edition-type card. Maybe if it's stripped-down Vega20 - minus the AI features - then I would believe it. But that might require an entirely new die, and I'm not sure AMD would bother with the expense just for a niche product that has no lower-end followup (imagine Vega FE with no Vega64 coming behind it in 2017). AMD has been pretty clear about Navi being what's there for the consumer market, and I'll stick to that vision of the future unless I learn otherwise from a reputable source. I am not all that hopeful about Navi arriving in any meaningful form before Q3 or Q4 2019, though. If they can get it done earlier without the hardware being a buggy mess, then great.

Also, Intel's idea to put a bunch of Atom cores together with advanced AVX instructions was the basis for Larrabee and Xeon Phi. I'll let you decide for yourself how well that worked out.

 

[jpiniero]

Vega 20 really isn't that different than Vega 10. The transistor count difference isn't that much. Now it would have to be expensive I reckon, hence the Frontier Edition.

 

[lifeblood]

While I doubt this rumor is true, let’s pretend for just a moment that it is completely true. Nobody expected RX 590 but AMD brought it out anyway for the midrange market, so this wouldn’t be totally unprecedented. With Navi looking like 2H of 2019, AMD may have decided to release a stopgap high level GPU to hold them over.

As a die shrink it would be cheaper than Vega 64, and should also be faster, they would just need to somehow disable the unused enterprise features. AMD could charge the same amount it does currently for Vega 64 (dropping the price of Vega 56/64) and people would buy it.

When AMD said Vega 20 is enterprise and Navi is their 7nm gaming chip during an investor call I’m sure it was the truth, then. Corporations are required to tell the truth on those calls, but they’re not forbidden from changing their minds afterwards. If somehow Vega 20 turned out to be a competitive gaming chip with little to no modification they may have decided to give it a try.

 

[bsp2020]

It is conceivable that AMD would release Vega II 56 & 64 using 12 nm and Vega II FE using 7nm in 1H. Since Polaris products covers up to only $300 price range, refreshing their $400, $500 and $1000 products makes sense. AMD can still claim that they have released the first 7nm gaming GPU to the market. It seems NVidia has done such a good job expanding the definition of "gaming market" that even $1000 GPU qualify as a gaming GPU these days.

That way, AMD has some new products in the market to hold them over until they release the real consumer oriented 7nm NAVI products in 2H.

 

[ub4ty]

The biggest thing that needs to occur for GPUs is a 7nm die shrink to somehow get these absurd power draws down. While any new features/performance would be welcomed, the biggest thing I need to see are significant power draw drops. It's getting to a point of absurdity now. CES is a big event for AMD. There are bold new markets that no one has even considered. Gaming is dead IMO until they tap into a new niche/indie futuristic vein that is void of sjwer/political/feminist/divisive/diversity/propaganda memes and instead focuses on something that touches the hearts of a future generation and compels them forward into the unknown. Cyber Punk 2077 seems most interesting because it transcends the B.S were currently steeped in and pushes the gamer into a futuristic landscape. There needs to be more things like this to even drive (sustain) the hardware sales IMO. Right now gaming is a drab experience centered on the rehashing and circling of old social themes.. Lipstick on a pig. It centers on profit churning the masses instead of doing anything ground breaking. Playing a dumpster fire politically charged uninspired mess isn't 'amazing' just because its in 4k... It's still garbage.

AMD is in a cushy position from this perspective... The games are garbage and nothing is really exciting or pushing demand for new hardware. So, there's Vega which pretty much delivers what a user could reasonably expect. Meanwhile, Nvidia is tripping over themselves attempting to convince consumers to buy new hardware for a gaming experience that is lackluster because the games and community therein sucks.

Historically, really great things develop in niches far away from a mainstream sphere that tends to run great things into the ground and/or adopt things once all the risk is gone and there's no soul left. This will likely never change. The big genuine risk taking for something bold and new needs to happen again in gaming. A return back to the grassroots spirit of gaming with the delivery of something that pushes the gamer forward into uncharted territory. No stupid reboots, no stupid back in time nonsense, no diversity agendas, no 'the future is female' idiocy.... Something that brings all ethnic groups/genders forward into an alien landscape of the future.

Look at a show like 'The Expanse' for what I mean. Almost everyone in the world is represented and you're all collectively catapulted into this wild new future... When that happens for gaming, everyone will be clamoring for the latest and greatest. Not for stupid FPS comparisons but because you will want to be there....

There's 3 I's of every business cycle (innovators, imitators, and idiots). We've been steeped in the idiot cycle for some time now on multiple fronts. What comes next is the end of the current absurdity and the start of something new. No one is at the forefront of this because I have yet to see it. Meanwhile, it'd be great if GPUs weren't such infernos and power hogs. As such, at a minimum, no one should be tripping over themselves until 7nm arrives on GPUs and even then .. not until there's a big time game that achieves what everyone in the gaming industry is failing at. That being said, that's why AMD is focusing on GPU compute. It's where the 'big leaps' forward are occurring and new markets being defined. Gaming is dead currently.

 

[thecoolnessrune]

I mean VEGA was announced at CES 2017 and didn't launch till August. I don't put it out of the realm of possibility with them announcing NAVI at CES and not launching until "2H 2019".

 

[DrMrLordX]

I mean VEGA was announced at CES 2017 and didn't launch till August. I don't put it out of the realm of possibility with them announcing NAVI at CES and not launching until "2H 2019".

Entirely possible. I would be pleasantly surprised if they could get Navi out by August 2019.

 

[happy medium]

I expect a 7nm prosumer card 15% faster than a Vega 64 in early spring.

 

[Guru]

How are you getting late 2019? In fact few AMD top officials have since then said that they are looking at a mid 2019 for their GPU's. In fact it seems as if Navi will already be here in Ryzen 3000 chips with integrated GPU's.

Seems to me once they have few batches of Ryzen 3000 out, they will probably start mass producing 7nm GPU's.

 

[DrMrLordX]

How are you getting late 2019? In fact few AMD top officials have since then said that they are looking at a mid 2019 for their GPU's. In fact it seems as if Navi will already be here in Ryzen 3000 chips with integrated GPU's.

There were some roadmaps showing Navi as late 2019/2020. It was awhile back.

Seems to me once they have few batches of Ryzen 3000 out, they will probably start mass producing 7nm GPU's.

They're already mass-producing 7nm GPUs (Vega20). They just aren't selling them to us.

 

[GodisanAtheist]

So for people not wanting to click on the link:

The tweet claims that Navi is not GCN eg. a new CU design which probably also gets away of the ROP limits and increases efficiency. That at least makes the leaks more believable as the power figures stated are way too good to just be another shrunk GCN part.

-AMD has gotten their money's worth and then some out of GCN and it's ever so slightly uplifting to dream that AMD has just even a solid product in the works.

Doesn't have to blow the doors off NV, just something that gets them back into the competitive landscape and as the arch will likely be unburdened by RTX hardware a possible return to efficiency.

 

[Guru]

There were some roadmaps showing Navi as late 2019/2020. It was awhile back.



They're already mass-producing 7nm GPUs (Vega20). They just aren't selling them to us.

Yeah, that's what I was alluding to. I mean mass produce 7nm desktop GPU's.

 

[ElFenix]

Speaking to AMD's CFO Devinder Kumar:

https://www.reddit.com/r/Amd/comments/9frjoe

It will be like this: Vega 20 for HPC (late 2018) > Epyc Rome (2019) > Ryzen 3000 > Navi.

So late 2019, early 2020.

this aged well.

 

[IEC]

this aged well.

Absolute certainty... while being a full year off and missing the Radeon VII.

 

[Muhammed]

this aged well.

Yeah it did, Navi is pushed to the back now, late 2019 is definitive now.

 

[jpiniero]

Absolute certainty... while being a full year off and missing the Radeon VII.

I think the RTX pricing convinced them to do it.

 

[GodisanAtheist]

I think AMD is just trying to be more aggressive.

Lisa Su isn't content to be the AMD of old and is likely putting a lot of pressure on various teams to think more like a competitor and less like an underdog. There is that old saying "A corporation is like a shark, it either moves forward or it dies". Sitting still in the GPU space isn't an option with a competitor like NV and AMD needs to make more moves like this in the future.

I believe AMD's CFO at the time of that statement was stating the honest truth as he knew it at the time. The truth just changed on him.